A small team I was scoping this week, part of a larger group, asked me for two things on a discovery call. Not an AI strategy. Not a chatbot. Two specific jobs. One: somewhere sensible to keep their files, so anyone can find anything in seconds instead of digging through folders nobody agrees on. Two: something to sort their shared inbox, the info@ address everyone half-owns and nobody really runs.
The timing for them was perfect as they were just starting a new subsidiary of a family office, so they were a light nimble team associated with a larger family office.
That is it. And it is the perfect first thing to hand an AI operator, which is why I am writing down what the work actually is. This is automating the back office with ai, the unglamorous version that quietly earns its place, and it is the same two jobs almost every established business I meet needs doing. But don't know how to do, and likely don't want to do.
If you run a law firm, an accountancy practice, a family office, a property company or a small brokerage, you can already picture this without me describing it. Decades of documents nobody can find. A shared mailbox that three people watch and nobody owns. The business runs fine, it makes good money, and underneath it the filing is held together by one long-serving person's memory. That is who this is for. Not a tech startup. A profitable, ordinary firm that never got around to the back office.
Job one: one tidy place for the files
The single most important decision here is to stop organising by folders and start organising by tags. Deep folder trees are where shared drives go to die. Two people file the same document in two places, nobody can find anything, and the structure works against you.
What to do, not how:
- Pick one home. For a team already on Microsoft 365, that is a SharePoint site they are already paying for. No new software.
- Tag, do not bury. Set a small set of labels (client, document type, status, owner, date) and tag each file once. Then you can read the same library a dozen ways: by client, by what is signed, by who owns it, filtered, grouped, searched.
- Make the tags required. You cannot save a file without labelling it. That is the rule that keeps the whole thing clean after I hand it over, and it is the difference between a system people use and a drive that rots in a month.
- Set up a few saved views so the library presents itself cleanly instead of as a wall of filenames.
- Optional, and this is the impressive part: an AI you can ask in plain English, "what is the latest signed agreement with this client," and get the answer with the source. That needs a paid licence per person who asks, so it is an add-on, not part of the basic build.
The trick here is not to ask the humans to do the tagging. The humans come up with the tags at the outset of the project, agree on them and let the AI do the tagging. The moment humans are asked to do the tagging, what's the point?
How I run it as a service: I design the structure, move the existing files into it (the part that earns trust fastest, because suddenly things are findable), set the rules, and then keep it clean every month so it does not drift back into a mess.
Job two: an inbox that files itself
"Organise the inbox by content" means something reads each message as it comes in, decides what it is (new enquiry, existing client, a bill, a supplier, internal, noise), files it to the right place and tags it.
What to do:
- Read and sort on arrival. Each email is classified the moment it comes in and moved to the right folder.
- Send the unsure ones to a human. Anything the AI is not confident about goes to a review folder rather than being filed wrong. Never guess. This is the same discipline I use when automating the month-end close: the machine does the bulk, a person decides the edge cases.
- Get a daily digest. A short readout of what came in and how it was sorted, so the team trusts it and can see what needed a human.
- Start simple, grow into the heavy version. For most teams the first build runs on the automation tools already inside Microsoft 365 (Power Automate plus an AI prompt). When the volume justifies it, the same job graduates to a custom build (Microsoft Graph plus Azure OpenAI) that is cheaper per email at scale and gives more control.
How I run it as a service: I define the categories with the team in their own words, because their buckets are not mine, then I run it and tune it as the business changes.
The part nobody likes but everybody needs
Reading email and documents with AI is processing personal information, so it falls under whatever data-protection law the business operates under (Europe's GDPR, and the newer laws modelled on it around the world). Two honest truths I put on the table early. There is rarely a data centre in your own country, so your files already live in whichever region your account was set up in. And the law usually does not require the data to stay home; it requires you to stay accountable for wherever it goes.
So part of the job is choosing and writing down where the data is processed, keeping to a shared mailbox rather than reading everyone's personal inbox, and having the agreement on paper. Done properly this stops being a hurdle and becomes a selling point: a clean, documented answer to "where does our data go," which most small firms cannot give today.
Why this is the first thing, not the only thing
I am open about the strategy because I write this build in public. The files and the inbox are low-risk and visible every day, and once I am the person who runs them, I am already inside the system with the trust to do more. The bigger work opens up naturally from there: the reporting, the close, the board pack that used to take a week, the controller-type build that is the real engagement.
On price, it is a one-off build fee to set it up and move the files in, then a monthly retainer to run it. Enterprise work is custom and quoted to each firm. The point of this first job is not the margin on an inbox. It is earning the seat.
Two easy jobs. They are also the two that every established, profitable firm I meet is quietly drowning in. If that is you, start there.
Monthly Revenues $11,000 | Clients 2 | Prospects 1 warm lead | Team: Me with Jan on standby
Day 97 of 365.