In order to see how well your chatbot is performing, you need to put some automated testing of your chatbot in place. This will help you get some objective measures of chatbot accuracy and performance, and will also give you some insight into areas where your chatbot could stand to improve. Keep in mind that testing cognitive applications is fundamentally different from testing more traditional applications. Traditional applications relied on path testing, and code coverage. Due to the non-deterministic nature of cognitive applications, we end up having to use statistical models for our testing.
I STRONGLY suggest using the Python notebooks listed below. These can be run on your own machine, or from within Watson Studio. They are a great starting point for your automated analysis of your chatbot. Over time you can improve these notebooks and expand on some of the testing techniques in them.
Log Analysis Workbooks — these two notebooks focus on slightly different areas, and do an analysis of your chatbot based on your logs.
My K-Fold Notebook — not as full-featured as the two notebooks above, but I like mine because it’s simple. For a new Python user, it also has a straightforward approach to k-fold testing that is easier to understand.
Some change management scenarios are enabled by using Watson Assistant’s new versioning capability. This versioning is a bit different than some versioning that you might be used to — it’s not like a typical software development check out, check in paradigm. Instead, it is more like a baseline establishment paradigm, where a version (or baseline) is created as a snapshot of the model at some point in time. The current state of the model (which you see in the UI) is always the tip, or development version.
So if you go into your Watson Assistant Plus instance, and go into your skill, you can just click on “Versions” in the left-hand nav bar, and you will see the versions screen.
In order to drop a new version (or baseline), just click on the “Save a new version” link in the upper right corner. You should then fill in something useful (think about naming conventions!) for the description, before pressing “Save”. We will talk more about naming conventions below, when we start to talk about some typical change management workflows.
Typical Change Management Workflows
A typical change management workflow will allow you to easily scope, contain, and then deploy changes to your chatbot. What I am proposing here is something that might not satisfy a “DevOps Purist”, since we have production and dev/test resources residing in the same service.
Keep in mind that this approach CAN easily be modified so that instead of moving the versions around within a Watson Assistant instance, we could export the model from one Watson Assistant environment (say the test environment), and then import it into another environment (say the production environment). These environments could be located in different resource groups, with individualized access and permissions tailored for each environment/resource group.
So without further explanation, here is a high-level view of a typical lightweight change management process for your Watson Covid-19 chatbot:
Get suggested changes for the next cycle. Save these in some agreed upon format/tool with changes for your Covid-19 bot.
Agree on changes and scope of changes for the next cycle. For those of you who are familiar with Agile development principles, think of this as sprint planning….
Apply these changes in the tip (development) version of your Covid-19 bot
Test development Covid-19 bot version — validate your changes — make final revisions
Run automated quality tests against development bot version (yes, I wasn’t kidding about automated tests earlier, do this now — and you will thank me later)
Create a new version in the Covid-19 bot — call it “<YYYY-MM-DD> Production”
Get formal sign off for promotion of new version to production.
Move the Assistant pointer for the production Covid-19 bot to the new version (“<YY-MM-DD> Production”)
Backup the Covid-19 bot, with an export of the dialog skill to JSON — store the JSON file with name “<YYYY-MM-DD>_Production.json”
Move the Assistant pointer for the development Covid-19 bot to the tip version
Go back to step 1 and do it again……
Note that I have been non-specific about tickets/issues/tools to be using. I just trust that you are using some sort of change control — no matter how simple. You want to get your stakeholders in the habit of making change requests and then APPROVING those requests — just so we avoid any “Why did you do that?” situations. You can implement this on the IBM Cloud (a free version of GitHub hosted on the IBM Cloud), where you can do real lightweight change management, or you can use your existing change management tooling and infrastructure. It’s completely up to you.
Change Management for the Advanced User
The simple change management process outlined above will work for simple deployments, and is probably robust enough for a chatbot that is not expected to be deployed on a long term basis. What about change management for more established chatbots? Or for chatbots that are part of a larger system that is currently under some more formal type of change management? What can we do in those cases?
In those cases, and for users wishing for a more robust and formal change management or DevOps infrastructure, you will want to read A comparison of DevOps choices for IBM Watson Assistant byLeigh Williamson. Leigh is a peer of mine who has spent some time thinking about this, and more importantly, he has spent some time actually implementing these types of change management systems.
So if you have been following this series of blog posts, you have deployed a longtail chatbot that answers Covid-19 questions. You have learned how to update that chatbot, and how to test and version control your changes. What’s left to do? At this point you know enough to be dangerous — it might be time to fill in some of those learning gaps. My blog posts tell you just what you need to know, but there are so many other Watson techniques and capabilities that you can use to make your end-user experience even better. Maybe you should check out some of the Watson Covid-19 specific learning resources that are out there.
In my earlier blog post, IBM Watson Fueled Chatbots — For Our Health, I talked about the costs and benefits of a chatbot, and showed you how a chatbot is within the reach of most organizations. I also walked you through a step-by-step guide of creating that initial chatbot. It’s not that hard to do, and the costs are minimal (or even zero if you take advantage of the IBM “Citizen Chatbot” campaign). Once you get that initial chatbot deployed, you have to update it. Given how fluid the current situation is in many locations, in terms of the guidance and response to the Covid-19 pandemic, you will be changing things OFTEN. So how do we do that? How do we maintain our chatbot?
In the following sections, I will briefly touch on how to do some basic things in Watson Assistant. You will want to find more complete guidance on all sorts of topics on our Watson Assistant landing pages, or even in the product documentation. I’ll try to make this as generic as possible since everyone will have slightly different implementations of things.
Getting Comfortable with Your Watson Chatbot
A good way to become familiar with your Watson Assistant implementation is to run some k-fold testing against it, and to look at and understand the results. The Python notebooks used for these can help you begin to understand what your chatbot is doing, how it is doing it, and gives you a foundation of test results to build on. I’ll cover this entire topic in a different post.
If you’re uncomfortable with this, you can just open up your Watson Assistant instance, and navigate into your Dialog skill. Once in there, go and look at your intents. Get familiar with the intents that you are looking for. You can export the training data for your intents. Just click on the “Export Intents” icon, and specify a CSV file to export to.
Once that file has been exported, you can open the CSV file and see all of the training data and intents. Use this same approach to export your entities and their training data, so you can familiarize yourself with those as well.
I will often keep a list of intents and their general topics on a single sheet of paper, along with entities and their focus, on my desk. It will help me keep things straight when I go in and begin to make changes to the existing dialog nodes. The typical list might look something like this:
#AnimalContact — Human and animal interaction and guidelines #AttendingFuneral — info on funerals and attendance … @phone — US phone number pattern xxx-xxx-xxxx @school_type — high school, college, etc
Simple Changes to Specific Responses
Making simple changes to the “canned” responses that our chatbot has is pretty easy. It’s also one of the best ways to begin learning and getting comfortable with the Watson Assistant.
So what can we do? Let’s start at the beginning (always a great place to start!), and jump into our Watson Assistant dialog skill. Once in, click on the “Dialog” menu option, to display the dialog tree. Remember that the dialog tree nodes are evaluated for each user interaction, and they get evaluated starting from the top, and continuing to the bottom of the tree.
