Headroom Data
Analytics and AI-safe data for app and subscription businesses

Numbers you can trust, for your team and for the AI tools now reading them.

Your dashboard, your billing and your AI assistant should all give the same answer. I check whether they do, fix it when they don't, and make sure your AI tools only see what they should.

Or email me here, at andy@headroomdata.com

The ten-question test Example · made-up figures

Your AI's answer, checked

Weekly questionAI saidChecked
Revenue last week, net of refunds✗ Wrong: refunds counted as sales £48,210£44,960
Trial to paid, last four weeks✓ Right, to rounding 11.8%11.6%
Day-7 retention of last week's new users✗ Wrong: reinstalls counted as new users 31%24%
Paying subscribers today✓ Right 3,4123,412
Churn last month✗ Wrong: failed payments left out 4.1%6.3%
2 right3 wrong The audit puts ten questions your business asks every week to your AI tool, and checks each answer against SQL I've verified on your own data.
What usually goes wrong

The numbers disagree, and now an AI is reading them too

The figures don't match

Analytics says one thing, billing another, the board pack a third. Refunds, reinstalls and failed payments are the usual reasons, and each one moves a decision.

AI tools can already see your data

Since 17 March 2026, enabling BigQuery also switches on its AI connector, and since August Mixpanel's is on by default for new free and Growth accounts. Most AI assistants can now be connected to a data warehouse.

Nobody checks the answers

An AI analyst answers every question with the same confidence, right or wrong. Until someone checks it against the real figures, you can't tell which is which.

Apps that hold children's or health data

The rules got stricter in 2026

If your app holds children's or health data, knowing which tools can reach it is where everything starts. The access map is the first page of the audit: every AI tool and connector, what it can read or change, and under whose terms. It's technical work on your data, not legal advice: your DPO or lawyer makes the legal calls from it.

What changed in 2026
  • 19 Jan 2026The Department for Education updates its safety standards for generative AI in education products. They bar any commercial use of personal data, model training included, without a confirmed lawful basis.
  • 6 Apr 2026NHS England's revised assessment form for digital health technology (DTAC) fully replaces the old one, which should no longer be used.
  • 24 Jun 2026The ICO reports on its audits of edtech providers. Among the gaps it found: incomplete data-flow mapping.
  • Aug 2026The ICO reports progress on its review of ten popular mobile games under the children's code, including whether profiling is off by default, and says it will consider formal enforcement action where appropriate.
Prices

Three fixed prices

Every job has two halves: your numbers, and the AI tools reading them. You always know the price before I start.

Start here

Audit

£1,200

Fixed price · five working days

Are your numbers right, what can your AI tools see, and do they answer correctly?

  • Tracking checked against what actually fires
  • Revenue reconciled to billing: refunds, double counts, failed payments
  • Retention and conversion checked, reinstalls and returning users included
  • The access map: which AI tools and connectors can reach which data
  • The ten-question test

A report ranked by money at stake, and a 30-minute call to go through it. The fee comes off Foundations if you book it within 30 days.

Included: one product, one analytics tool, one billing source, one warehouse, one AI tool for the ten questions, the last full month. Report within five working days of access. Full scope on request.

Fixed for Firebase exporting to BigQuery, with RevenueCat. Other set-ups: a fixed quote after the free call.

Book the audit

Analytics Foundations

£9,500

Fixed price · two weeks

One set of numbers everyone agrees on, and AI tools you can trust with them.

  • Metric definitions written once, for your team and your AI tools
  • Clean models in your warehouse: conversion, retention, lifetime value, payback
  • Billing reconciliation and a dashboard your team trusts
  • AI tools set up to work from checked totals, never customer records
  • A one-page AI data policy
  • A morning brief of yesterday's numbers, written by AI from the checked figures

Included: one product, one warehouse, up to two billing sources, eight metrics, one dashboard, one AI connection. Usually follows the audit, which comes off the price.

Fixed for Firebase exporting to BigQuery, with RevenueCat. Other set-ups: a fixed quote after the audit.

Ask about Foundations

Fractional Data Lead

£2,950 a month

One day a week · monthly

A part-time head of data, for when you need one but not five days a week.

  • A weekly review of your numbers
  • New questions answered inside that day, not queued for a quarter
  • The ten-question test re-run every month
  • New AI tools checked when you connect them

Included: one product, one fixed day a week. Three-month minimum, then a month's notice. Days don't roll over.

Ask about the fractional role

Prices exclude VAT.

How it works

From one number you don't trust to numbers you do

  1. A 30-minute call

    Tell me which number you don't trust and what decision is waiting on it. There's no charge.

  2. Access you control

    You give me a read-only account. I work inside your own systems, and nothing about your customers leaves them.

  3. The audit

    Within five working days of access: a report ranked by money at stake, and a call to go through it.

  4. Fix it, then keep it right

    Foundations fixes what the audit found. The fractional role keeps it that way.

Tools I work with: GA4 and Firebase, BigQuery, Redshift and Postgres, Looker Studio, Mixpanel, Amplitude, PostHog, Stripe, RevenueCat and Shopify, and the AI tools that read them, from ChatGPT, Claude, Gemini and Copilot to the analysts built into each product.

About

I build and run apps myself

I'm Andy Rippington. I spent nearly four years as a data scientist at First Touch Games, the Oxford studio behind Dream League Soccer and Score! Hero, whose games have been downloaded over a billion times. I worked across analytics, data engineering, machine learning and automation, mostly in SQL and Python.

Today I build and run my own apps through Half a Cow Games Ltd, and Headroom Data is the consulting side of that company. The tracking, warehouse models and AI guards I set up for clients are the ones I run on my own apps.

I explain things plainly. If a number is wrong, you'll know why, what it has been costing you, and what to do about it.

Questions

What people ask first

Will you need our customers' personal data?

No. I work inside your systems, through a read-only account you control, and nothing about your customers leaves them. Every job comes with a data processing agreement.

Do you use AI on our data?

Yes, on table structures, code, synthetic data and totals across groups too large to pick anyone out. That is part of how the price stays fixed. Never on records about individual people, unless you ask for that in writing. If your policy keeps AI tools away from even table structures and totals, I'm not the right fit, and I'll say so before you book rather than after.

What access do you need?

Read-only, and only to what the job names. For the audit on the fixed-price set-up: the BigQuery dataset holding your Firebase export, your RevenueCat account or its export, and the AI tool you want tested. The data processing agreement is signed before any of it is issued, and you revoke it when the report is in.

Is this legal advice?

No. It's technical work on your data and the tools that read it. Your DPO or lawyer makes the legal decisions, and the audit gives them an accurate map to make them from.

Where are you, and who do you work with?

Oxford, UK. I work remotely with app and subscription businesses anywhere, in English. Business clients outside the UK aren't charged UK VAT.

Who would I be contracting with?

Half a Cow Games Ltd, trading as Headroom Data. It's my company: I build apps through it, and I'm the one doing your work. Its registered details are at the foot of this page and on every agreement and invoice.

Contact

Start with the number you don't trust

Email me which number it is and what's waiting on it. I'll reply myself, and the first call is free.