In-House AI vs Agency
The Advert Says One Role. It Contains Four Jobs.
Deciding to build AI capability inside the business usually starts with writing a job advert, and the advert quietly asks for four different people: somebody who can sit with operations and work out what is worth automating, somebody who can build it, somebody who can get it into systems that were never designed for it, and somebody who will still be testing it eighteen months later. The fourth is the one that decides whether any of it still works — and it is the one nobody enjoys.
Hiring in-house vs Using an agency
You own the capability · You rent the capacity
- What the first six months buy
Someone learning your business, which is genuinely valuable.Working systems, and no context - Knowledge of how you actually operate
Deep, and it compounds. This is the real case for hiring.Partial, and re-explained each time - Breadth of the role
One person covering scoping, building, integrating, testing.Different people for each - What happens between projects
They need work, and inventing it is a real cost.Nothing. You stop paying - Who does the unglamorous maintenance
Them, if the role was written to include it.Whoever the contract names - How you know the work is good
Hard. Few businesses can interview for this well.Hard, and you can see prior work - What it costs to be wrong
A hire, a notice period, and a rebuild.A contract ending
Row four is the one that gets skipped in the business case. An automation specialist with no automation to build is an expensive person looking for problems, and the problems they find will be real but not the ones that were worth solving. That is not a hypothetical failure — it is the most common way an in-house AI hire goes quietly wrong, and it looks like success for about nine months.
Hire when the work is continuous, not when the project is big
The clearest signal that hiring is right has nothing to do with project size. It is whether there will be automation work every week in two years’ time. If your operation changes constantly — new suppliers, new processes, new systems, a product that keeps moving — then the work is continuous, the person will always have something worth doing, and their accumulating knowledge of how you operate is worth more than anything an outsider can assemble. In that situation you should hire, and we would say so.
The mistake is hiring because one project is large. A big build is a burst of work, and bursts are what outside capacity is for. When the build finishes the burst ends, but the salary does not — and what follows is either a genuinely underused person or a search for new automations to justify the role. The second outcome is the expensive one, because automations invented to fill a diary are rarely the ones the business needed.
There is also a version of this decision that is neither option, and it deserves saying: often the right answer is to train somebody who already works for you. Someone in operations who knows every exception in the process is closer to being useful here than an external hire who knows the technology but not the business — because the technology is learnable in months and the operational knowledge takes years. That path is slower and it is frequently the one that sticks.
The shape we see work most often is neither pure: build the first systems with outside capacity while somebody internal sits alongside it, then hand over deliberately, with the outside help stepping down rather than stopping. The point of an agency is to become unnecessary in a defined way, and one worth using will describe what that looks like before you sign anything.
An automation specialist with no automation to build is an expensive person looking for problems.
Signals that point one way or the other
Read them against your next two years, not against your next project.
There is work every week, indefinitely
Hire. Somebody accumulating knowledge of how you operate will outperform any outside help within a year, and the role will not run dry.
One large build, then quiet
Do not hire for a burst. The salary continues after the build ends, and filling the gap with invented automations is how the role goes wrong.
Somebody internal already knows the exceptions
Train them. The technology is learnable in months; their knowledge of every edge case in your process took years and cannot be recruited for.
You cannot interview for this
Most businesses cannot, honestly. If nobody on your side can tell good work from confident work, an outside build with a named handover is the safer first move.
The work spans four disciplines
Scoping, building, integrating, testing. If you need all four immediately, one hire will be strong at two of them and the other two will slip.
You want both
Usually the right shape. Outside capacity builds the first systems with somebody internal alongside, then steps down on a stated schedule rather than stopping.
What people ask about this choice
When does hiring in-house clearly win?
When there will be automation work every week in two years, and when your operation changes often enough that knowing it deeply matters more than knowing the tools. In that situation the accumulating knowledge of how you actually work is worth more than anything an outsider can assemble, and we would tell you to hire.
Can one person really do all of it?
Some can, and they are rare and in demand. Most people are strong at two of the four — scoping, building, integrating, testing — and the one that slips is usually testing, because it is the least interesting and the least visible. That gap is worth planning for rather than discovering.
Is an agency not just more expensive per hour?
Per hour, yes. Over a year it depends entirely on whether there is enough work to keep somebody busy. A salary is a fixed cost that continues between projects; outside capacity is one you switch off. That is the actual comparison, and it is a question about your workload rather than about rates.
Should we train somebody we already have?
Often, and it is underrated. Someone in operations who knows every exception in your process is closer to being useful than an external hire who knows the technology but not the business — the technology is learnable in months, the operational knowledge took years. It is slower and it tends to stick.
What is the most common way an in-house hire goes wrong?
The work runs out and nobody notices for months. The person starts finding automations to justify the role, those automations are real but were not the ones worth building, and it looks like success for about nine months. Hiring for a burst of work rather than a continuous stream is the usual cause.
How is this different from your agency vs freelancer page?
That page is about buying from outside — who is on the other end, and what happens when they are unavailable. This one is about building the capability inside and whether the role is really one job. Different risk: that page is about continuity, this one is about whether there is enough work to fill a seat.