At some point after the second or third failed AI experiment, the conversation inside most growing companies turns to hiring. If the tools are not working, maybe the answer is a person who owns this properly.

It is a reasonable instinct and it is often the wrong first move. Not always. But often enough that it is worth walking through the comparison properly rather than by gut feel.

What each option actually costs

Start with the money, because the gap is wider than most people assume once you count everything.

An in-house automation or AI engineer is not just their salary. It is salary plus employer contributions, plus recruitment cost, plus tooling and infrastructure, plus the three to six months of ramp before they ship anything that changes a number in the business. Depending on your market that lands somewhere between roughly 1.5 and 1.8 times the base salary in true first-year cost. In the UK, US or Gulf markets, a competent mid-level hire in this space is not cheap, and a genuinely senior one is expensive enough that most 20 to 150 person businesses cannot justify it for a workload that is, in truth, a handful of workflows.

External help is priced differently. A scoped diagnostic engagement is typically a low four-figure sum. A fixed-price build on a single workflow is usually a few thousand. You are buying a defined outcome rather than a permanent salary line, and if the engagement goes badly you have lost a project budget rather than a year.

The important number is not which is cheaper per hour. External help is almost always more expensive per hour. The important number is cost per validated outcome, and that is where the comparison usually flips.

The evidence points one direction for smaller companies

There is data on this, and it is unusually clear.

The MIT NANDA study of AI adoption in 2025 found that organisations which bought from specialist vendors or built partnerships succeeded roughly 67% of the time. Organisations that built internally succeeded about one third as often. That is a very large gap, and it held across the deployments they analysed.

The reason is not that internal engineers are less capable. It is that an internal hire starts with no pattern library. They have seen your business and possibly one or two others. Someone who has implemented the same class of workflow twenty times knows within an hour which parts will break, which integrations are unreliable, and which processes are not worth automating at all. That knowledge is the actual product. It is difficult to acquire and it does not come with the job title.

Where an in-house hire genuinely wins

This is not an argument that you should never hire. There are clear conditions where internal is correct.

  • The work is continuous, not finite. If automation and AI will generate ongoing proprietary work for years, a permanent owner makes sense. If it is four workflows and then maintenance, it does not.
  • The domain knowledge cannot leave the building. Some businesses have process knowledge that is genuinely too sensitive or too intricate to hand over. That is a legitimate reason to keep it internal.
  • You already know what to build. This is the big one. An internal hire is excellent at execution against a clear brief. They are expensive and slow at working out what the brief should be.
  • Scale justifies it. Past a certain size the volume of automation work keeps a person busy permanently, and at that point the salary is cheaper than the equivalent in project fees.

Where an external partner genuinely wins

  • You need to know what is worth doing. Diagnosis benefits enormously from having seen many businesses. It is the highest-leverage thing to buy externally.
  • You need speed. A hire takes two to three months to find and another three to ramp. A scoped engagement starts in days.
  • The scope is finite and definable. One workflow, one department, one clear bottleneck.
  • You cannot yet justify the headcount. Which is the honest position of most companies in the 20 to 150 employee range.

The mistake that costs the most

The expensive error is not choosing wrong between these two options. It is choosing either one before you know what should be built.

We see this constantly. A company hires a capable automation engineer, and that person spends their first four months doing stakeholder interviews and process discovery, because nobody had done it. That is a senior salary spent on producing a document. The document is necessary. Paying for it that way is not.

The equivalent error on the external side is hiring an agency to build something specific before anyone has checked whether that specific thing is the constraint. You then get a well-built automation on a process that was not costing you very much, and the return never materialises.

Both failures share a cause, and it is the same one behind the 95% of AI pilots that produce nothing. Nobody mapped how work actually moves before committing money. The tool, or the person, gets bolted onto a mess that nobody had described accurately.

The sequence that works

For most businesses in this size range, the order that produces results looks like this.

First, get the map. Understand where work actually stops, which steps exist only to shuttle data between systems, and what each of those is costing per month in real terms. This is a short, finite piece of work and it should be bought externally because breadth of pattern recognition is what makes it good.

Second, build one thing. Pick the highest-impact, lowest-effort item on that map and get it working properly, end to end, with a named owner. One working automation changes the internal conversation more than any presentation.

Third, decide on headcount with evidence. By now you know how much automation work your business actually generates, and whether it is a permanent job or a series of projects. That question answers itself once you have real data, and it is a far better decision than the one you would have made at the start.

Where to begin

Our AI Audit exists for step one. In 14 days we map how work moves through your business and hand back the three to five automation opportunities worth pursuing, ranked by impact, effort and readiness, with the reasoning shown. It is $1,500, or SAR 5,600.

The roadmap is yours regardless of what you do next. Some clients hand it to a new internal hire as their onboarding brief, which is a sensible use of it. Some ask us to build the first workflow as a two-week fixed-price sprint. Either way you are making the hiring decision with evidence instead of a hunch, which is the entire point.