Punthub is built and owned by The Automation Agency. It is not a client engagement, and we label it that way everywhere it appears. It is here because it is the clearest evidence we have of something running unattended for a long time — which is the thing clients are actually buying when they ask whether a nightly system can be trusted.
A prediction product is only useful if the data behind it is current. That means collecting from several sources every evening, running the models, reconciling what comes back, and publishing before anyone looks in the morning.
Done by hand, that is an evening job, every evening, forever — and one missed night makes the product wrong rather than late.
We wanted to know whether we could build something that genuinely needed nobody, and then live with it long enough to find out.
Scheduled collection from six live data sources each evening.
Seven prediction models run over the collected data.
A reconciliation pass across the results, so figures agree before anything is published.
An overnight sync into the database the site reads from.
A front end that renders the output as prediction cards through bespoke plugins.
Loud failure: if a stage does not complete, it says so rather than publishing yesterday again.
Each of these is set out in how we count — what it measures, the basis, how it was checked, and where we have not finished checking.
Runs on: Python · scheduled collection · Supabase · Next.js. Listed because people ask, not because it is the point — the choice of tool follows the problem.
The £1,500 Process Audit maps where your team's time actually goes, ranks what is worth automating by return, and gives you a fixed-price quote against each one — in writing, before you commit to building anything.