Take a look at the top node in your dialog tree. You can see that it has a name, followed by a section asking if the assistant recognizes some event. This is evaluating the user’s input, and looking at the intent and possibly the entities. The next section describes what happens if that particular condition is met. This section is the first place to focus — since most new chatbots are built with some “pre-built” intents, entities and dialog nodes. They also have text that may not match what you want to say, or how you want to say it. Take this example from my earlier IBM Watson Fueled Chatbots post. Look at the dialog node titled “Naming of virus”. There is a long description of the virus name and origination — you might want your response to be a bit more brief. Go ahead and change it. Then use the “Try it” button to open a dialog and test out your change. Type in “why is it named covid-19”, and you should be able to see what your modified response looks like.
When dealing with chatbots where you may have reused the data and settings from an earlier chatbot, it is often best to step through each node in a dialog tree, and assess the answer being given back to the end-user. Is this answer accurate for my end users? Does it fit in with the overall approach of my chatbot? Is the personality of my chatbot maintained? For some tips on effective chatbots (and the importance of chatbot personality), check out some of the articles on chatbot success factors and best practices, like Conversational Assistants and Quality with Watson Assistant — Revisited and Best Practices for Building and Maintaining a Chatbot.
Pro Tip: When writing out the text in a Dialog answer, use the markup language instead of HTML tags. Markup language will get processed properly by speech engines and integrations, while HTML tags will not always be properly processed. For example, use “you should **ALWAYS** wash your hands”, instead of, “you should <B>ALWAYS<\B> wash your hands”.
What Do Our Users Want?
The next steps to take in doing some ongoing improvements and changes to your chatbot involve making some changes to what you are answering. What types of questions are your end-users asking? What new conditions have come up, conditions that need to be addressed? Do we have changes to existing guidance offered by our chatbot?
In order to find the answers to these types of questions, we’ll take advantage of something that is unique to Watson Assistant Plus editions, and not available in the traditional Watson Assistant instances. We’ll utilize the analytics associated with the operation of our chatbot.
Go into your Watson Assistant instance, and then drop into the dialog skill that you are interested in. Then click on the “Analytics” option in the left-hand nav bar. You will see a screen that looks similar to the one shown below. It has a few graphs and some interesting looking data. What should we focus on?
I tend to look in certain areas when I initially begin to look at chatbot performance.
Weak understanding — this number represents the user statements that our chatbot had a tough time figuring out. This is an obvious area to address.
Active Users — this number shows how well your solution is being adopted. It can also help you anticipate any increase in costs.
Top Intents and Top Entities — these two lists give you an idea of the areas that are being hardest hit with questions. You can see what people are asking about most often, and track how this might change over time.
When looking at the weak understanding box, click on the “View Logs” link. This will now show you logs of all of the conversations that your chatbot has had over some time period. The first view you see here is filtered — it is showing conversations where no entities were detected, and where no intent was detected. These are the questions that you had no answer for. When looking to improve your chatbot, this is typically a great place to start. See what kinds of questions are NOT being answered.
Gather a list of the legitimate questions that you want to address. Begin to organize these into rough groups. See if there are patterns. Some things that you should be looking for:
Do you see 5 to 10 different user inquiries that are asking the same basic question? That is a candidate for a new intent.
Do you see some new terms showing up a lot? Are particular slang terms beginning to emerge? These can be added to existing entities, or could be new entities that you want to identify.
Do you have any user questions that should have gone to a particular intent, but didn’t? Those user statements (commonly referred to as utterances) should be added to the training data.
What Types of Things Do We Deal With?
OK — so you have determined some areas and types of questions that you are not answering. How do we want to address these things?
The first thing to do is to improve your existing intents. You should add user utterances that were incorrectly classified, or went unclassified, to the training data for the intent that they should map to. The only exception to this is if you have more than 40 examples already in place for a particular intent. In that case, you should begin to remove similar utterances from the training data for that intent — for each one that you add, you should remove one. Otherwise, you may run into the problem of overfitting.
The next thing to do is to identify any new entities that you have. I suggest that you recognize and get these entities defined in your Watson Assistant dialog skill. After you have added these entities to your dialog skill, you should then go and begin to look at the individual dialog nodes that you already have, and determine if any of your responses can be modified to provide more targeted information based on these new entities.
Then it’s time to look at your potential new intents. Keep in mind that one or two utterances about something odd is NOT an intent. For any quality intent that you want to create, you should have at least 10 utterances. You should also check those utterances and make sure that they do not “collide” with, or replicate, utterances associated with other already existing intents.
The most simple scenario to handle with a dialog node is just directly answering a question linked to some intent. Here is an example:
At the top of the node you can see it’s name. The name is only important as a way for you to identify a node — what it does, what it answers. Make it brief, but descriptive. Next to the name is the “Customize” button, we won’t be using that for our simple dialog node example. Next is the trigger condition — in this example, we are triggering on “#Whythename”. The hashtag in front indicates that this is an intent (#intent), if there was an ampersand in front of the name then we would be referencing an entity (@entity). You can link intents and entities, as well as context variables, to further refine your dialog node actions. Since this is a simple dialog node, we’ll just trigger on a single intent. Finally, there is the response section. This response is just some text, but responses can take many forms. Try to keep your text responses simple and brief, you want to help your users, not bury them with details.
Now We Know Everything, Right?
Not really. You know how to do some of the easier things to improve your chatbot. At this point, you need to begin implementing some automated testing of your chatbot. This will help you get some objective measures of chatbot accuracy and performance, and will also give you some insight into areas where your chatbot could stand to improve.
So the next thing to do is to implement some automated testing and the application of some DevOps principles. We’ll save that for our next blog post.
With Covid-19 and all of the changes in our world, it’s interesting to think about how this impacts businesses all over the world. What can a chatbot do that is relevant in our new situation — with stay-at-home orders, social distancing, and the closing of non-essential businesses? Chatbots can do something that most local organizations need — communicate to your customers and users with a limited amount of staff, on a 24×7 basis. How many businesses are unable to staff call centers — wait times on many “help lines” can be up to two hours long. Chatbots enable you to “talk” with your customers and give them the information that they want, when they want it. Coupling your chatbot with some speech services can give you a solution that will work for your call center. A recent study by Forrester Consulting indicated that companies saw a benefit of over $5 per contained conversation handled by Watson Assistant. So what are you waiting for?
A lot of companies and organizations would consider chatbots, and other AI technologies, but they feel like those technologies are too expensive. They might feel like they don’t have the expertise to build a chatbot on their own. They might have been right — in the past. This week, IBM announced it’s “Citizen Chatbot” campaign. This is targeted at government, healthcare, and other public agencies, and it provides for 90 days of FREE access to a fully functional chatbot with Covid19 information and guidance that is kept up-to-date with content from the CDC. It allows you to quickly and cheaply try chatbot technology to see how it can work for you.
That sounds wonderful, but a lot of my readers are not in those industries. Even though they are not in those industries, they still have a need to share Covid-19 information with their employees, business partners, and customers. Some of this information might be specific to the products and services that they provide. So if I can’t qualify for the “Citizen Chatbot” campaign, is there a way that I can do this for myself? How much effort would it be? How much would it cost?
The Simple Chatbot Solution
The most simple chatbot has three components: Watson Assistant, Watson Discovery, and your web page or mobile application.
Your Web page or mobile application will be the user interface for your chatbot. It is what the user interacts with. It handles login (if needed), sets the colors and presentation of your application, what the forms look like, and all of the things that your end-user sees.
