Blog · 17 November 2026
Why we run BDQ Claude-first
How we use Claude to keep up with the work of a much bigger team, and what went wrong along the way.
Chris Bland
CEO and co-founder, BDQ
Blog · 17 November 2026
How we use Claude to keep up with the work of a much bigger team, and what went wrong along the way.
Chris Bland
CEO and co-founder, BDQ
We're a small consultancy doing work that, done properly, needs a much bigger team. We couldn't do it all well by hand, so we built ways to have AI do the repetitive parts, with a person checking what matters. Here's what that looks like, and what we've learned.
A consultancy our size still has to keep on top of everything a larger one does: our website and how it shows up in search, the analytics behind it, the finance inbox, and the service desks we run for customers. Each is a steady stream of small jobs that need doing properly, and none of them can be skipped.
So AI does the repetitive work, and our small team gets through more than it could by hand.
Mostly, a person works in Claude and asks it questions. When it needs live information, Claude reaches into the system that holds it over MCP (Model Context Protocol), an open standard for connecting AI assistants to other tools.
Our website is the clearest example. I used to read our Search Console and Plausible reports and work out what they meant. Now I ask in plain English, "how is the site doing?" or "is the Cherwell page catching anything?", and Claude pulls the live figures and explains them. When it thinks a page should change, it proposes the change with the evidence attached. I approve it or I don't, and Claude Code makes the approved changes.
A question about the site used to take me an hour or so to answer from the reports. Now it takes a few minutes, and I see the answer alongside our actual marketing material and plans.
I've written about that one in more detail: I stopped reading Search Console reports. Now I ask.
Right now, nothing changes on our website until I've approved it. That check is what makes it reasonable to let AI do the work at all, and it's what we'd expect from anyone else's AI too.
It hasn't all gone smoothly, and it's worth saying so.
In its first week, the analytics connection told me one of our blog posts had dropped out of search results. It hadn't. The comparison it ran was only looking at part of the data. It was fixed within days.
Before we trusted its numbers at all, we checked them against a Search Console export we'd taken by hand. They matched.
We now check every number on our own website against a source. That check found a statistic on our homepage that we couldn't back up, and it came off.
Check the numbers before you trust them. We compared ours with an export we'd taken by hand.
Start with read-only access. Our analytics connection can read our data and nothing else.
Keep a person on the decisions, and make the checking easy, so it actually happens.
You don't really know what it's like until you try it on your own questions.
Trying to do something similar? Talk to us. If you run a service desk, you can also see Claude working with Freshservice.
Book a 30-min consult
Tell us roughly what you’re facing. We’ll come back within one business day with thoughts, references, and a sense of whether we’re a good fit.
More reading
I used to work out what our search data meant by reading reports. Now I ask Claude in plain English, and it answers from the live figures.
Read story →Rovo, Freddy AI, Halo AI or Atomicwork? A platform-neutral guide to AI in IT service management and how to choose the right fit for your…
Read story → Help Desk MigrationDiscover why outdated support data can reduce AI accuracy and how a smarter migration strategy can improve the quality of your AI-ready…
Read story →We use cookies to keep the site working and to power features like our live chat. Analytics are cookieless. See our privacy policy.