44 answers, including the awkward ones.

Anything a person does by copying between tabs on a schedule: routing, follow-ups, reconciliation, reporting, filing. If the steps can be written down, they can usually be automated. If they can’t be written down yet, that’s the first problem to fix — and we’ll say so rather than build on top of a process nobody agrees on.

High volume, low judgement, and a clear rule for what happens next. Invoice intake, lead routing, order processing, onboarding checklists, report assembly. The more often it happens and the less it needs an opinion, the better the return.

When the process changes every week, when nobody can agree what the steps are, when volume is low enough that a person doing it occasionally is cheaper than a build that has to be maintained, or when being wrong is expensive and there’s no sensible way to review. We turn work down on these grounds.

Usually it replaces the worst part of several jobs — the copying, the chasing, the re-typing. Most clients redeploy the hours rather than the people. We’re not going to pretend that’s always the case, but it’s what we see.

That depends entirely on the process, which is why the discovery call starts by counting: how often does this run, how long does it take, how many people touch it. If the numbers don’t justify a build, we’ll tell you on the call.

Partly. The mechanical steps run automatically and stop at the point where judgement is needed, with everything gathered and ready for a person to decide in seconds instead of minutes. That is usually a bigger win than trying to automate the judgement itself.

There’s no price list; every build is scoped against the process it replaces. A two-step Zapier fix and a self-hosted agent platform shouldn’t cost the same. You get a fixed price and a fixed date in writing after a free 30-minute discovery call, before anything is built.

No. You pay once for the build. No retainer, no licence from us, no lock-in. Subscription suits the vendor and rarely suits you if the thing runs for years.

Only your own usage — API calls, platform seats, model tokens — billed by the provider at their published rate with nothing added by us. We’ll estimate it during scoping so there’s no surprise.

The workflows, the code, the prompts, the credentials, written documentation and a recorded walkthrough. The test we hold ourselves to: could another developer pick it up tomorrow and not need us? They could.

Buy hours when you need them. No standing agreement, no minimum. Plenty of clients never come back, which is the intended outcome.

A deposit to start and the balance on delivery, against the fixed quote. Larger builds can be split across milestones.

Sometimes. Small budgets suit narrow, well-defined builds — one process, one clear outcome. They don’t suit “automate our operations”. Come with the smallest annoying thing rather than the biggest.

A workflow follows steps you wrote. An agent decides which steps to take. Workflows are predictable and cheap; agents handle variety and ambiguity. Most good builds use both — a workflow for the spine, an agent for the messy part.

When they’re trusted with decisions whose cost of being wrong is high, and nobody checks. We put a human approval step exactly there, and use plain rule-based logic where determinism matters more than cleverness.

It’s real and it doesn’t go away. What changes the outcome is design: grounding answers in your own documents, constraining the output format, validating against a source of truth, and putting a human on anything irreversible. We’ll tell you which parts of a build carry that risk.

Yes — that’s RAG. Your policies, contracts, product docs and history become the source, so answers come from your business rather than the open internet. Quality depends heavily on how the material is organised, which is often the real work.

OpenAI, Anthropic, or open models you host yourself — chosen per build on cost, quality and where the data is allowed to go. We don’t resell model access, so we have no favourite to push.

Yes, where the models and your infrastructure allow it. It costs more to set up and less to run, and it’s often the answer when data can’t leave your environment.

Yes. Extraction, classification, matching against existing records, and filing. Accuracy varies by document quality — we test against your real documents during the build rather than promising a number up front.

Whichever fits the job. Zapier for simple glue and the widest app library, Make for visual complexity and branching, n8n when you want self-hosting, control and no per-task pricing, code when no platform fits. We don’t resell licences, so the recommendation is only about the build.

No per-execution pricing, your data stays in your environment, and no ceiling on complexity. The trade is that somebody has to run it — updates, backups, uptime. We’ll set it up and document it, but it becomes yours to keep alive.

It happens, and it’s a real risk with per-task pricing at volume. We design so the logic is portable where we can, and we’ll tell you during scoping if a build is going to be sensitive to it.

Usually. We audit what’s there, tell you what’s worth keeping, and quote the rebuild separately from the new work so you can see what you’re paying for. Inherited builds with no documentation sometimes cost more to untangle than to redo — we’ll say which it is.

Yes. Zapier to Make, Make to n8n, cloud to self-hosted. Not everything maps one-to-one, so we check what actually transfers before quoting.

Yes, when it’s the right answer — a scheduled script, a small service, a queue. Platforms are convenient until they’re a constraint.

Anything with an API, which is nearly everything. Beyond that: webhooks, scheduled exports, email parsing, browser automation and scraping where a system genuinely offers nothing else.

There are usually still options — scheduled exports, email or file drops, browser automation. They’re less robust and we’ll say so plainly, along with what could break and how you’d know.

Yes, and it’s harder than it sounds. Two-way sync needs rules for what happens when both sides change: which wins, what gets logged, how loops are prevented. Agreeing those rules is most of the work.

Designed for from the start — batching, queueing, backoff and retries. It’s the unglamorous part that decides whether a build holds at volume.

Usually, as far as that platform’s export allows. We check what can actually come across before quoting, so there’s no surprise on switchover day.

Yes. Live, on one screen, pulling from wherever the numbers actually live — instead of a spreadsheet somebody rebuilds every Monday.

The build alerts before you notice. Every workflow ships with error handling, retries and a notification path you choose — email, Slack, whatever you actually watch. Because you hold the credentials and the documentation, you’re never locked out of your own system.

Only if you want that as a separate arrangement. By default the build monitors itself and tells you. We’d rather you didn’t need us.

In your own accounts and the platforms you already use. We don’t keep a copy, and we don’t need one to do the work.

The engineers working on it, under credentials you grant and can revoke. After handover we no longer have access unless you keep us on. We’ll ask for the narrowest permissions that make the build possible.

Not by us. Where a build sends data to a model provider, we use API access with training disabled and tell you exactly what leaves your environment and when.

Yes, and you should. Testing is a stage of every build: you run it against real work and try to break it, and we tune until it holds.

Book the free 30-minute discovery call, or ring us. Bring the process that annoys you most.

A description of the process as it actually runs — not as it’s supposed to. How often, how long, who touches it, what goes wrong. Screenshots help more than diagrams.

Most ship in two to four weeks. A single narrow workflow can be days; a platform with several integrations and an agent takes longer. You get a date with the quote.

A discovery call, an onboarding call to collect access, and then testing. Testing is where your time matters most, because you’re the one who knows how the process should behave.

Access to the systems involved, the rules the process follows, and someone who can answer questions during the build.

Yes — US, EU and MENA. Calls are booked in your time zone, not ours.