On the weekend of 4 and 5 July 2026 we took part in Rise of the Builder: The Xero App & Agent Hackathon, run by Encode Club with Xero at Encode Hub in London. We built the first version of Customer Compass over the two days. It reached the finals and won the Cashflow Accelerator category, the category judged by Xero.
It has been live in the UK since July 2026, at customercompass.ai/uk. This is a short account of where it came from, and of what a weekend build does and does not prove.
A weekend is long enough to show that an idea works. It is not long enough to show that it can be trusted with someone's accounts.
Where the idea came from
In the week before the hackathon we asked accountancy practitioners we know a plain question: what functionality is Xero missing? Customer Compass is what we built in answer.
It is an AI decision agent for service businesses on Xero. Each week it reads your Xero data, surfaces the few client and revenue moves worth making (the client going quiet, the one you are under-pricing, the cash you are leaking) and drafts the first action for each. Reporting tools show you charts you have to interpret. Customer Compass gives you a ranked worklist and the first step, ready to review.
What a weekend build proves
We built it using AI-assisted development, which is how we work in any case. A weekend with those tools is enough to get from a question to a working demonstration that judges can assess. That is worth knowing. It means an idea can be tested in front of people who know the platform before anyone commits a budget to it.
It proves less than it appears to. A demonstration has to work once, on the data in front of it, with its builder in the room. A product has to work every week, on a stranger's books, with nobody watching. A hackathon does not test any of that, and a prize does not change it.
What a product needs that a demo does not
The version that is live today rests on decisions that matter more than the speed of the first build:
- Every metric, ranking and figure is computed in code and unit-tested. The AI writes the plain-English explanation and drafts the action. It never invents a number
- Actions run in Xero only when you approve them. Writeback is off by default
- Each action is idempotent, so an approval cannot fire twice, and is audit-trailed to Xero's own history
- It never moves money and never deletes records
- Advice comes with its reasoning, and nothing is presented as more certain than it is
AI tooling made the first version quick. It did not make those decisions for us, and they are the reason the product can be put in front of a business owner.
Where it is now
Customer Compass went live in the UK later the same month. It works in your Xero organisation's own currency, and there is a demo organisation to explore at customercompass.ai/demo. The product page sets out the 24 decision signals it reads and how the approval step works.
If you run a service business on Xero, or you advise people who do, get in touch.