The Watson Assistant piece will drive the “conversation” with the user. It determines what the user wants, and then provides the needed information or guidance. It handles the more simple questions, the frequently asked questions (FAQ’s) and the general questions that your end users may have.
When questions get too specific for the Watson Assistant to handle, they are passed off to the Watson Discovery service. The Discovery service will read (or ingest) a series of documents or web pages, and will then search through this knowledge base (commonly called a corpus) for answers to those more complex or complicated questions. These are often referred to as “long-tail” questions.
Mobile applications may use the Watson Speech services to provide voice inputs and outputs to your Watson Assistant service. These additional services would incur an additional cost when deployed, but would allow your chatbot to relieve some call center volume (which could be a huge cost savings for you). These are not necessary for a chatbot deployment, but represent the ability to add some capability to your chatbot.
The Simple Chatbot Costs
The costs for any solution can be difficult to fully quantify. What will the software development costs be? How much maintenance will need to be done? How many people have to be on call to support the solution? I don’t know your labor and development costs, so I leave the calculation of those costs to you. What I DO know, is the cost of the Watson services that a chatbot would use. So let’s look at our simple chatbot.
We use a single Watson PLUS instance. This single instance is going to cost $120 for each 1000 unique users every month. So if our bot is servicing 10,000 users in it’s first month, with some of them calling in multiple times, my monthly cost for the Watson Assistant service is going to be $1200.
We use a single Watson Discovery instance. We were careful when creating our Discovery instance, and created an x-small size instance, since our knowledge base is less than 50k documents. This size Discovery service will cost us $500/month.
We will use a single Speech-to-Text instance. These instances get billed based on usage — at a rate of $0.02 per minute. If your usage gets really high, this price per minute will drop (see the Speech-to-Text catalog page for service pricing). If we assume 5 minutes of speech per call (which is a loooong call…), we would have a bill of about $1000 for the month.
We will use a single Text-to-Speech instance. These instances get billed based on usage — at a rate of $0.02 per thousand characters. If your usage gets really high, this price per thousand characters will drop(see the Text-to-Speech catalog page for pricing info). If we assume 4000 characters of speech per call (which is a lot of talking in each call…), we would have a bill of about $800 for the month.
Note: IBM currently provides a Watson Assistant for Voice Interaction (WAVI) bundle of Watson Assistant (WA), Watson Discovery (WDS), Speech-to-Text (STT), Text-to-Speech (TTS), and Voice Gateway (VGW) for only $650 per month!! It provides 10K minutes of call time and 10K unique monthly web users.
With all of these costs added together, we see a total monthly cost of $3500 per month (Assistant’s $1200 added to $500 for Discovery, and $1000 and $800 for the Speech services) for handling these 10000 unique customers. That sounds like a lot of money, and it is. But now let’s look at the benefits that we will see.
I am going to make some assumptions, and use some “typical” numbers. I encourage you to do the math shown here, using your own numbers and costs, and see what your particular situation is. For this example, I will assume that we have a local call center that has been struggling to handle those 10,000 calls a month. We’ll assume that our chatbot can deflect only half of those calls (most customers report far more significant call deflection rates). So we figure that 5,000 calls a month are deflected. Let’s take that $5 per deflected conversation benefit from earlier, and we can cut it in half — to $2.50 — let’s be conservative in our benefit estimates. Multiplying those, we end up with a benefit of $12,500. Keep in mind that we are trying to be conservative with our benefits calculations. Now all of sudden our $3500 a month in costs doesn’t look so bad — in fact we have a net benefit of $9000. If we ramp up with more users per month, the costs, and benefits, increase on a fairly linear scale.
Note: Simple math with the WAVI bundle mentioned above also results in a fairly linear cost/benefits curve, with only $650 in costs against a benefit of $12,500 for those first 10000 users.
What Happens Now?
Many of you will be wanting to start building out your chatbot, but your organization will want to manage the effort and will want some sort of project plan. So let’s give them one — and begin to set expectations.
Looking at this simple chatbot project plan, you can see the high-level tasks involved in rolling out a chatbot. Provisioning resources is simple, and you can do all of your provisioning in an hour or less. Keep in mind that you should do this within a functional IBM Cloud account (see my post on Getting Started Right With the IBM Cloud).
You then will need to do some work on your dialog skill, first importing some pre-built intents, training data, and dialog flows. This is simple and should take less than an hour. Updating these to fit your particular situation will take a couple of days.
In parallel to your dialog work, you can create a search skill, and implement and execute the web crawling needed to populate your knowledge base for handling those “long tail” questions. All of these tasks can be done during the first few days of the project.
At this point you will need your existing development and operations teams to work together in getting the code to deploy your chatbot widget added to your homepage (or wherever you decide to host your chatbot). I strongly suggest doing this in some kind of staging or test environment. You can then test in this environment, and eventually deploy these changes to your production environments. All in less than a week.
After this initial deployment, you will have some longer workstreams to execute on, as well as some recurring work. For the recurring work, you’ll need to set up some iterative process for reviewing your chatbot results, and identifying improvements that need to be made. You will need to identify areas where things are changing, and where your chatbot needs to change.
The two longer workstreams that are needed to be done involve the automation of the testing process and the automation of the promotion/deployment processes. These are necessary if you want to make your chatbot a smoothly functioning and updating application. The processes and tools that you use here should be able to be used for ANY chatbot that your organization creates.
Let’s Get Ready for Building!
Before I launch into an overview of the steps that you need to take to build your very own Covid-19 Chatbot, you should be aware that there are now hundreds of “How to build your own Covid-19 chatbot” articles out there. Many of them are really good, and you might prefer their approach to what I have here. So I’ll list a few of them here — take a look at them and use the approach and the instructions that seem most straightforward to you.
Simple Covid-19 Chatbot — a great video that shows you how to do this in 4 minutes. He moves fast, so you’ll probably have to start/stop the video quite a bit, but it is simple and straightforward.
The first thing to do is to get a “functional account” in your domain. We touched on this earlier (my post on Getting Started Right With the IBM Cloud), but it is important. Make sure that your “owning” account is not tied to a single person — but is more of a service account that multiple people can administer.
Using that “functional account”, get yourself an IBM Cloud account. Some organizations may already have an IBM Cloud account, in which case, you’ll just need to make proper arrangements with the account owner.
We need to do some basic housekeeping and setup of our Cloud account, so we can more easily manage access to resources. We’re going to want to create a new Resource Group and some Access Groups for our chatbot project. So first go to “Manage -> Account”, and then click on “Resource Groups”. You’ll see a list of resource groups (maybe just a default resource group — it depends on if this is a new or existing account). On this screen, we’re going to click the blue “Create” button to create a new Resource Group.
Call your new Resource Group something informative, like “Covid_Chatbot_RG”. Make sure to add the “_RG” at the end of your Resource Group name — it’s never to early to start using naming conventions.
Now we want to create some Access Groups — so we can easily add users to our project based on the role that they will play. So select “Manage -> Access (IAM)”, and then click on the “Access Groups” option. Now we’ll create three access groups, and name them with our resource group name, followed by the role, and then an “_AG” (which shows they are an access group). So for my three access groups, I created the following: “Covid_Chatbot_Admins_AG”, “Covid_Chatbot_Developers_AG”, and “Covid_Chatbot_Users_AG”. After creating these access groups, you should now see them on the “Access Groups” page.
