Thompson Intelligence
Email Book a call
Connecting your data

You already have the data. It just doesn't talk to itself.

The accounts package knows what you invoiced. The inbox knows what was promised. The till knows what sold. None of them talk, and that gap is where the answers live. This joins them up, read-only, without you changing a thing.

Your data is not missing. It's just scattered.

Most owner-run businesses already hold everything needed to answer their hardest questions. The problem: the answer needs three systems read at once, and no human does that every Monday.

Xero · invoices & ledger Shopify · orders Stripe · payments Email · enquiries & promises QuickBooks · books Sage · accounts & payroll Google Sheets Excel · and its CSV exports HubSpot · or whatever CRM The till / EPOS The job book nobody has replaced The one that is only in someone's head
The last one is the real job. Every business has a rule that lives with one person: which customers get the old price, which supplier is always late, what "urgent" actually means here. Connecting the systems is the easy half. Writing that down so a machine can apply it is the half that makes the answers correct.
The systems audit

Before anything gets built, somebody has to look at all of it.

Most businesses I speak to are running five or six systems that each hold part of the same story, and nobody has ever put them side by side. The systems audit is me doing that: what you have, what each one is actually saying, where they contradict each other, and what you would get if they were joined up.

What I do

  1. List everything. Every system, spreadsheet and inbox that holds a number somebody relies on. This is usually longer than the owner expects.
  2. Read what each one holds. What it is authoritative for, what it duplicates, and what it is quietly wrong about.
  3. Put them side by side. The same customer, order or invoice in three systems, compared. This is where the mismatches surface, and they are always there.
  4. Say what could connect, and what it would give you. Named pairs, in order of what they are worth, not what is technically interesting.

What you get

A written document. Every finding referenced to the system it came from, every figure labelled with where it came from, and a ranked list of what to do about it. Written to be read by somebody who was not on the call, so it survives being forwarded to a partner or an accountant.

The deliverable

This is the document. Read one before you buy one.

A worked example of the full systems audit, run end to end on an invented homeware wholesaler running five systems. Every mismatch referenced to the two systems it sits between, every figure labelled with where it came from, and the connections ranked by what they remove. Same format as the spreadsheet audit, because it is the same job at a bigger scale.

  • 5 systems reviewed
  • 8 mismatches, all referenced
  • £14,280 unbilled, found
Open the example audit ↗ Download the PDF

No client data appears in it, and it prints to A4 exactly as it appears on screen.

First page of the worked example systems audit: Barrowfield Interiors, five systems reviewed, showing the summary and the first three findings. Open the full report ↗

Then you choose. Two of these four cost you nothing more.

Do nothing

Sometimes the answer is that your systems are fine and the problem is somewhere else. You will have paid £900 to stop wondering.

Hand it to your own people

If you already have someone technical, an IT firm or a developer you use, the report is written so they can do the work without me. It names the systems, the records and the fixes. That is a perfectly good outcome and it is why the document is written to be read by somebody who was not on the call.

One automation

If the real cost is a job being done by hand between two systems, that is £750 and it does not need a dashboard.

The dashboard build

If the answer is that you cannot see the whole picture, this is what the audit was deciding. £1,500, less the £300 credit.

Five moves, and none of them ask you to change systems.

You keep using exactly what you use now. Nothing gets migrated, nothing gets replaced, and reporting never writes back into your accounts.

01Where it lives

We list every place a number lives, including the spreadsheets and the inbox. Usually there are more than anyone expected.

02Read-only keys

Connections that can look and cannot touch. Set up in your accounts, revocable by you, at any time, without asking me.

03Reconciled

The unglamorous part. Records matched across systems until the totals tie. This is where duplicates and gaps surface.

04Your rules applied

Your vocabulary, your categories, your exceptions. The bit that makes it your business's data rather than generic data.

05Something acts on it

A dashboard, a briefing, a chase list, an alert. The plumbing is the same underneath whichever one you buy.

Uses thisSpreadsheets & auditsThe audit reads your workbook the same careful way, and often ends with "this should be connected, not copied".The audit → Uses thisAutomationsAn automation is this plumbing plus an action on the end: draft the reply, update the row, send the alert.Automations → Uses thisDashboardsThe full version: everything joined, reconciled overnight, and written up as answers by 07:00.The dashboard →

The route your data takes, start to finish.

The same nightly pipeline every build runs on. Read-only at the edges, one locked workspace in the middle, and a written briefing at the end.

Read-only by default. Anything that writes back is agreed with you first One locked workspace per client, isolation proven with a test Never used to train anyone's models

The platforms I build on

Nothing homemade where it matters. Every system runs on infrastructure used by millions of businesses, with a clear exit path if anything needs replacing.

Intelligencethe models doing the analysis
Claude logoClaudeAI analysis
OpenAI logoOpenAIAI models
Data & hostingwhere your workspace lives
Supabase logoSupabaseIsolated database
Vercel logoVercelHosting
Automationthe moving parts between your tools
n8n logon8nwhere the workflows run
Zapier logoZapierAutomation
Make logoMakeAutomation
Slack logoSlackComms
Notion logoNotionKnowledge
Your sourcesthe systems your data already sits in
QuickBooks logoQuickBooksAccounts data
Sage logoSageAccounts & payroll
HubSpot logoHubSpotCRM & enquiries
Gmail logoGmailEmail systems
Stripe logoStripePayments data
Xero logoXeroAccounts data
Shopify logoShopifyCommerce data

How your data is protected

Isolated to your company

Your data sits at rest in London (Supabase), walled off with row-level security. Client data is never mixed, pooled or shared.

Encrypted end to end

Data is encrypted while it travels and while it sits on disk, on every platform in the stack.

Never used for training

Your data is processed to answer your questions, not to train anyone's models.

You own everything

The dashboard, the systems and the data are yours outright. If we part ways, it all keeps working.

Access on your terms

You decide who logs in. Credentials live in your accounts wherever possible, not mine.

A written audit trail

Every build ships with documentation of what connects to what, and why.

Three companies touch your data. Here they are, and here is where they sit.

The section above is what happens to your data. This is who holds it and in which country, which is a different question and the one an accountant will ask.

1 · Supabase, LondonYour data at rest, in your own locked workspace. You get read-only access to your own database, included in the welcome pack.
2 · VercelRuns the dashboard and the overnight jobs. SOC 2 Type 2 and ISO 27001 certified, like Supabase.
3 · Anthropic API, USWrites the daily briefing. The one hop that leaves the UK, flagged rather than hidden. Not used for training, deleted within 30 days. If UK-only processing is a requirement, there is a route for that.

Connections are read-only by default, so reading your numbers can never alter them. If a job you commission needs to write something back, we agree that first and it is named in the build. Nothing changes in your systems that you have not asked for.

How the figures earn trust

A dashboard is only useful if you believe it. Mine are checked three ways before you ever see them.

1Approved platforms

Every platform is vetted before it touches client work: security posture, track record, and a clear exit path if it ever needs replacing.

2Verified figures

Figures are reconciled against your own numbers and cross-checked with an independent parse of the raw data. My first build matched £11M of revenue to the penny, and caught a £5,500 error the old software missed. These are real figures from a real reconciliation, kept anonymous at the client's request.

3Reviewed by a human

Nothing ships on autopilot. I review every briefing before it reaches you, and the findings document shows its workings, not just its conclusions.

Bring the messiest export you have.

If your data can be read, I can make it answer questions. If it can't, I'll tell you that on the call for free.

Book an intro call