The Fastest, Simplest Way To Implement Data-Driven Gen-AI Solutions For Your Organization

I recently wrote an article about Accelerating Enterprise Gen-AI Use Cases which spoke a great deal about the hype behind Gen-AI, the value it can bring to an organization, and the importance of time-to-value when implementing this technology. This article is a follow-on to that post and is designed to be more of a “How-to” guide. Once you are done reading this article, I hope you come away with a better understanding of what it takes to bring your Gen-AI ideas to life.

Let’s start with WHAT NOT TO DO. Here are some of the pitfalls to avoid:

• Assuming a certain amount of hallucination or lack of control is acceptable
• The “One Bot to Rule Them All” approach
• Failing to validate results
• A Tech-team only approach

Pitfall #1: Assuming a certain amount of hallucination or lack of control is acceptable

This scenario should never be acceptable. You put your career at stake if you lack control over the results of your chatbots. The solution that you implement should offer consistent and reliable results. Is this even possible with a Gen-AI solution? The answer is yes but only if developed correctly. Avoid an approach that completely relies on unstructured data using a Vector database. Leveraging a structured, graph approach is a much safer bet. Why? You have greater control of what the Gen-AI solution will produce. Control and security are your friend. Learn more about Vector vs Graph here.

Pitfall #2: The “One Bot to Rule Them All” approach

It is amazing how the concept of having multiple, contextual chatbots isn’t really talked about in Gen-AI discussions. It's as if the concept of building a single chatbot is so hard that discussing multiple chatbots becomes a non-starter. It is critical to remember that an approach that offers greater control and more contextual results will always be preferred. Do you really know what is buried in your unstructured data that you feel safe enough to expose this to everyone? Avoid this pitfall by creating multiple, compartmentalized chatbots. When building a solution that derives data from structured data, this is a must have.

Pitfall #3: Failing to validate results

With Gen-AI you cannot do enough validation and testing. This is especially true given how new this technology is. Behind every production chatbot there must be a team of people monitoring, testing and improving the results. Recent headlines suggest that a runaway, hallucinating chatbot can cost real money. This is also true for internal chatbots that employees will become reliant on. Bad data will produce bad results which will produce poor decisions. This too can cause considerable damage. Avoid this pitfall by including validation and feedback mechanisms in the scope of your project. The easier you make this, the better the results.

Pitfall #4: A tech-team only approach

Simply put: a bunch of data geeks fooling around in the basement with Gen-AI will not warrant the attention of those with a budget. Your efforts must be grounded in real-world value. To avoid this pitfall, make sure you are partnered with stakeholders that bring real business pain to the equation. For example, how difficult is it for supply chain analysts to understand high risk scenarios or bottlenecks in material supply? How much time spent researching can be reduced by a data-connected chatbot? Per my previous post, the more specific you can be, the better. Find yourself a business champion to partner with.

The Proper Focus is Everything

Now that we know what to avoid, let’s now discuss WHAT TO DO. Here are some ideas that will help you find greater success:

• Focus on Time to Value
• Focus on Context
• Trust Leads to Adoption; Adoption Leads To ROI

What To Do #1: Focus on Time to Value

I cannot beat this drum enough. The longer your project horizons, the greater the chance of failure. This formula has proven time and time again with the Iron Triangle. Time impacts cost, scope, and quality. Therefore, focus on initiatives that bring results in weeks versus months. Building the “One Bot to Rule Them All” is naturally a recipe for disaster. Instead focus on a limited, more agile approach. Seek out tech partners that provide a broad set of capabilities that can be implemented quickly. You are looking for simplicity and repeatability. Anything that requires code (e.g. Python) should be avoided at all costs.

What To Do #2: Focus on Context

Developing contextual chatbots that will be leveraged in a variety of ways and implemented in multiple scenarios sounds difficult but it really isn't if you come at it from the right approach. Naturally you will need the right tech partners to achieve this (ahem… Process Tempo). Using our Supply Chain example, imagine a supply chain “aware” chat bot that does not require any special prompting or querying magic on behalf of the end user. Now imagine this chatbot accessible from key supply chain dashboards. Users will be able to ask questions of the data they are looking at and instantly validate the results. Delivering this contextual capability will turn you into a rockstar! It is about applying this technology in a way that derives maximum impact. In this case placing a smart bot next to your data multiplies the value of both as it makes this interaction contextual. Once you try this for yourself, you will never go back to boring old dashboards again!

What To Do #3: Trust Leads to Adoption; Adoption Leads To ROI

So time to value is important but how much value are we really talking about? Chatbots must be implemented with a massive audience in mind in order to overcome the likely costs involved. Given that there will likely be multiple, competing Gen-AI initiatives going on simultaneously, how will your approach be different? Your approach will succeed if it is one that is focused on trust.

If you go to a restaurant for the first time and the service is horrible, you will likely never go back. The same is true for a shiny new chatbot that the “greatly admired and adored” IT department spent months delivering. All it takes is one bad experience to ruin things.

You build trust with your users in three ways: 1) You spend a considerable amount of time validating and perfecting the results; 2) you capture important feedback from your users; 3) You share information and operate with a high level of transparency.

With this said, here is what you need to do:

• Develop a strong validation capability by providing an environment that your team can use to monitor the chatbots performance, usage and key metrics. Does your project roadmap include this? If not, you had better add it. It is better if your team uncovers problems first!
• Implement a feedback mechanism that makes it easy for users to contribute their feedback, knowledge, and experience. Imagine if a user spots bad data that never gets fixed. The trust they have in you and your solution will forever be damaged. Now imagine if they are able to flag the issue and then actually see progress on its resolution. Imagine the trust that will be built up between the user and the team managing this environment! This can only be achieved by combining validation, feedback and workflow. Again, if this is not in your project roadmap, you need to add it. Users are already wary of chatbot hallucinations and are also well aware of data quality issues. You will need to overcome this double-negative.
• Broadcast your results both good and bad early and often. Developing dashboards that provide statistics on the use of your chatbots and how they perform will go a long way in securing future funding for your efforts. In a well-managed environment, you will be able to demonstrate a steady increase in adoption and a steady decline in negative feedback. This level of transparency and improvement is what creates career growth!

Here is what we learned: The Fastest and Simplest Way To Implement Data-Driven Gen-AI Solutions

By focusing on what to do, and what not to do, your chances of deploying successful data-driven Gen-AI solutions greatly increases. Implementing these solutions faster and simpler requires a broader perspective and a proper tech stack that can support this approach.

This is your must-have tech stack:

• A Large Language Model (LLM). Preferably a 3rd or 4th generation model that can be deployed in a secure cloud or within your firewall. Most organizations that we have worked with are, for the time being, reluctant to use SaaS solutions.
• A knowledge graph (versus a vector only approach). You will want to load your data into a structured graph for performance, consistency, control and security reasons. This is extra work but there is really no getting around it. Check out www.neo4j.com for inspiration
• A code-free means of creating multiple, contextual chatbots. Check out www.processtempo.com?
• A platform for testing, validating and analyzing key metrics. This platform will need to be usable by both technical and non-technical stakeholders
• Integrated workflow features that can capture and process end user feedback

Your roadmap must extend beyond implementing a Large Language Model (LLM) and a chatbot interface. It must be geared towards a much more robust and well thought out vision.

'The Roadmap for Greater Success

I have purposely used the plural form of “solution” because your vision should really be about repeatability. We are only scratching the surface when it comes to the benefits of Gen-AI and because of this you should focus on developing a foundation that will enable you to produce multiple Gen-AI solutions. This is a new space. You and your team should own it! To succeed requires the proper vision and the proper foundation. I hope we got you thinking!