The last thing to do with access groups is to set permissions for the access group. From the “Access Groups” page, click on the “Covid_Chatbot_Admins_AG” group (or your equivalent). Once the access group comes up, click on the “Access Policies” tab, and click the blue “Assign Access” button. This screen is where we will assign access for each of the Access Groups. Here is what you need to set for each group:
For the “Covid_Chatbot_Admins_AG” group:
Select “IAM Services”, “All Identity and Access enabled services”, “Resource group: Covid_Chatbot_RG”, and “All regions”. Check every box for platform access (Viewer, Operator, Editor, Administrator), service access (Reader, Writer, Manager), and resource group access (Viewer, Operator, Editor, Administrator). Note that you may need to click on the “View All” link to see all of your options for resource group access. Press the “Add” button at the bottom of the window.
Select “Account Management”, and “All Account Management services”. Check only two boxes for platform access (Viewer, Operator), and press the “Add” button at the bottom of the window.
Finally, press the blue “Assign” button in the right-hand pane, to assign these permissions to this access group.
For the “Covid_Chatbot_Developers_AG” group:
Select “IAM Services”, “All Identity and Access enabled services”, “Resource group: Covid_Chatbot_RG”, and “All regions”. Check two boxes for platform access (Viewer, Operator), and three boxes for service access (Reader, Writer, Manager). Press the “Add” button at the bottom of the window.
Select “Account Management”, and “Support Center”. Check three boxes for platform access (Viewer, Operator, Editor), and press the “Add” button at the bottom of the window.
Finally, press the blue “Assign” button in the right-hand pane, to assign these permissions to this access group.
For the “Covid_Chatbot_Users_AG” group:
Select “IAM Services”, “All Identity and Access enabled services”, “Resource group: Covid_Chatbot_RG”, and “All regions”. Check one box for platform access (Viewer), and one box for service access (Reader). Press the “Add” button at the bottom of the window.
Select “Account Management”, and “Support Center”. Check three boxes for platform access (Viewer, Operator, Editor), and press the “Add” button at the bottom of the window.
Finally, press the blue “Assign” button in the right-hand pane, to assign these permissions to this access group.
The last thing to do with access groups is to set your account into the admin access group. From the “Access Groups” page, click on the “Covid_Chatbot_Admins_AG” group (or your equivalent). Once the access group comes up, click on the “Users” tab, and click the blue “Add users” button. Now select your account, and then click the blue “Add to group” button.
Now we want to people invited to our project area. Select “Manage -> Access (IAM)”, and then click on the “Users” option. Now you can add the email addresses (IBM Cloud accounts) of the people you want to invite to your effort by clicking on the blue “Invite users” button. Add your users in groups, separated by commas, and assign each group of email addresses to a particular Access Group by clicking the blue “Add” link for the access group that you want assigned. This will give our new users access to the areas of the account that they need access to.
Now we’ll log in using your own account (in my case that might be firstname.lastname@example.org), so we can check out these changes and get started with our chatbot.
Getting Services Created
The first steps in this project involve getting your Watson services deployed. We’re going to try to keep this as simple as possible — if you are looking for more complex deployment examples, see some of the links in the previous section, or reach out to me and I can probably point you in productive directions.
Now we want to make sure that we create these in the right place, so log into the IBM Cloud in your own account, and switch your context to be operating in the newly created account (where the chatbot is going to “live”). To make this context switch, go to your top bar, and look for your current context. Click on it and then choose the more general account (and not your personal account).
Now let’s start creating some services. Click on the “Catalog” menu option, and then look at the AI offerings. You will want to choose the Watson Assistant. When you click on it, it will give you some options for creating your service. You will want to choose a geographic region (or data center) where you want to create your instance. Choose one that is relatively close to your physical location. Then select a plan (choose the Plus plan). Now you will need to select a service name (I have a naming convention, so mine is called “WA_Covid_Chatbot_Tox”) and a resource group (we’ll use the one we created, “Covid_Chatbot_RG”). Once this is all set, click the blue “Create” button in the right-hand pane to create your Watson Assistant instance.
While this instance provisions, let’s go and create the service that will handle our long-tail questions. Click on the “Catalog” menu option, and then look at the AI offerings. You will want to choose the Discovery service. When you click on it, it will give you some options for creating your service. You will want to choose a geographic region (or data center) where you want to create your instance. Choose the same one as before. Then select a plan (choose the Advanced plan). Now you will need to select a service name (I have a naming convention, so mine is called “WD_Covid_Chatbot_Tox”) and a resource group (we’ll use the one we created, “Covid_Chatbot_RG”). Once this is all set, click the blue “Create” button in the right-hand pane to create your Watson Discovery instance.
Begin Doing Some Assistant Work
So now we are ready to do some damage. Let’s go to our IBM Cloud dashboard, and take a look at our services/resources. Under the services section, you should see the two services that you created. Because you were smart and you used naming conventions, you know by the names what services these are. So you click on the name beginning with “WA” to open your Watson Assistant instance. Once it opens, click on the blue “Launch Watson Assistant” button. You’re now into the Watson Assistant tooling.
You now see a page with your Assistant displayed, it’s called “My first assistant”. You can see that it has one skill, under Dialog, called “My first skill”. Click on the second icon (called “Skills”) in the left-hand nav bar to pull up your list of skils. You should now see a list of all of your skills (you only have one right now). Click on the three dots in the upper right-hand corner of the skill, we need to rename this thing. Select “Rename”, and then rename your skill to “My Covid Skill” or something similar. Now just click on the main area of the box — and you are now in the skill tooling.
We’re starting from scratch, but we want to get this deployed quickly. So lets grab some work and data that has already be done for us. Click on the three dots in the upper right of the display, and select the “Import” option. We are going to import some JSON that represents a pre-built generic Covid-19 chatbot. You will want to use the fileCovid Json as the import file.
Note: You need to download this file to your local machine to make this work. If you’re not a GitHub person, then just click on the “Raw” button, and a new page will be displayed in your browser window. Once this is displayed, in your browser, just save the page by clicking “File -> Save Page as…”, and save it in JSON format. This content is from the “Call for Code” competition.
You can now poke around and take a look at the Intents that were added to your skill. Also take a look at the Entities, and the Dialog nodes. Don’t change anything — but look around at things. Take some time, after the import of new data your Watson Assistant needs to take some time to learn (called training) all of the new things that you just threw at it. After a couple of minutes, you can click the “Try it” button, and you should see a pane open up on the right-hand side of your browser. This is where you can test and check out your chatbot. Make sure that there is no flag that says that Watson is still training — and then try typing in some questions for your chatbot.
After that, you can now take a look at some of those dialog nodes. Click on one of them — and see how the responses are returned to the end user. Those conditions, the ones with the hashtag in front, are based on your intents. Those conditions with the @ in front of them are based on your entities. Using the detected intents and entities, we determine what the end user is asking about, and we supply an appropriate answer to them.
Let’s Look at the Long Tail
Now we need to configure our other service, the Watson Discovery service. Let’s go to our IBM Cloud dashboard, and take a look at our services/resources. Once again you smile a little, you were smart and you used naming conventions, so you click on the name beginning with “WD” to open your Watson Discovery instance. Once it opens, click on the blue “Launch Watson Discovery” button. You’re now going to see the Watson Discovery tooling.
