Recurring engagement
The build is the easy part. Six months later the data has changed, the model has been updated twice, and someone in accounts has quietly gone back to doing it by hand. Operations is the part that stops that happening.
From $3,000/month · month to month after an initial quarter
The problem
None of these throw an error. That is what makes them expensive — the workflow keeps reporting success while the output gets worse.
A workflow reads a field that gets renamed, a form gains a question, a supplier changes their invoice layout. Nothing errors. The automation keeps running and quietly starts producing worse output than the manual process it replaced.
Providers deprecate versions, adjust defaults and update behaviour on their own schedule. A prompt tuned in March against one model is running against a different one by autumn, and nobody re-tested it.
The moment an automation is wrong twice, someone starts doing it by hand again and stops telling you. The workflow still shows as running. The savings have already gone.
A new hire gets added to a group, a folder gets shared, a connector gets approved. The governance you configured on day one drifts, and the first anyone hears about it is during an audit.
What you get
Sampled output checked against ground truth on a defined cadence, so degradation is caught by us rather than reported by a customer.
Access, data classification, retention and connector approvals re-checked against the policy, not assumed to have held.
We track deprecations and behaviour changes across the tools you run, and re-test before they land on you.
Processes change. Automations get amended to match, in-scope, without a new statement of work every time.
A defined allocation each month for building the next thing, so the backlog moves without a separate project.
One person who knows the estate. Not a ticket queue, not a rotating pool.
Pricing
Per-user pricing makes sense for managed IT because cost scales with people. AI operations does not — it scales with how many workflows matter and how much they are trusted with.
For a business running a handful of automations that matter. Monitoring, governance review, change tracking and in-scope amendments.
Everything in Operate, plus a standing build allocation each month. For companies working through a backlog rather than maintaining a finished one.
For operations where AI is load-bearing across several functions and the work needs someone in the room rather than on a call. Scoped individually.
Build and operate are separate agreements. You are never locked into a monthly fee to keep using something you already paid to have built. If you would rather run it in-house, we document it and hand it over.
Questions
Because the failure mode is drift, and drift is continuous. A build hands you something that works today. Operations is what keeps it working when the data, the models and the process all change underneath it — which they will, within months.
No. They keep the estate running; we own the AI layer on top of it. Most clients keep the provider they have. If you would rather have one accountable party for both, we will take that on too, but it is a separate decision.
Most clients here did. Build and operate are separate agreements deliberately, so you are never locked into ongoing fees to keep something you already paid to have built. If you want to run it yourself, we hand over documentation and get out of the way.
Rarely, and we would usually say no. Operating an estate we have not mapped means guessing at what matters. If you have automations running and want them taken over, the assessment is how we find out what we would be signing up to.
Month to month after an initial three months. The first quarter is where the monitoring baselines get established, and leaving before that means paying for setup without the benefit. After that, thirty days notice.
We will not take on an estate we have not mapped. Two weeks, $2,500, credited in full against whatever comes next.
