Most assessment products are a scoring engine with a PDF on the end. Which works fine, right up until the moment someone wants to know why.
This is the first post in a weekly series focused on building MyAcumen from the ground up using AI-First principles. The decisions we made, why we made them, and how we went about it. Aimed wider than my usual sales audience.
You know the shape. You answer a few hundred questions, a scoring engine averages them, and back comes a report of bar charts and paragraphs about what people in your band tend to be like. The charts are fine. The trouble is that’s where it stops, so it’s useful for twenty minutes and then it goes in a folder.
Everything written around those charts was written before anybody knew anything about you. Fixed band tables mean a score of 41 and a score of 69 return the same paragraph, even though those are two very different people with two different problems. Your own specific question, the one you actually needed answered, was never in scope.
That’s why AI had to sit at the centre of MyAcumen rather than being added at the end. Four things forced it.
It had to be secure. This data is personal and it affects careers, so it has to live in one place with proper access control, not on spreadsheets across people’s laptops.
It had to be connected. Four assessments, plus 360 feedback from colleagues if you choose it, are only worth having if they can be read together. Kept apart, they’re four PDFs and a folder.
It had to be interactive, because the value is in your second and third question and nobody can predict what those will be.
And it had to go deeper than the charts. We build plenty of those, from years of knowing which questions people actually ask, and they report the facts well. What a chart can’t do is explain itself, and the explanation is the part you can act on. That has to be written from your own answers rather than picked off a shelf.
For anyone building something similar, here’s how it works.
Every response is stored at question level with row-level security, not just the rolled-up score. That one decision makes everything after it possible, because there is nothing underneath an average to ask about.
Scoring stays fixed. The same rules every time, no model anywhere near it, so the numbers hold still and we can show our working.
At report-build time your full record goes to the model, which writes your analysis from your real numbers rather than looking up the paragraph that matches your band.
The chat layer reads that same record, so the report and the conversation can’t contradict each other.
biggest issue was a major change – we built the prototype on a no-code database. But when we tried to plug in AI functionality, it just wasnt possible. we spent weeks exploring workarounds but soon came to the conclusion that it just wasnt going to work.
So we made the tough decision to totally rebuild on an AI First platform. Where the data and the AI engine could operate together, seamlessly.
What you get though is so worth it. You read your results at nine on a Sunday night and if something doesn’t sit right you can ask about it there and then. The answer you need comes back about you rather than about a generic group of people who scored similar to you. Best of all, you haven’t had to book a call with anybody to get this insight.
One test is worth putting to any assessment being sold to you. Ask what happens when you have a question the report didn’t expect. If the answer is “book a debrief call”, you’ve bought a document, not a diagnostic.