The first thing that we need to do is to create our Discovery environment and choose a size for our Discovery instance (for more details on this, check out my blog post on Watson Discovery at the Size You Want). So in your Discovery window, find the environment icon (in the upper right corner), click on it, and then click on the blue “Create environment” button. The size of your environment is important — it determines the cost of your Discovery service. We wanted a quick chatbot, and we don’t anticipate a huge amount of documents needing to be stored, so we choose the “X-small” option, under the development tab. Once you find this option, click on the blue “Set up” button to begin setting up your environment, which will take a few minutes.
Once your environment is all set up, we’ll set up a simple collection and a simple web crawl to populate that document collection. You’ll notice that there is a button to “Connect a data source”. Press it. Next we’ll select “Web Crawl” as our data source. Now we get to add URL’s of the sites that we want to crawl for information. Let’s begin with these sites (you may want to add your own information sources):
As you add these URL’s, click the small gear icon that comes up with them as they are added. Then configure the number of hops for each data source. The web crawl will follow links, and links from those pages, and this is the depth of that link following that will be used. Once you have added your sources, click on the blue “Save & Synch” button. This will start your initial web crawl of data.
Your web crawl has now begun. On this screen, you can see the name of your new collection, which is just some random collection of characters. We want to change that. We’re going to call our new collection “CDC_Covid_info”, so click the small pencil icon next to that random name, and change it. So now we’ll just wait for those sites to get crawled and processed. This takes a little while, so we’ll just continue on and let it continue in parallel.
Note: Don’t panic if/when your Discovery web crawl begins to display numbers of documents that have failed. Documents fail to get ingested for a number of different reasons. Usually we just cast a wide net, and if some pages don’t get properly picked up, then we’ll find that information somewhere else.
Bringing it Together
Check your watch — how long has it been since we started? All we need to do now is connect these services, and then deploy our chatbot. Let’s start by connecting the services.
Go back into your Watson Assistant instance, and let’s look at the Assistants, which you can see by clicking on the top icon in the left-hand navigation pane. You can now see your only assistant — click on the rectangle to go into the assistant. Once inside, you can see your recently imported dialog skill. If you click on that box, you go to that familiar tool where you can modify your intents, entities, and dialog nodes. Let’s not do that right now, instead let’s look BELOW our dialog skill. There is a space there for a Search Skill. Let’s go and add one of those!
So press the blue “Add search skill” button, and you will be asked to name your search skill. Let’s name it the same thing as our document collection in Watson Discovery (“CDC_Covid_info”)— that way we won’t be confused or misled later — and hit the blue “Continue” button. On the next screen we need to choose our Watson Discovery instance (“WD_Covid_Chatbot_Tox”), and our web crawler collection of documents (“CDC_Covid_info”). Then press the blue “Configure” button, which will bring you to the last configuration screen. This screen allows you to tailor the information returned to your user. For now, we’ll just stick to the default values. So the last thing to do is to press the blue “Create” button to establish the collection as your source for answers to any “long tail” questions that your chatbot gets.
Now we need to make sure that we can get there. So let’s go into the Dialog skill for our chatbot. Scroll all of the way to the bottom, and find the dialog node called “Anything else”. Click on this node and open it up. If you look at it, it basically states that if it sees anything not covered in your intents, it will fire off a webhook. We don’t have time to do webhooks today (see some of the great materials on webhooks by Mitch Mason, Morgan Langlais, or this simple webhooks demo), so we’ll get rid of this. Click on the minus sign next to all but one of your options under the “Assistant Responds” section. For the last remaining option, click on the small gear, and configure the response. Under the “If assistant recognizes”, the value should be “anything_else”. Under the “Assistant Responds” area, select “Search skill”. Once you’ve done this (which tells the chatbot if it cannot detect an intent, to send the question to the search skill — or Discovery collection), press the blue “Save” button.
Testing and Releasing
Your chatbot is done — it might not feel that way, but it is. So let’s check it out. First we’ll want to do a basic check out of the answers to common user questions. So go back into your Watson Assistant instance, and let’s look at the Assistants. You know how to get there by now — if you don’t, then scroll back up and figure it out again.
Click on your assistant, and in the upper right of the window, you will see a section for integrations. In that section is a link that says “Add integration +”. Click on that link.
You have a few options here, but right now we’re just going to use the “Preview link” option. This is there so you can test out your chatbot with a private link, before you have to expose it to the world. So click that option, and just leave the default name for your preview link, and press the blue “Create” button. Once you do — a link will be generated that will allow you to see your chatbot in a web browser. Click on that link and ask some questions, check out how your chatbot is working.
When you are ready — and you should be ready pretty quickly — it’s time to deploy this chatbot. Let’s go back to the assistant page, and click on that link that says “Add integration +” again. This time we’ll add a “Web chat” integration. So click that option, and again we will just leave the default name for your name, and press the blue “Create” button. Now we see some options to customize your UI, and as you change these options you can see a sample window to the right of what your chatbot will look like. Once you are done playing with the title bar name and the colors that you want, you can just copy the code in the following section, and have it added to the web page where you want your chatbot. Simple as that.
Now you need to update this chatbot. You should learn more about how Intents work, how Entities can be used to help make your chatbot responses more targeted, and how Dialog nodes work. I like the Watson Assistant landing page (full disclosure — I help curate this page) for links to more learning resources. You can also take some of the online courses offered if you prefer to learn that way. The BEST way to learn is by doing. Make changes to your chatbot and see how it reacts. Don’t be scared — if you ever want to try something out, just duplicate your skill first, and then work on the copy. Then you always have the original to go back to if you really screw something up.
You will want to make sure that your responses haven’t needed to change due to changing conditions, and you may want to add new intents for new situations that your users might ask about.
If you’re really stuck, or just overwhelmed with the possibilities, you can always reach out to me. I can help you talk through your options, and I can give you some good advice to move forward with. Just reach out to me at email@example.com.
Note: I updated the original Conversational Assistants and Quality blog post in February 2020 to add a link to a much better testing notebook that I discovered, and to do a slight rewrite of that section. This blog post is a complete update to that original post – and it restates a lot of what I highlighted in the original post. The BIG difference is the new Python testing notebook – which is located out on GitHub, as CSM-Bot-Kfold-Test.
In early February of 2020 I was informed of this great blog post and Python notebook, on How to Design the Training Data for an AI Assistant. I REALLY liked this Python notebook MUCH better than my original k-fold notebook (from August of 2019). The other nice thing is that you can discover this Python notebook in the catalog in Watson Studio, and just apply it and have it added to your Watson Studio project. The only big difference with this notebook is that you need to have your testing data in a separate CSV file – it doesn’t break up “folds” based on your training data. It didn’t even do folds – just straight training and testing data.
I wasn’t a big fan of that approach, I liked my basic approach of pointing at only a Watson Assistant instance, and using all of the training data in a series of k-fold tests. Nobody wants to manage this data, that data, this file, that file….. it’s an opportunity to screw things up. Most of my customers are NOT AI experts, they just want a suite of tools that they can point at their chatbot engine that will allow them to do some automated testing of their chatbot. I have also noticed that many will use ALL of their training data, and not hold back some as test data. Doing k-fold testing using all of the training data in an existing Watson Assistant instance addresses this.
However, I really liked some of the analysis that they had done of the training data, and some of the other insights that they provided. So I decided to dive in and spend a little time merging the best of both of these approaches together. First, let’s start with some basic “rules” that you should be following if you are developing a chatbot.
