AI consultancy versus building it yourself: what makes sense for an SME?
AI consultancy versus building it yourself: that is the question the moment you have decided something needs to happen with AI. Do you hire advice, or put your own people on it? Both can work, and both go wrong for the same reason: the gap between whoever decides what should happen and whoever builds it.
We are Ubert, and we are not neutral — we both advise and build, so we have an interest in that third way. That is why this page also states honestly when you are better off building it yourself, and when you actually want an adviser with nothing to sell. Read it with that in mind.
What each option really means
Hiring AI consultancy means someone from outside looks in, names the opportunities and delivers a plan. Building it yourself means your own people — or a new hire — work it out and put it together. In both cases the question is not "who is smarter", but "where does the knowledge of your process sit, and who turns it into something that works".
The mistake is treating it as a strict choice. In practice the winning setup is usually a mix: someone with experience who builds it, together with the people at your company who do the work every day. The process knowledge is yours; the building knowledge you hire in. Pull those two apart and you pay for the handover between them.
The hidden cost of building it yourself
Building it yourself looks cheap: no agency, no outside hourly rate. The bill comes later. Someone internal builds something that runs as long as they are there; when they leave, you have software nobody understands anymore. That is key-person risk, and at a company of twenty people it is real.
On top of that, almost everyone underestimates maintenance. An integration that works today breaks the moment your accounting package gets an update. AI models change, vendors alter their APIs. Building it yourself is not one project but an ongoing commitment.
- Key-person risk: the builder leaves and nobody knows the system anymore
- Maintenance nobody planned for: updates, changed APIs, new models
- Learning time: your people do it on the side and make the mistakes an experienced builder already knows
- No second pair of eyes on security, GDPR and the EU AI Act
- The work that piles up while your best employee wrestles with AI
The hidden cost of separate AI consultancy
Buying advice separately feels clean: first know what to do, then build. On paper that is right. In practice there is a handover between advice and delivery, and that is where it usually goes wrong. The adviser describes what needs to happen in your words; the builder has to guess what was meant and inventories everything again.
You pay for that twice. And advice written without knowing how it will be built often describes a more expensive solution than needed — someone who never builds does not notice that an integration that is one line on paper takes three weeks. The classic end result is an advisory report in a drawer, because there is nobody at your company to guard it for a year.
The third way: advice and building with one partner
There is a middle road that avoids the downsides of both: one partner who advises and builds, together with your people. No handover between thinking and doing, because it is the same person. No key-person risk, because it runs on standard technology with documentation and you get the code and your data. And the advice is realistic, because whoever has to build it will not dream up a three-week solution where one day will do.
The honest downside: someone who advises and builds earns from the delivery. That does not make them independent. That is why we also say "do not do it" when something delivers too little, and point you to an independent adviser when you need a judgement with nothing to sell — for a bank or a tender, for instance.
- No handover between adviser and builder, so you pay for the inventory once
- Realistic advice, because the same person has to deliver it
- No key-person risk: standard technology, documentation, and the code and data are yours
- Security, GDPR and the EU AI Act are part of the build, not an invoice afterwards
- Start small, measure, expand — instead of a big plan fixed in advance
When building it yourself is the right call
Building it yourself is sensible if you have the knowledge in house and keep it. If you have your own developer or an IT team that gets time for it and stays, building internally is often best: the process knowledge and the building knowledge then sit in one place, and you depend on no one.
It also works for something small and well-defined: a script that converts one file, a simple integration you click together with a standard tool. Hiring an agency for that is a waste. The line is at "does this need to keep running for years and grow with my business" — then maintenance becomes the real question.
- You have your own developer or IT team that gets time and stays
- It is small, well-defined and clickable with a standard tool
- It does not need to grow with your business for years
- You deliberately want to build knowledge in your own team, and accept the learning time
How to make the call for your business
Do not just cost the build, cost three years. Put maintenance, the risk if your builder leaves, and the time your own people spend next to the build cost. Then the comparison often lands differently than at first glance.
- Decide whether this is a one-off or something that must run and grow for years.
- Check whether you have the build and maintenance knowledge in house and keep it, not just now.
- Cost three years: building plus maintenance plus risk, not just the first bill.
- Weigh who has the knowledge of your process — that almost always sits with your own people.
- In doubt, do the free AI scan and a core session first; then you know what it is and decide with numbers.
Frequently asked questions
AI consultancy versus building it yourself: which is cheaper?
On the first bill, building it yourself is cheaper; over three years usually not. Building it yourself adds maintenance, learning time and the risk that your builder leaves; separate consultancy makes you pay twice for the handover between adviser and builder. So always cost three years, including maintenance, not just the build.
What is the risk of building AI yourself in an SME?
The biggest risk is key-person dependency: someone builds something that runs as long as they are there, and when they leave nobody understands it anymore. Maintenance is also almost always underestimated — package updates, changed APIs and new models break what used to work.
Can I not just separate advice and building?
You can, and sometimes you must (for an investment decision or a tender you want an independent judgement). But for ordinary automation, separating often works out expensive: the builder inventories again what the adviser already knew, and advice written without build knowledge often describes a costlier solution than needed.
What is the third way between consultancy and building yourself?
One partner who advises and builds, together with your people. The process knowledge is yours, the building knowledge you hire in, and there is no handover between thinking and doing. What matters is that such a partner also dares to say "do not do it" and hands you the code, data and documentation, so you are not locked in.
When is building it yourself the better choice?
If you have your own developer or IT team that gets time and stays, or if it is small and well-defined and clickable with a standard tool. The line is at things that must run for years and grow with your business; then maintenance becomes the deciding question.
What we do for this
Further reading
Projects rarely fail on the technology. They fail on messy data, a pilot that never ends and nobody who owns it.…
Buying is almost always the starting point, building it yourself looks more attractive than it is, and outsourcing only …
Advice without building produces a report; building without advice produces the wrong system. This is how you work out w…
Other topics
Torn between hiring and building yourself?
Do the free AI scan first: five minutes, a score and a price indication. Then, in a core session of half a day on your own shop floor, we look at what it would deliver for you and whether building yourself, outsourcing or building together is the wisest — with an honest go or no-go.