Northgate is not badly run. It is a firm whose systems were each bought to solve one problem, and which now spend their days handing work back and forth between one another through people.
Across fourteen interviews the same shape came up in every department. A fact is captured once - a tenant's date of birth, a certificate expiry, a repair report, a rent receipt - and then a person carries it from one system to the next by hand. Nobody described this as a crisis. Several people described it as simply how the job works.
That is what makes it worth attacking. There is no failing system to rip out and no rescue to perform. There is a large, boring, measurable volume of carrying, and carrying is the thing machines are actually good at.
Fifty distinct opportunities came out of the sessions, grouped into nine projects. What follows is not all fifty. It is the order we would do them in, and why.
Everything else waits. These three are chosen because each one is self-contained, visible within weeks, and does not depend on the data work that comes later.
Three people are already paying for AI out of their own pockets and pasting who-knows-what into a consumer account. That is the actual risk position today, and it does not improve by waiting. Move Northgate onto a governed business tier with a data processing agreement, so tenant and landlord data can lawfully be used. Add scheduled automation so useful work runs whether or not the person who built it is in the office. Train everybody once, properly, and put the prompts that work in a shared place.
This is cheap, fast, and it converts an unmanaged risk into a managed capability. It is also the thing that makes the next eight projects legal.
Around 120 repair reports arrive each week through Fixflo, email and the phone, and every one is read by a person before anything happens. An hour of the facilities manager's morning goes on sorting them. Priority depends on the order things land in, so a leak can queue behind a dripping tap. Nothing chases the contractor afterwards, and 31% of jobs sit open past thirty days. Forty calls a week come in asking where a job has got to.
We triage the queue automatically - category, priority, property, contractor - and put it in front of the helpdesk as a queue to approve rather than a queue to sort. Chase-up runs on a schedule without anybody remembering. The tenant gets a status they can see, which is where most of those forty calls go.
The helpdesk keeps every decision. The machine does the reading and the chasing.
Gas safety, EICR, EPC, fire risk assessments and legionella are all tracked in a single spreadsheet, roughly 1,400 rows, maintained by hand by one person. His own summary of the position was that if he were hit by a bus, that spreadsheet is the compliance function. Nine gas certificates lapsed in the last twelve months - none for long, but nine is nine - and fire risk assessment actions sit inside the PDFs rather than on anybody's task list.
Certificates arrive by email already. We read them where they land, pull out the expiry and the remedial actions, update a live register, and book the renewal before the lapse rather than after it. The compliance officer stops being the system and starts supervising it.
Two further early projects - Arrears Engine and Tenancy Onboarding - are equally strong and equally ready. We have held them at second wave only because three simultaneous programmes is the most a firm of this size absorbs well.
Nothing here is decided by the machine. It reads the overnight reports, works out what they are, finds the property, picks the contractor the rules say to pick, and drafts the dispatch. A person still presses Approve. The two panels are the whole product: what needs you, and what was handled while you were asleep.
Six of the fifty opportunities cannot be built properly until Northgate can answer a simple question: where does a document live? Today the answer depends on the year it was filed and the person who filed it. SharePoint holds recent files, Qube holds some property documents, and a Dropbox account holds everything before the 2022 migration. Nobody currently at the firm knows with confidence what is in that Dropbox, whether it duplicates SharePoint, or whether anything in it is the only copy.
This is the gating project, and it is why Portfolio Memory sits in the middle of the plan rather than at the end. Service Charge Studio, most of Listing Studio and all of the Dashboard are downstream of it. Attempting any of them first produces a tool that is right about recent data and silently wrong about everything older.
This is also the honest answer to the question the board will ask, which is why the dashboard cannot come first. The dashboard is four days of work sitting on top of ten days of unglamorous data work. Built in the other order, it looks impressive for a fortnight and then nobody trusts it.
Twelve weeks, with something working in front of people from week three.
Each programme runs alongside the existing process until the people doing that job say it can be switched off. That is not caution for its own sake - it is how you find the fifteen edge cases nobody mentioned in an interview.
Day rate, no retainer, nothing paid before it is built. Software is bought at cost and billed on to you at cost.
Governed AI seats for the whole firm are the only significant new line, at roughly £25 per person per month. Fixflo, Alto, Qube, Goodlord and SharePoint are all already paid for, and seven of the nine projects need no new vendor contract at all. Model and API usage on the automations is itemised on the invoice at what it actually cost, with no margin - on this volume, tens of pounds a month rather than hundreds.
The register is the full picture and it is all yours regardless of what happens next - fifty opportunities, sourced and ranked, whether we build any of them or not.
If the board wants to start, start with the first wave. Seven weeks, three working programmes, and a decision point that arrives with real numbers attached.