We are not accepting clients at this time.

Book a diagnostic Book a call

Two offerings. One test for both.

Not "does it repeat" — agents handle variation. Could you write down what a good outcome looks like? If you could, it can be handed off. Finding which pieces of it are costing you most is what the discovery is for.

  • Custom automation
  • AI workflows

Two offerings, not two products. Most builds are some of each.

The split is where the work sits: steps a rule can describe end to end, and steps that need something read and judged first. For each one: what it looks like on a bad Tuesday, what we would actually build, and where it stops.

  1. Custom automation

    Today
    The same information gets typed in twice because the two systems holding it do not talk, and the Monday report gets rebuilt by hand from both of them.
    What we build
    The path between the systems, written down with the person who currently walks it, and then built: what starts it, what it carries, what your rules say to do with it, and what happens when a field is empty or a record does not match. It runs on the trigger rather than on somebody remembering, and every run leaves a plain-language record of what moved and where it landed.
    Where it stops
    The rules. The system applies the ones you gave it. Changing one is a decision, and it comes back to a person before anything runs differently.

  2. AI workflows

    Today
    Somebody reads what came in, works out what it is, digs up the history that belongs with it, and writes the reply, the record or the summary — forty times a week, slower every time the queue backs up.
    What we build
    A model does the reading, the matching and the first draft, working from your own documents and in your firm’s wording, and shows what it based the answer on. What it produces is logged against the job so the next person does not start over. When it is not confident it stops and hands the whole thing to a named person with the history already assembled, rather than guessing.
    Where it stops
    The judgement and the customer. A price, an exception, an unhappy client — those stay with the people who are already good at them, and reach them faster because nobody is buried in lookups.

Neither of these is bought as a module. A build is scoped against your work, in your systems, and priced once — so which side of the line it falls matters much less than what the discovery finds in your week.

How a build runs

What we need from your side.

Less than a software rollout, more than nothing.

  1. A few hours from the people doing the work

    Not just you. The dispatcher, the estimator, whoever quietly holds the schedule together. We need to watch the job being done, and later we need those same people to try to break what we built.

  2. Access to the systems the work already lives in

    Your CRM, your inbox, your job software, your drive. Read access to begin with; write access only where a system has been agreed to write. Your systems and your data stay yours — we build alongside them, never on a copy.

  3. Your exceptions, out of one person’s head

    Every business has the rule, and then the forty times the rule does not apply. The exceptions are the build. If they live only in one person’s head, we need that person for an afternoon and a phone number after that.

  4. One person who can decide

    Someone who can say "yes, that is how we do it" without calling a meeting. A fixed scope only stays fixed if decisions land in days rather than weeks, and that is the one thing we cannot do for you.

The diagnostic itself asks for none of it — free, 60 minutes, and one person who knows where the week goes. Everything above only starts if you decide to build.

Book a diagnostic

Nothing we build speaks to your customer without a person in the way.

Not as a default we might quietly move later. The checkpoints are named in the scope before the build starts, and the only way one comes out is if you take it out, for a step you name, in writing.

Every action a system takes is logged in plain language — what it did, what it read, and why it went the way it did. That matters most on the day something is wrong, because the question is never "did the AI do it" but "what exactly did it do, and to which job".

Checkpoints we do not remove

  • A person approves anything that goes to a customer.
  • A person supplies the numbers that are judgements — a price, a discount, a date.
  • A rule only changes because someone changed it, in writing.
  • Anything the system is unsure of goes to a named person, with the history attached.

Judgement, exceptions and the awkward customer stay with the people who are already good at them. The point is to give those people the hours back, not the decisions away.

"Our business is too specific for this."

It is a fair worry aimed at the wrong thing. What people mean by too specific is almost always the exceptions — the customer on a different price list, the county with its own permit form, the fact that Tuesdays are not like other days. Off-the-shelf software cannot hold any of that, which is exactly why you are still doing it by hand.

We are not fitting you to a product. There is nothing to configure and no template to bend, so the exceptions are not an obstacle to the build — they are the specification. We write them down with the person who knows them and build to those. That is why the discovery interviews the people doing the work instead of arriving as a questionnaire.

Where we say don’t

  • The process is broken, not slow. Automating it makes the wrong thing happen faster and more often. Fix it on paper first; it is usually cheaper than anything we would build.
  • It is worth less than the build. Agents moved this line and it is lower than it was, so rare work is not automatically out. But if the hours back do not clear the cost, you hear it in the discovery rather than after the invoice.
  • Nobody can say what correct looks like. Not work that varies — agents handle variation. Work where two people in the room disagree about the right answer. There is nothing to build against yet, and that is a decision to make, not a system to buy.
  • It needs a judgement call every time. Then it is not a definable outcome, it is the job. Those hours are the ones the rest of this is meant to protect.

A 60-minute diagnostic, then a fixed scope. Nothing owed either way.

Book a diagnostic

Free, no pitch deck. We work out whether there is anything here worth building — including when the answer is that there is not.