Getting Started with Your Conversational Assistant
Back in July of 2019, I was working with a group of like-minded people inside of IBM, and we decided to create an IBM internal chatbot that would capture a lot of the “institutional knowledge” that some of our more experienced members knew, but that didn’t seem to be captured anywhere. We wanted our newer team members to be as effective as our more seasoned members.
We spent a week or two coming to a common vision for our chatbot. We also mapped out a “growth path” for our chatbot, and we agreed on our roles. I cannot begin to stress how important this is – Best Practice #1 – Know the scope and growth path for your chatbot. We had a good roadmap for the growth of our chatbot. We mapped out the scope for a pilot, where we wanted to be to release it to our end users, and a couple of additional capabilities that we wanted to add on once we got it deployed.
My boss graciously agreed to be our business sponsor – his
role is to constantly question our work and our approach. “Is this the most cost-effective way to do
this?”, and, “Does that add any value to your chatbot?”, are a couple of the
questions he constantly challenges us with.
As a technical guy, it’s important to have someone dragging us back to
reality – it’s easy to get focused on the technology and lose sight of the end
Our team of “developers” also got a feel for the roles we would play. I focused on the overall view and dove deeper on technical issues, some of my co-workers served primarily as testers, some as knowledge experts (SME’s), and others as served as UI specialists, focusing on the flow of conversation. This helped us coordinate our work, and it turned out to be quite important – Best Practice #2 – Know your roles – have technical people, developers, SME’s, architects, and end users represented. If you don’t have people in these roles, get them.
Starting Out – Building A Work Pipeline
As we started, we came together and worked in a spreadsheet
(!?!), gathering the basic questions that we anticipated our chatbot being able
to answer. We cast a pretty wide net
looking for “sample” questions to get us kickstarted. If you are doing something “new”, you’ll have
to come up with these utterances yourself.
If you’re covering something that already exists, there should be logs
of end user questions that you can use to jumpstart this phase of your project.
Next, we wanted to make sure that we had an orderly development environment. Since our chatbot was strictly for internal deployment, we didn’t have to worry too much about the separation of environments, so we could use the versioning capabilities of Watson Assistant. Since our chatbot was going to be deployed on Slack, we were able to deploy our “development” version on Slack, and also deploy our “test” and “production” versions on Slack as well. These are all tracked on the Versions tab of the Watson Assistant Skill UI. This gives us the ability to “promote” tested versions of our skill to different environments. All of this allowed us to have a stable environment that we could work and test in – which leads us to Best Practice #3 – Have a solid dev/test/prod environment set up for your Conversational assistant or chatbot.
How Are We Doing? – K- Fold Testing
As we started out, we began by pulling things together and seeing how our conversational assistant was doing in real-time, using the “Try It” button in the upper right-hand corner of the Watson Assistant skills screen. Our results were hit and miss at first, so we knew that we needed a good way to test out our assistant.
In February of 2020 I was informed of this great blog post and Python notebook, on How to Design the Training Data for an AI Assistant. I liked this Python notebook MUCH better than my old K-fold notebook, but I liked my approach better. So I went to work combining the best of both worlds into a new Python notebook. My new Python notebook does this – and provides some great insight into your chatbot. Go and find it on GitHub, where it is stored as CSM-Bot-Kfold-Test.
This highlights our next best practice – Best Practice #4 – Automate Your AI Testing Strategy.
As we let our automated training process take hold, we noted that our results were not what we had hoped, and that updating things was difficult. We also learned that taking time each week to review our Watson Assistant logs was time well spent.
It was quite difficult to add new scope to our conversation agent, so we looked at our intents and entities again. After some in-depth discussions, we decided to try a slightly different focus on what we considered intents. It allowed us to make better use of the entities that we detected, and it gave us the ability to construct a more easily maintained dialog tree. We needed to change the way that we were thinking about intents and entities.
All of this brings us to our next piece of wisdom – Best Practice #5 – Be Open-Minded About Your Intents and Entities. All too often I see teams fall into one of either two traps.
Trap 1 – they try to tailor their intents to the answers that they want to give. If you find yourself with intents like, “how_to_change_password” and “how_to_change_username”, then you might be describing answers, and not necessarily describing intents.
Trap 2 – teams try to have very focused intents. This leads in an explosion of intents, and a subsequent explosion of dialog nodes. If you find yourself with intents like, “change_password_mobile”, “change_password_web”, “change_password_voice”, then you have probably fallen into this trap.
We found that by having more general intents, and then using context variables and entities to specify things with more detail, that we have been able to keep our intents relatively well managed, our dialog trees smaller and better organized, and our entire project is much easier to maintain. So, if our intent was “find_person”, then we will use context variables and entities to determine what products and roles the person should have. Someone asking, “How do I find the program manager for Watson Assistant?”, would return an intent of “find_person”, with entities detected for “program manager” and “Watson Assistant”. In this way, we can add additional scope without adding intents, but only by adding some entities and one dialog node.
Why K-Fold Isn’t Enough
One thing that we realized early on was that our k-fold
results were just one aspect of the “quality” of our conversational
assistant. They helped quantify how well
we were able to identify user intents, but they didn’t do a lot for our
detection of entities or the overall quality of our assistant. We found that our k-fold testing told us when
we needed to provide additional training examples for our classifier, and this
feedback worked well.
We also found that the “quality” of our assistant improved
when we gave it some personality. We
provided some random humorous responses to intents around the origin of the
assistant, or more general questions like, “How are you doing today?”. The more of a personality that we injected
into our assistant, the more authentic and “smooth” our interactions with it
began to feel. This leads us to Best
Practice #6 – Inject Some Personality Into Your Assistant.
Some materials from IBM will break this down into greater
detail, insisting that you pay attention to tone, personality, chit-chat and
proactivity. I like to keep it simple –
it’s all part of the personality that your solution has. I usually think of a “person” that my
solution is – say a 32-year old male from Detroit, who went to college at
Michigan, who loves sports and muscle cars, named Bob. Or maybe a 24-year-old recent college
graduate named Cindy who grew up in a small town in Ohio, who has dreams of
becoming an entrepreneur in the health care industry someday. This helps me be consistent with the
personality of my solution.
We also noticed that we often needed to rework our Dialog tree and the responses that we were specifying. We used the Analytics tab in the skill we were developing. On that Analytics tab, we would often review individual user conversations and see how our skill was handling user interactions. This led us to make changes to the wording that we used, as well as to the things we were looking for (in terms of entities) and what we were storing (in terms of conversation context). Very small changes can result in a big change in the end-user perception. Something as simple as using contractions (like “it’s” instead of “it is”), will result in a more informal conversation style.
The Analytics tab in Watson Assistant is interesting. It provides a wealth of information that you
can download and analyze. Our effort was
small, so we didn’t automate this analysis, but many teams DO automate the
collection and analysis of Watson Assistant logs. In our case, we just spent some time each
week reviewing the logs and looking for “holes” in our assistant (questions and
topics that our users needed answers for that we did not address), and trends
in our data. It has helped guide our
evolution of this solution.
This blog post identifies some best practices for developing
a chatbot with IBM Watson Assistant – but these apply to ANY chatbot
development, regardless of technology.
Practice #1 – Know the scope and growth path for your chatbot
Practice #2 – Know your roles – have technical people, developers, SME’s,
architects, and end users represented
Practice #3 – Have a solid dev/test/prod environment set up for your
Conversational assistant or chatbot
Practice #4 – Automate Your AI Testing Strategy
Practice #5 – Be Open Minded About Your Intents and Entities
Practice #6 – Inject Some Personality Into Your Assistant
Now that you have the benefit of some experience in the
development of a conversational assistant, take some time to dig in and begin
building a solution that will make your life easier and more productive.
