Logic First, AI Second: How We Put AI Into a Business System

Everyone is throwing AI at everything. We take a simpler view: keep it simple, and use AI only where nothing else can do the job.

When we put AI into a business system, we build the human logic first. Then we use AI to enhance or manage that logic. Here is how that played out on a real build, and how we think any business should approach it.

Where we used AI in Protée OS

Protée OS sends outbound briefs to vendors. We wanted the quotes that came back to land inside the software, so users stayed in it. We also wanted the vendors, who sit outside the software, to keep working the way they already do. Everyone uses email, so vendors reply by email.

That left a gap: turning an emailed quote into data the software can use. This is where AI came in, as a simple and very specific data extraction step. It reads the quote and pulls out the details, and the system takes it from there.

Reusing the same logic for a lead funnel

Once the MVP was built, we needed a way to get people into the software quickly. We reapplied what the software already did, in a different order, and built the world's first image-to-event-scope pipeline. We used the same AI we had for the moodboard, tweaked its guidelines, and pushed the output through our logic engine. The result is an instant event scope from an image, with a user sign-up at the end.

The guardrails we put around AI

Every use of AI in our builds is isolated into a specific function. It does one job, and the rest of the system doesn't depend on it behaving like a person.

Because the human logic comes first, you can switch the AI off and the software still runs. We think that matters. If AI is restricted by governance in a couple of years, a business with a core foundation of human logic and its own logic engine is instantly ahead of everyone else. It's a market leader overnight: all we have to do is turn off the AI and the software still runs on the human logic. We're working on giving AI more to do, cautiously, and we're planning for both futures: one where AI is largely uninhibited and one where it's controlled.

What didn't work

Asking AI to parse a quote without the logic engine behind it. On its own it had no clue, and the results were random. The logic engine is what makes the extraction reliable.

Where AI doesn't belong

Anywhere near money. We use AI to extract prices from a quote, but never to take or make payments.

How to decide if AI belongs in a workflow

Ask whether you can rely on the outcome. If AI gets it right nearly every time, it enhances the workflow. If a person has to check the result, or redo it when the AI hallucinates, it's a hindrance, because you end up doing the work anyway.

So build the logic first, then bring in AI for the things that can't be done any other way.

If you're weighing up where AI fits in your own system, that's the kind of work we do in AI development. Book a discovery call and we'll work through it with you.