I counted the tabs one morning. Xero, because someone owes me money. SharePoint, for the pipeline and the task list. Outlook, for enquiries. Bookings, for what's on today. Analytics, for the website. Cloudflare and GitHub, to remind myself what I actually run. LinkedIn, because that's where half my referrals come from.
Ten. Ten sign-ins, ten different ideas of what "recent" means, and only after all of that could I get to the question I'd actually sat down to answer: what should I do today?
The frustrating part is that nothing was broken. Every number existed. Every number was correct. Each of those products is perfectly good at its own job and entirely uninterested in anything outside it — which meant the job of assembling them into one picture fell to me. Daily. From scratch. And I'm not that reliable.
The problem wasn't the clicking
If it were just time, I'd have shrugged. Ten minutes of tab-hopping is annoying, not fatal.
The real cost was the stuff that only shows up when you put two systems next to each other. An invoice quietly ageing past sixty days while the project it belonged to was still humming along happily. A domain renewal that was fine, fine, fine, and then a fortnight away. A warm enquiry that got a reply and never made it onto the board. None of those are visible in any single tab. They're visible in the gaps between them, and the gaps were being watched by my memory — which is not a system, it's a hobby.
So the picture existed in exactly one place: my head, rebuilt every morning, never written down. That's fine right up until the morning you're tired, or travelling, or deep in a build and not looking.
What I didn't want to build
The obvious move is to pull everything into one database and report off it. I've built those. I didn't want one here, for a reason that took me a while to articulate properly: a second copy of your business is a second thing that can be wrong. And the first day it disagrees with Xero, you don't fix the pipeline — you stop trusting the dashboard. Then you're back to ten tabs, plus a dashboard nobody believes.
The other thing I didn't want was an eleventh tab. If the answer to "too many products" is "buy another product", something has gone wrong.
So the rule I gave myself was: a window, not a warehouse. The hub owns almost nothing. Every figure is fetched from the system that actually owns it, at the moment I look at it. If Xero says something different, Xero is right and the screen updates. The only things stored are the ones no other system holds — the list of everything we run, a map of how it connects, and the goals.
Sense, recommend, act — and building the brake first
What came out of that is a hub with three layers, and the order matters.
Sense is boring read-only plumbing, and it's most of the build. Pipeline and tasks from SharePoint. Cash and receivables from Xero. Enquiries, bookings and calendars from Microsoft 365. Traffic from Analytics. Uptime, domain expiry and platform headroom from the platforms themselves. All of it on one screen, behind one sign-in.
Recommend is one button. It bundles all of the above along with my goals and asks Claude a single question: given this, what should I do next? It comes back ranked, and every item cites the data it came from, so I can disagree with it properly.
Act is the narrow bit, and I built the handbrake before I built the engine. Anything the system wants to do — create a task, chase an invoice, draft a reply — arrives as a proposal. It sits there, inert, until I approve, edit or reject it. Approve and it happens for real, in the real system, and gets logged. That isn't a policy I've promised to follow; it's the shape of the code. The model can only produce proposals. Only the button executes.
The part I didn't expect
I assumed the hard bit would be the integrations. It wasn't — that's just careful plumbing.
The hard bit was the goals. An AI can rank things, but only against something you've actually committed to, and I discovered I hadn't. I had a vague sense of direction and a lot of opinions. Sitting down and writing monthly, yearly and five-year targets specific enough that a machine could judge against them was uncomfortable, took a couple of evenings, and was easily the most valuable thing to come out of the whole project. The dashboard was almost a side effect.
What actually changed
I'd love to give you a percentage. I can't, honestly, and I'd rather not make one up.
What I can tell you is that the morning reconstruction stopped happening. The picture is on the screen before the coffee. The things I used to catch by accident are now watched continuously, by something that doesn't get tired. And "what should I do next?" has an answer with its working shown — which means I can argue with it, which is the whole point.
The other thing that happened is that it turned into the clearest demonstration of our work we've got. When I explain to a client what "one screen for your whole business" means, I'm no longer describing something hypothetical — I'm showing them the thing I run my own business on. There's a full case study here, screen by screen, if you want to see how it's put together.
We're all monkeys. Tools make us advanced — but only the ones that fit in a head this scattered.