Four things we are good at.
We are not a general development shop, and we turn down work that does not fit. What we do is take manual handoffs out of companies that have outgrown their spreadsheets but have no appetite for replacing software that already works.
Where the time goes.
These come up in almost every company we talk to. The details vary. The shape rarely does.
| Handoff | Usually owned by | What it looks like |
|---|---|---|
| Form → CRM | Ops · daily | Enquiries get retyped into HubSpot. A week later somebody merges the duplicates. |
| Won deal → Billing | Finance · weekly | Closed deals are re-entered into Xero. By month end the two systems disagree and someone has to work out which is right. |
| Invoice → Payment | Finance · weekly | Somebody opens the ageing report and decides who to email. It is an awkward job, so it slides down the list. |
| Ticket → Engineering | Support lead · daily | Zendesk threads get summarised into Linear by hand. Most of the useful detail does not survive the trip. |
| New customer → Live | CS · per account | A checklist run across five tools. Whatever gets missed comes back as a support ticket about three weeks later. |
| Systems → Board pack | Ops · monthly | The same spreadsheet, rebuilt from the same three exports, every month. |
What we build.
Four things. We would rather be genuinely good at these than passable at everything.
Getting your systems to talk to each other.
Your tools all have APIs. Connecting them is the easy part. The work is in what happens when a request fails at 3am, or a record turns up without a field everyone assumed was mandatory.
Integrations usually break in the same handful of ways. A rate limit during a busy hour. A webhook delivered twice. A partial failure halfway through a batch, so half the records moved and half did not.
We make every job safe to re-run, so a replay cannot double-charge someone or create the same lead twice. When something does fail, it posts into whichever Slack channel your team already has open, rather than into a log file nobody has opened since it was set up.
The admin screen your ops team keeps asking for.
One page that does the job currently spread across a shared spreadsheet and four browser tabs. Built for the handful of people who will use it every day.
Internal tools have a different job to customer-facing software. Nobody needs persuading to use them, so they do not need to be beautiful. They need to be quick, hard to get wrong, and clear about what just happened.
We build the screen around the decision the person is actually making, which is not usually the same shape as the database underneath. It runs in your cloud, behind your existing login.
The Friday job that runs without anyone.
Reconciliation, chasing, provisioning, the monthly report. Work that happens on a schedule and only gets noticed when it does not.
This is where headcount quietly goes. It never appears on an org chart, but you can see it in a calendar: the Friday export, the month end reconciliation, the deck someone assembles by hand every Monday morning.
We automate the assembly and leave the judgement alone. Anything that sends money or emails a customer still waits for someone to click. Everything leading up to that does not need to.
Good at reading things. Bad at arithmetic.
We use models to read messy text, which is what they are reliably good at: enquiries, tickets, supplier documents. We do not use them to work out a number that ends up in a report.
Turning a rambling email into structured fields is a genuinely good use of a language model. Adding up a ledger is not, and companies that let a model do it end up presenting figures they cannot defend when someone asks.
Anything we ship with a model in it comes with a set of held-out examples, so you can see how accurate it is before it touches production and keep an eye on it afterwards. When it is not confident, the item goes to a person rather than getting a guess.
Quite often the honest answer is that a plain rule would do the job better. When that is the case we write the rule and tell you why.
Not sure which of these you need?
Most people are not, before the map exists. Working that out is what the first half day is for.
Start with the working session.
Half a day, fixed price. You come out of it with a map of your handoffs and a view on which ones are costing you most. Hire us afterwards or do not; the map is yours either way.