Let me ask you something. In your organization right now, what percentage of your employees use AI as much as you do?
When I asked that question at a recent AI Accelerator Institute event, fewer than 10% of hands went up. And these were chief AI officers. People whose literal job is to get AI working inside their companies.
That gap tells you everything. We’re not struggling with AI because the technology isn’t good enough. We’re struggling because we haven’t figured out how to make it work for everyone, consistently, across real organizational data.
That’s the problem we’ve been trying to solve at PromptQL. And what we’ve learned along the way has fundamentally changed how we think about AI in the enterprise.
The goal most organizations are quietly working toward
Here’s where I think every organization with serious AI ambitions is trying to get to: more than 90% of employees using AI for more than 50% of their working day. Not as a novelty. Not as an occasional shortcut. As the actual way work gets done.
If you’re a chief AI officer, that probably sounds familiar.
It might even be your mandate. And yet, when you look around your organization, you’ll likely find that AI adoption is concentrated in a small pocket of enthusiastic early adopters, while the rest of your workforce either dabbles occasionally or avoids it altogether.
There’s a reason for that. And it comes down to accuracy…
When AI gives you a wrong answer, you notice. You correct it.
Maybe you try again, and then you quietly stop trusting it for anything important.
Multiply that experience across thousands of employees, and you’ve got an adoption problem that no amount of internal comms or training sessions will fix. People use tools that work. When AI works, they use it. When it doesn’t, they don’t.
So the real question is: why does AI on data perform so poorly, and what can actually be done about it?
For expert advice like this straight to your inbox every other Friday, sign up for Pro+ membership.
You’ll also get access to 300+ hours of exclusive video content, a complimentary Summit ticket, and so much more.
So, what are you waiting for?
The benchmark that changed how we think about this
Early in 2024, we were getting what felt like genuinely strong accuracy results with our customers. We wanted to benchmark those results properly, to have something concrete we could point to. So we looked at the existing industry benchmarks.
They were, honestly, pretty disappointing. Most of them were text-to-SQL benchmarks built on datasets that bore almost no resemblance to what real enterprise data environments look like. Simple schemas, clean data, single databases.
We were already working with design partners whose data lived across MongoDB, SQL Server, and Postgres simultaneously, with queries that crossed all three.
The benchmarks weren’t measuring the right thing.
