Build or buy: choosing between custom AI and off-the-shelf tools

Every leader looking at AI eventually reaches the same fork in the road. Should we buy an off-the-shelf tool, or build something designed around our own work? Both answers can be right. The mistake is choosing one by habit rather than by looking at the work itself.

What each option really means

An off-the-shelf tool is a product built for many organisations at once. You subscribe, configure it and roll it out. It is quick to start, the vendor maintains it, and it improves over time without effort on your part.

A custom system is built for your organisation specifically. It connects to your own systems, follows your own rules, and reflects the way your experts make decisions. It takes more effort at the start, but it fits the work rather than asking the work to fit it.

Five questions to decide

You do not need a lengthy evaluation to get to a sensible answer. Start with these questions about the specific workflow you have in mind.

• Is this work the same in most organisations? Writing meeting notes or summarising a long document looks much the same everywhere. Assessing a credit exception at your bank, under your policies, does not.
• Does it depend on your own data and systems? If the work needs information spread across several internal systems, a generic tool will only see part of the picture.
• Does it involve judgement that sets you apart? If the way your team makes this decision is part of why clients choose you, handing it to a generic product may flatten that advantage.
• What happens when it gets something wrong? If an error is a minor inconvenience, simplicity wins. If it carries regulatory, financial or safety consequences, you need controls designed for your context.
• Where must the data stay? If information cannot leave your environment, check whether a product can genuinely run inside it, not just promise to protect data elsewhere.

When off-the-shelf is enough

Buy when the answers point towards common, low-risk work. General writing help, transcription, translation, searching public information and personal productivity are all well served by mature products. Building your own version of these would usually waste time and money that could be spent on work that is truly yours.

It is also sensible to buy when you are simply exploring. Letting teams try general tools is a good way to build familiarity and surface ideas for where deeper work might pay off.

Buy for the work everyone does. Build for the work only you do.

When the work is specific enough to build

Build when the work is particular to you. Typical signs include:

• the process involves your own documents, records and systems talking to one another;
• the rules are yours, shaped by your regulators, your risk appetite and your history;
• people must approve outputs, and you need a clear record of what was proposed and why;
• you have tried a generic tool and found your teams working around it rather than with it.

In these cases, a custom system is not a luxury. It is the only way to get something your experts will trust and your risk team will accept.

It is not all or nothing

In practice, most organisations end up with both. Generic tools handle everyday tasks across the workforce, while a smaller number of custom systems take on the workflows that matter most. A custom system can also use commercial AI models underneath; building does not mean reinventing every component, only designing the whole around your work.

The useful habit is to make the choice workflow by workflow, rather than as a single company-wide decision.

How Sort AI approaches the choice

We are a forward deployed engineering company, so we build. But we will tell you when a workflow does not need us. When it does, our engineers embed inside your business, build from first principles on your real data, and deploy in your cloud or on-premise. We prove the system on one queue, judged by your experts, and then reuse what works across other teams.