I just worked with a customer this week on an issue that they had – and the solution didn’t seem obvious, so I figured that I would share it with a larger audience.
My customer has a couple of Watson Discovery instances in the IBM Cloud environment. These instances are supporting a cognitive application that they have acting in an expert assistant role – providing quick access to guidance and information to their associates. One instance is a Discovery instance which is supporting the production environment, the other is supporting their development and test environment. Both instances are Small sized.
They realize that they would like to save some money by using a smaller size instance for their development and test environment, where they think they only need an X-Small sized instance. They asked me for some guidance on how to do this.
This is not as simple a request as it might seem at first. The issue is that once you move into the Advanced sized instances (instead fo the free Lite instances), your Discovery instances begin to cost you money. They also can be upgraded from one size to a larger size, but they cannot be shrunk. Why? We can always expand and allocate additional resource to an instance, but we cannot guarantee that there will be no loss of data when shrinking instances. So we don’t shrink them.
It’s probably best to start by looking at the various sizes and plans. Looking at the Discovery page on the Cloud dashboard gives you some idea of the costs and charges, but it is not easy to read. Instead, I find that the help pages on upgrading Discovery, and Discovery pricing, are much more helpful. The table on each of these pages are informative and when they are combined, they give you the basics of what you need to know (this is accurate at the time of publishing – November 2019).
One other IMPORTANT difference between the plans is this: Each plan gives you a single environment that supports up to 100 collections and free NLU enrichments. The only exception is the X-Small plan, which will only support 4 collections. Also, note that you may also pay additional for news queries and custom models.
What Needs To Be Done
In order to “downsize” one of their Discovery instances from Small to X-Small, the customer will need to migrate the data themselves. What makes this difficult is that they will only have 4 collections available to them in the X-Small instance, instead of the 100 that were available in their Small instance. So they need to take these steps:
Create a new Discovery instance, with a size of X-Small.
Select the 4 (or fewer) collections that will be used in the new X-Small instance.
Re-ingest documents into the 4 new collections.
Delete the old development and test Discovery instance.
Creating a Discovery Instance of a Certain Size
The issue that my customer ran into was this: How do I create a Discovery instance of a certain size? When I look at the Discovery page on the Cloud dashboard , all I see is that I can select the Advanced plan – but no option on what size to use. So how do you do it?
It’s simple, and it’s outlined in the help docs in the section on upgrading Discovery. You first need to go in and create a new instance of the Discovery service with the Advanced plan. After you do this, the service will take some time to provision. You’ll need to wait patiently while this is done – it’s usually less than 2 minutes.
Now open your Discovery instance, by clicking on the link, and then choosing the “Launch Watson Discovery” button on the Manage page. You will now see the Discovery instance come up, and you will click on the small icon in the upper right corner of the browser to bring up a dialog that will allow you to “Create Environment”.
Then you will be able to select the SIZE of the Discovery instance that you want. You will see a dialog that looks similar to what is shown below:
See that you can choose from three different menus: Development (which shows the X-Small option), Small/Medium (which shows the Small through Medium-Large options), and Large (which shows Large through XXX-Large). Choose the size that you want, and then hit the “Set Up” button. This will create your Discovery environment, in the size that you want.
What If I Want To Increase the Size of my Discovery Instance?
In the above case, we had to do some specific actions to get a new instance created in a size that we wanted. We also learned that if we wanted to SHRINK in size, we needed to create a new instance and migrate data to the new instance.
What if I have been using Discovery for a while now, and I want to INCREASE in size? How do I do that? it’s actually pretty simple, and it’s also documented in the online help, in the section on upgrading Discovery. it just provides a link to the API, but not a lot of additional explanation. I’ll give you a bit more, so it’s a bit more clear.
If you look at the Discovery API reference, you’ll see a section on Update an Environment. This is the API call that you can use to upgrade your environment (and thus, the size of your Discovery instance). The API call is spelled out in this doc, and you can get examples for whatever language or way that you want to generate this API call by just selecting a type for the example in the black window on the right. In the example below, I chose to look at this in Python.
Just make sure to use the “size” parameter in your API call, and make sure that you use the right code for the size that you want (the label from the table earlier in this post). That’s all there is to it.
In the past couple of weeks, I have seen a few comments from my customers complaining about the lack of “sufficient” API documentation for the various Watson API’s. I used to like to point my customers to the Swagger API documentation, but I can’t seem to find it anymore. So I asked some of my fellow IBM folks if they knew where these pages were. They didn’t know, they just had some vague notion that they were no longer supported.
I miss those Swagger API pages – so I found out how to get them. The IBM development teams no longer host these pages, but you can generate them for yourself, whenever you want, but just following this short little guide.
Go Ahead – Get That Swagger
Step 1 – Figure out which API you want to generate a Swagger page for. Go to the IBM Cloud catalog, and select the service that you want to see. For the purposes of this example, I’ll go and look at the Watson Assistant service.
Step 2 – Get to the API Documentation page by clicking on the link titled View API Docs – as shown below.
Step 2a – You can skip all of this hassle by just going to the IBM Cloud API Docs page, and then selecting the specific API documentation page that you are looking for (which in our case is Watson Assistant v1). This is much quicker – and easier to bookmark and remember.
Step 3 – You are now on the Watson Assistant API (V1) page. Look for the ellipsis in the white text portion of the UI, as shown below, and click on it. Save a version of the API by selecting Download OpenAPI Definition. This will download a JSON file to your local machine.
Step 5 – In the Swagger editor window, select File -> Import File. Then select your recently downloaded Watson API JSON file (from Step 3).
You can now look at the Swagger API version of the Watson API documentation. This allows you to see all of the API calls for the service, along with the various parameters, and the responses. It also allows you to try to use the interface in an interactive manner. Pretty nice!!
With the end of the year quickly approaching, it is a great time to look back on the past year, and to look forward in anticipation for what is coming in 2018.
2017 was an interesting year. I saw an explosion in the development of chatbots of various different types. Some were very simple, others used both Watson Conversation and the Watson Discovery service to provide a deeper user experience – with an ability to answer both short tail and long tail questions. I saw a huge uptick in interest in basic Cloud approaches, and a lot of interest in containers and Kubernetes. I expect that both of these trends will continue into 2018.
In 2018 I expect to see the IBM Cloud mature and expand in it’s ability to quickly provide development, test and production computing environments for our customers. I also expect that more people will become interested in hybrid cloud approaches, and will want to understand some best practices for managing these complex environments. I am also excited about some of the excellent cognitive projects that I have seen which could soon be in production for our customers. I also expect that some of our more advanced customers will be looking at how cognitive technologies can be incorporated into their current DevOps processes, and how these processes can be expanded into the cloud.
I hope that your 2017 was a good one, and I hope that you have a happy and safe holiday season.
I’ve been working with IBM Watson and cognitive computing for a while now, and I have seen a lot of really amazing and innovative applications of cognitive technology by our customers.Customers take combinations of the cognitive capabilities of Watson, and combine them to provide capabilities, insight, and intelligence that is unique and game changing.
Often these efforts begin as small prototypes, showing the potential of the application being developed.They are bare boned, lack an attractive UI, and just show the promise of what could potentially be a new and powerful capability.They have no security, and don’t address any of the other non-functional requirements of any enterprise application.
