The world of generative AI is evolving rapidly, and as such, the solutions and capabilities around how people leverage the technology are as well. The chatbot seems to be the biggest trend and many people's entry point into the generative AI space.
The first four months of the year were the era of the single purpose chatbot. Many of the chatbots developed during this time had a backing data store (database, vector store, etc.), and people could ask questions about that particular data set and receive natural language answers, essentially commoditizing the knowledge of that specific data source. As the industry is getting comfortable with this, the needs of users and the business are dictating that the chatbot does more and more things. We are in the era of the multi-use case or multi-intent chatbot.
With this new wave of use cases, I challenge you to change the way you view chatbots. The chatbot is no longer the entry point to access data; instead, it is the user interface layer that interacts with any workflow. That workflow is backed by any number of data sources and allows your user base to interface in natural language, reducing the training burden.
Let’s walk through an example of how one should think about a chatbot.
The Business Problem
No good solution starts with a solution - there has to be a problem. What is a pain point in your business - either internal or external? Is there a long lead time for a certain report? Is there a silo of knowledge that is continually the bottle neck? Is there a workflow that involves multiple teams doing multiple steps? Are there manual steps that slow down a process?
For the sake of this article - let’s pick the staffing process that any services company goes through.
The problem statement - it is slow, it involves multiple teams of humans, it requires specialized knowledge from those humans, and those humans are busy.
The process - the process kicks off when a new contract gets signed. The contract has resources doing a thing that requires skills. Upon signature, a resourcing team begins the coordination process of getting all of the resources, with the right skills, who are available assigned to the project. The first step is understanding the skills necessary for each role based on the contract. This understanding usually involves talking to the tech team to understand the contract. Once we know the skills - then they have to find available resources with those skills. In this scenario we have a skills database of self reported skills, which is never perfect. So even after looking up a person with skills or if we don’t find someone with the skills, input from the delivery leadership may be required.. All of this back and forth with the various people can lead to a lengthy process.
The Data
Generative AI solutions are only as good as the data that support them. Understanding the data is key to knowing if a solution will be successful.
In the business problem above we have two problematic steps - the interaction with the tech team, and the interaction with delivery leadership. If we can commoditize that siloed knowledge, then we can make this a single team problem to solve, streamline the process greatly, and reduce the burden on other teams.
The first problem - what are the skills on the contract. At Caylent our contracts are pretty detailed so all of that information is generally included. So the data source to solve this problem is the contract, but it may require a technical resource to interpret.
The second problem - who is available and whether they have the right skills. This is really two problems, but they need to come together as a single answer.
Who is available - we use a tool to manage everyone's assignments. This tool has the dates everyone is booked. This tool has a query interface for us to interact with the data. This data sounds easily accessible.
Who has the skills - this is harder due to self reporting of skill ratings. Some people are more modest in their ratings of themselves than others. Looking for a three star rating in a skill is not going to produce a group of people who are all the same level at that skill.
But, if we compare that person's three star rating against everything else they have self-rated, we can make some assumptions about that skill. If a given person has nothing above a three star - then we should classify anything at that level as their “highest rated skills”. Looking at the available data from a different perspective can provide us with additional insight.
The Solution
Just build a chatbot - done! Not quite.
For the first problem, we have clean data in an accessible place. We should be able to leverage generative AI to help with the specialized skill set needed to extract skills from a contract.
However, for the second solution our data is in two separate places. Further, we have already identified that some of that data isn’t the cleanest to base decisions on.
For this particular use case, a simple solution is a small data pipeline that can combine the two sets of data so we know when people are available and what skills they have in a single data source. We can also use this pipeline to enhance some of the skills data and add an additional category of “top rated” and “second rated” in addition to the five star and four star.
Once we have “clean” data, now we can use a chatbot to enable people to interact with that data. In our case, here we are enabling one team to ask in natural language, “what are the skills needed for the roles on this contract?” and once they have that answer, they can then ask, “what engineers do I have available next week with Python skills?”. With these two questions we have significantly reduced the touchpoints needed for this team to complete this workflow.
In Conclusion
Start with the business case - what is the problem that needs to be solved. Next, look at the data - does the data exist to answer this question and is that data clean enough to trust. And then finally, you can use generative AI as your interface to commoditize that data. Then, once you solve this problem, you repeat the process for another business case working it into the same chatbot making the bot even more powerful and giving it the tool to address multiple intents.
If you have a business case but are unsure about your data cleanliness or need help cleaning your data: Caylent has years of experience helping clients solve this very problem. Once you have clean data, if you need help building a chatbot as the interface for people to make decisions based on this data, we have 100s of client use cases worth of expertise to help you avoid common mistakes and build an extendable chatbot that can continue to drive business value. Caylent’s Multi-Model AI Companion marries your data with state-of-the-art AI models to create an AI companion that can dramatically enhance your productivity and streamline operations. Get in touch with our experts to learn how you can take advantage of AI.
Clayton Davis
