I found out on a Thursday afternoon, during a meeting that was supposed to be about something else entirely.
We were twelve people. A small but serious product team — two developers, a designer, a couple of operations people, a handful of business-side folks who had accumulated SaaS subscriptions the way organisations accumulate Post-it notes: gradually, inevitably, without anyone quite remembering authorising most of them. I was the person nominally responsible for making sure we used our tools well, which in practice meant being the person everyone asked when something wasn't working and the person nobody asked before signing up for something new.
The meeting had been called to discuss our AI tooling strategy, which in 2024 had become a question every team above a certain size was trying to answer. We had ChatGPT Plus accounts. We had a Notion AI workspace. We had three people using Jasper for different content tasks. We had a developer who'd quietly started using GitHub Copilot. We had a marketing manager who'd purchased a Claude.ai subscription on her personal card and was expensing it monthly.
It was our designer, twenty minutes into the meeting, who asked the question that started everything: "Hang on. Are Notion AI, Jasper, and Claude all doing the same thing for us?" The room went quiet in that specific way that means everyone is simultaneously doing mental arithmetic and arriving at an uncomfortable answer.
The answer was: mostly, yes. With minor variations in interface and occasional differences in output quality, three separate subscriptions were addressing the same core need — AI-assisted writing and thinking — for different people on the same team, with zero coordination, at a combined cost of £340 a month.
That was just the writing tools. When we mapped the whole stack, the number was worse.
I spent the following month building a spreadsheet. This is, as I acknowledged to myself at the time, a very developer way to respond to discovering that spreadsheets are insufficient — to build a more sophisticated spreadsheet. But even as I was doing it I could see the shape of what was actually needed: not a static document but a queryable intelligence layer that could answer questions the spreadsheet couldn't.
Questions like: which of our current tools does this capability? Questions like: we need something that does X and integrates with Y and costs under Z per seat — what exists? Questions like: if we dropped three of these tools and consolidated onto one, what would we lose?
The AI tools market had exploded to three thousand plus products by that point, and it was changing every week. A static spreadsheet went stale faster than you could maintain it. What you needed was a database that was itself kept current, combined with something that could interpret and reason across that data in natural language.
stackaudit.co.uk is that thing.
The tool database was the first thing I built and it remains the most important thing I maintain. Three hundred and seventeen AI and SaaS products, each with feature-level data, current pricing, integration capabilities, and a last-verified timestamp. It's scraped weekly — not because the product names change, but because features ship, pricing tiers shift, tools get acquired, and the recommendation that was accurate three weeks ago may quietly point to a capability that's now been deprecated or superseded.
The anti-hallucination architecture was non-negotiable from the start. Every recommendation Claude makes in the stack audit comes from a verified database record injected into the prompt as structured data. Claude never recommends a tool from training memory. It receives the database query results as a <verified_tools> XML block, ranks and interprets them, and returns product_ids that can be validated against the database. If the query returns zero results, the tool returns a NO_DATA response. It does not improvise.
This distinction — between an AI that interprets verified current data and an AI that synthesises from training knowledge — is the entire product. The AI tools landscape from training cutoff is not the AI tools landscape today. The gap between them is where bad recommendations live.
The redundancy detector is the feature that converts the most users from free to paid, and I think I understand why.
There's something psychologically clarifying about seeing it laid out in black and white. Not "you might have some overlap" but "you are paying £840 a month for three tools that do the same thing, here are the specific capabilities they share, and here is a consolidation path that would save you £560 a month and lose you nothing material."
The first person I ran it for professionally — a COO at a forty-person fintech, over coffee, as a favour — sat across the table from me and read the output in silence for what felt like quite a long time.
"This is a board presentation," she said finally. "I can take this to our next board meeting." That was the moment I knew it was a product rather than a side project.
The ROI calculator came from the same instinct: operations people don't just need to know there's waste, they need to quantify the fix. The automation ROI calculator takes a manual workflow — hours per week, team size, average fully-loaded hourly cost — and returns a payback period and twelve-month net saving across three implementation scenarios. It also generates a 150-word CFO brief. That's not a feature I planned. It emerged from watching people use the tool: they would run the calculation and then immediately ask "but how do I explain this to my finance team?" The brief is the answer to that question, pre-written.
stackaudit.co.uk is a B2B product for operations managers, heads of technology, and AI leads at twenty to five-hundred-person companies. The problem it solves gets worse as teams get larger and AI adoption accelerates — which means the market is expanding every quarter without any effort from me.
The PLG mechanic that I'm most proud of is the shareable report. When an operations manager runs a redundancy analysis and shares it with their CFO, the shared link shows "made with stackaudit.co.uk" and a "run your own audit — free" CTA. One shared report inside a company Slack can trigger three to five team signups. I track this with PostHog referrer data and it's consistently the highest-converting acquisition source.
The business started with a team of twelve people and a Thursday afternoon meeting that was supposed to be about something else.
I wonder sometimes how many other Thursday afternoons are happening in offices across the country right now — the same quiet calculation, the same uncomfortable arithmetic, the same question that doesn't yet have a tool to answer it.
It does now.
stackaudit.co.uk is free to try. The AI Tool Selector runs 5 free queries per month. The full redundancy analysis is available on the Pro plan at £29/seat/month.