As these prototypes evolve, they are used to generate interest in further funding their development – and this is where things begin to get tricky.Up to this point, the prototype has displayed great promise and raw capabilities with relatively little effort.The development team is either very small, or limited to people doing this in their “spare time”.Getting the thing 50% or 75% completed has taken a relatively short amount of time, cost very little, and has been done through the efforts of a small group of people.
Once the enterprise concerns need to be addressed, the costs and effort begin to balloon.Adding fully functional security to a project requires effort, as does the need to build a scalable and robust architecture.Management begins to think that the team has gotten dumb, as the “last 25%” of a project takes twice as long and costs twice as much as anticipated.It is at this point that many organizations abandon their cognitive efforts.
Some organizations do continue their efforts, and when they do, they are rewarded with a new cognitive capability that provides them with benefits that they cannot get with their existing tools and systems.So what makes these organizations successful, while others are not?I’ll explore this a bit in this series of blog posts, and discuss what you can do to improve your chances for success.
Getting Some Terminology Straight
This post is focusing on the organizational and business concerns that you will encounter in the development of a cognitive application. One of the largest sources of confusion between the business people and the technical staff is in the terminology used to describe the maturity of a cognitive application. Terms like “prototype”, “MVP”, “pilot”, “proof of concept” (or POC), and “production” get thrown around, and there are different understandings of what they mean.
I asked a bunch of Watson architects for their definition of these terms, and we all had some differences of opinion. So we discussed it, argued a little bit, and generally hashed it out. We ended up with this progression of application maturity:
Demonstration (Mockup) – this is exploring technology for understanding, developing, and managing requirements. This is something that is done so people can say, “Now I see what you mean…”
Proof Of Concept (POC) – this is a collection of services and code that show what cognitive services can do, with a functional but not fully polished UI, that show the basics of a technology. It helps determine the feasibility of implementing some cognitive capability, and focuses on the highest risk area. It is meant to a focus on figuring out what you don’t know and can only figure out by putting code on machine (and in the case of Watson content, by pushing data through our APIs). The POC is not really polished, and could be completely thrown out (depending on what you learn) before moving to the next phase.
Prototype – This has a more fleshed out architecture than a POC, consider it a big brother to a POC. This is a simple collection of services and code that show the basics of a technology, and the beginning of an architecture for a product. It should handle most of the primary use cases for a product. Still not fully polished, it could be considered a bit “rough around the edges”.
Minimum Viable Product (MVP) – this is the next stage. A collection of services and code that provide a fully functional use case, with a “happy path” and some alternative paths fully realized. It can stand on it’s own. A fully functional UI, but potentially a limited set of functionality and capabilities, that may only serve a subset of the intended user population. The core of the architecture has been completed, and there may be extension points for anticipated future enhancements.
Pilot – an initial release of the MVP application to a small subset of the anticipated user community. Used to help solicit end user feedback and to allow for the “hardening” of non-functional pieces of the architecture (performance, scalability, etc.). It’s where you learn to operationalize the application (even though you should have been doing this all along).
Production – A production system is the full system suitable for use by the complete target user group with the level of robustness and functional and non-functional capabilities that meets stakeholder expectations.
You might agree with this progression, or you may think we’re wrong. It doesn’t really matter, what does matter is that your technical and development teams and your business stakeholders have a common understanding of the terminology used and the progression that it implies. Otherwise you run the risk of having a huge misalignment of expectations. Technical people will know that a large amount of work is needed to mature and “harden” a prototype into an MVP, while your business stakeholders may see the prototype and think to themselves, “we’re only a week or two away from being able to make this available to everyone!”.
Have a real solution to a real problem
One of the key things to being successful is having a good idea for a cognitive application.Many people come up with “cool” ideas for applications, and even create new and inventive capabilities using the IBM Watson cognitive services.This is great – but it can lead to you burning time on projects that will never be funded or fully supported by your organization.These types of things should be done as “open source” projects.Develop your idea, learn more about the capabilities, and share your ideas with the world – while demonstrating your understanding of the cognitive space.
For those of you working for an organization that you want to fund and support your idea, you need to make sure that your idea provides a real solution to a real business problem.It should provide a competitive advantage if not save your organization money, or both.You should be able to roughly quantify the benefits of your cognitive application.If you want your organization to spend money to develop your application or solution, they have to have a reasonable expectation that your efforts will be worth the time and money spent. It also needs to have a clear benefit to the end user. If nobody wants to use this, then how much benefit can it really have?
The best way to do this is to build a two slide slide deck.I know, I know…. Powerpoint and slide decks are evil – but in this case your effort will save you time and provide a vision for your efforts in the future.Slide one should be a quick description of the problem that your cognitive application or solution will solve.Describe the problem, and highlight the negative impact that it has.Get quotes from leaders within your industry or organization that highlight the negative impact – and use them.Slide two should be a high level description of your cognitive application or solution.Don’t get too detailed – just describe how you will solve the problem, and then highlight the benefits of your solution. Don’t neglect to highlight the consequences of failing to solve the problem, it helps put the effort into context. Now make sure that EVERYONE on your team has access to this slide deck, since EVERYONE (developers, business analysts, project managers, and the receptionist) on the team should be aware of what your primary goals are, and should be able to communicate that.
I know that this is “Business 101” kind of guidance, but you would be amazed at how many teams are unable to explain WHAT they are building, and WHY it is important. Even worse is when the technical leadership thinks that they are building one thing, the business leadership thinks they are building something else, and the development team is actually building something that is neither of the two. It is also important that you keep this deck updated.Keep in mind that situations change, and that the benefits of your solution may change based on your competitors, the business environment, and other outside factors. The key here is to keep it simple, brief, and easy for EVERYONE involved to understand. You need to be able to “sell” this idea to anyone who questions your use of people, time and money – from the most senior Executive, to the most junior developer.
So now we have an idea, a mission for the team, and some succinct benefits that your cognitive application or solution can provide.That should help us articulate our vision and our mission to our technical stakeholders and our business stakeholders. We should have ALL of the people involved having a similar understanding of how things will progress, and what it is that we are actually building. In Part 2 of this series of blog posts, I will explore the use Agile development methods, and how to dig into the real work of developing a cognitive application.
Thanks to my fellow Watson architects for their input and observations on cognitive application maturity. The version that I shared in this blog post bears little resemblance to what I originally started with….
An ASK is a shared set of code and other assets that developers can use as a starting point for doing development of cognitive applications that use the Watson services. It also provides a sample application so you can see what an end user will see. What developers REALLY like about these ASKs is that they include real working code – and provide a starting point for some of the more common use cases being implemented with Watson (like the Answer Retrieval ASK). You can simply fork (or download) the GitHub project containing the ASK code, and see all of the code needed for a working cognitive application. I know that a lot of bootcamp teams will use ASKs as a quick way to put together the basic infrastructure for their cognitive use cases – it gives them a starting point that executes and works. They can then modify the UI and the business logic to fit their own needs.
This week our development team (led by Lakshmi Krishnamurthy) is releasing a new ASK. This ASK is implemented in Node,js, and it uses the Watson Conversation service to create a Text Message Chatbot. It’s pretty simple, and you can see a quick demo of the text message chatbot online. This can serve as a great starting point for your own work in creating a chatbot that serves a need that you have.