All posts

In-House or Partner - Building your AI operations system

Chloe hawking3 November 20265 min read

Somewhere in your business, someone has probably built something with AI that genuinely impressed the leadership team. It took a fortnight, it works, and the obvious next question has been asked in at least one meeting: if our own people can do this, why would we pay anyone else?

It's a fair question, and in plenty of cases the answer is that you shouldn't. The more useful question is a different one, though, and it decides whether that impressive first build turns into something the business can rely on: who owns this system every Monday morning from here on?

Why building in-house is the right instinct

The barrier to building has dropped dramatically, and the people who understand your operation best are already on your payroll. They know where the work gets stuck, which customers need special handling and which workarounds hold the place together. This is knowledge no outside firm walks in the door with.

Building in-house also keeps the learning inside the business. Every system your own people build makes them better at spotting the next opportunity, and for contained jobs close to a single team, such as summarising reports, drafting routine correspondence or tidying data, there's often no reason to involve anyone else.

So where does it get harder?

It gets harder earlier than most people expect. Getting from a promising demo to something the business uses every day means connecting it properly to the systems you actually run on, redesigning the process around it, ensuring it is secure and controlled, testing it against the awkward cases and bringing the team along. This is where plenty of internal builds stall well before launch.

The build is the visible part, and it's also the smallest part of the cost. Once a system is live, the work changes shape: the underlying models get updated or retired, an upstream process changes and breaks an integration without anyone noticing, outputs drift and need checking, running costs creep up as usage grows, and sooner or later the person who built it is hot talent thanks to this experience and moves on.

In business terms that adds up to continuity risk, operational risk and a running cost that rarely made it into the original business case. For a business of 20 to 500 people it lands harder, because the specialist skills needed to keep systems healthy are among the hardest to hire.

So when do you build, and when do you bring in help?

A simple test covers most situations.

  • Build it yourself when the job is contained, the risk is low, it sits close to the team that will run it, and you have people with the time and interest to keep it healthy.
  • Bring in a specialist when the work spans several systems, changes how a core process runs, touches customers, needs to go from demo to dependable quickly, or has to be maintained and improved for years.

Most established businesses end up with a mix, and the part people tend to forget is the handover. Agree at the start how much capability should move across to your own team over time, so that how dependent you are on any partner is a choice you've made.

Whichever way you build, some things belong to the business and should never be handed over. Every system needs a named business owner who is accountable for what it does and whether it's delivering. The knowledge of how the process actually works, the stewardship of the data and the decisions about what the system is allowed to do all sit with you, and so does the work of getting people to use it.

These roles are what make any partnership succeed, and a business that outsources them loses control of the outcome, however good its partner is.

Common worries about working with a partner, and how to settle them

A widely discussed 2025 study from MIT's NANDA initiative found that AI solutions bought from specialist vendors or built through partnerships succeeded about twice as often as purely internal builds.

However Most leaders weighing up a partner have the same three worries, and each of them can be settled before any work begins.

Will it actually deliver? Plenty of businesses have paid for advice that sat on a shelf or a tool nobody used, so the scepticism is earned. The real test is whether a partner is prepared to share the risk. At KARST, every piece of work is scoped against a commercial return, and where the work suits it we tie our fees to the outcomes it delivers, so we're measured on the same thing you are.

Will we be stuck? How much you depend on a partner should be a deliberate choice. Some businesses want a fully managed service, where the partner keeps the system healthy and improving. Others want to become self-sufficient, with their own team taking over maintenance and evolution and the partner's job being to get them there. Many settle on a hybrid, running day-to-day operations in-house and calling on specialist support for the bigger changes. At KARST we offer all three, agree the approach with you at the start, and expect it to shift as your team's capability grows.

Is our data safe? For us this is non-negotiable, and the standard is set by your needs. Where data is hosted, who can access it and how its use is recorded should all be agreed up front. 

A good partner raises these questions before you do and builds to your requirements from day one.

What to ask of a partner

A good partner brings something your own team can't easily build on its own, and that's breadth. They have usually seen a version of your problem before, in another business, another industry or their own time running operations, and they know which approaches held up and which ones fell over a year later. That outside view keeps a system from being built in a vacuum, shaped only by how your business has always done things, and it often surfaces a better way to run the process before any build begins.

If you do bring someone in, ask more of them than a working system. Ask who is accountable for the result once it's live, how they will build your team's capability as they go, and whether their fees move with the outcome the work delivers. The partner worth having will also tell you plainly which parts you should build and run yourself.

What all this means for your business

The cost of building has become a small share of the total, while the cost of owning, running and improving what you've built has stayed put. That shifts the decision away from who can build it and towards who will be accountable for it in a year's time.

The right model is the one that gives you a system you can rely on at launch and long after, with someone accountable for it every week, whoever built it. The businesses that settle that question early will get far more from their first impressive build than the ones still admiring the demo.

KARST is an operations and technology firm that designs and implements Intelligent Operating Systems to deliver commercial outcomes. Built by operators, powered by AI, measured in outcomes.

Written by

Continue the conversation

Want to talk about how this applies to your business?

Book a consultation session and we will work through it together.