Drive the AI conversation with Reference Architecture

by Chris PehuraC-SUITE DATA — 2024/05/17

Email us at to see how we make your plans work.

The worst thing you can do for an AI solution is assume that it’s swappable. Worse, that it’s a swappable tech. Don’t assume you can just plug AI into your organization, turn the key, and expect it to work and run fine. Every time you put something new into your organization there’s always a benefit, and always a cost. And those costs, those downsides, need to be discussed and emphasized during all your conversations when you’re talking about AI. You lose something when you have AI. And people don’t talk about that.

There are three main conversations you can have about AI. You can have the business conversation. You can have the technical AI conversation. And then there’s a middle ground where you talk about how AI is used. This a conversation being directed by conflicting priorities of business requirements and technical requirements.

Wrangling Strong Personalities

This middle ground conversation is the trickiest to have because there are always strong personalities involved. These strong personalities want to take over the conversation. They want everyone to take their direction. The problems come when that direction conflicts with the culture and personalities involved in making AI work at your company. Some of these personalities hold PhDs. Some of them are proclaimed experts. Some have written books. Instead of having a collaborative conversation these strong personalities degrade it into a turf war.

The only way to get difficult people to work together is by guiding them to stay on the same page. You do this by using something called a reference architecture. This is not just a picture slapped together. It has to look neat. It has to look credible. It has to be something that everyone will buy into and accept. People need to chose that picture and allow that picture to guide them through the conversation. Here’s an example of a reference architecture that I use to structure a conversation.

Seeing to Talking

I break up the conversation visually into high priority areas to talk about. (Analysis, Data, Market, etc.) Each area has components or topics that naturally come up when talking about that given area. And the color of each component means something for that area. Typically what is talked about for AI is the same at most companies. The conversations differ based on the priority of the topics being talked about. At your company when talking about the AI engine, the most important thing might be the model or the specific details of the AI hardware. For Data, it could be Big Data or the Data Lake. You’ll find components stay the same, but their color and position in the diagram changes.

Blue — What is Most Important?

Blue is something that is considered the most important for that area. What’s a good blue for you might be different at your company. Hybrid solutions I feel is key when you’re talking about AI engines. All the best AI are cocktails of different AI approaches. Large Language Models, a hybrid solution, is a good example. Business Intelligence is important at a lot of companies when you’re talking about Data. Again, for your organization it might be different.

Gray — What is the Price Tag?

The gray components are topics about containing costs and removing impediments. If we go to the Sales and Production area, we see machines and tools are in gray. The right machines and tools help with productivity which is a cost reduction approach.

Green and Yellow — What makes us Competitive?

The green components are things that make your company competitive. It could be things like improvements. It could be things like specific procedures. Anything seen as key to making your company competitive. And then we have the yellow. This is not what just makes the company competitive. It does more than that. It is what makes your company excel and be leagues above the rest. What things the company does or can do to blow away the competition. In Strategy and ROI, that could be business models.

Red — What Stops Us?

Now, we have to talk about the red. These are the showstoppers, the things that could go wrong. And when they do, they go wrong big. These are things that impact just not just the area but potentially everything in the diagram. For Strategy and ROI that is risk management. Working with lawyers, regulatory, and litigation hammered into me that revenue, costs, and lawsuits hinge on risk management. I strongly stress risk management in everything, especially with AI design.

Make it Look Sexy

It is an art to develop reference architecture diagrams. I strongly advocate that you put as much as you can in the diagram without making it look cluttered. The diagram has to be attractive. It needs to be symmetrical, beautiful. Some people do a mixture of pictures with text for these diagrams. I shy away from that format because most people remember the pictures and forget the text. You want your reference architecture memorable.

Build Street Cred

After the conversations are completed, your next step is to provide a consolidated summary. This could be for a whitepaper, business case, or business proposal. You modify the reference architecture so it matches up with the summary. You want to make it look more credible and align up what people talk about and think. Some of the guidelines that I follow are the same guidelines on how you do visuals and visualizations. Things involving position, color, how your eyes navigate through the diagram. You want people to do simple comparisons between the different areas. You want people to make connections. That is what makes reference architecture diagrams extremely powerful and actionable.

Build Ownership

When using reference architecture don’t take one off-the-shelf that’s completely filled in. That’s not a good idea. You want to have one that is mostly blank, something that you could have people fill in during the group conversations. You want this because you want people to engage and have a sense of ownership in the diagram. Once people have that ownership they will fight tooth-and-nail to defend the diagram and the resulting actions and decisions based on that diagram.

Final Thoughts

It’s always good to plan how you’re going to position the reference architecture diagram. The best thing you can do is deep research into the company. Be very prepared before you start using the diagram as the conversational center piece. Because once these AI conversations start you don’t have a second chance with that diagram. You want to have a good idea of what should be in each box. That will allow you to nudge and nod people in the right direction. You may not get everything that you expect but it is always close enough. And the people involve will be more invested to make that diagram a reality.

Email us at to see how we make your plans work.