The market is splitting around different kinds of work. The organisations that prove where AI earns its place before standardising will have more choice, less switching pain and a much better chance of making the investment pay.

I think a lot of organisations are making their first serious AI decision backwards.

They are comparing the major platforms, sitting through demonstrations, scoring features, negotiating licences and trying to pick the one that will become the organisation’s standard. It feels responsible. A large commitment deserves a proper selection process.

But the selection starts before the organisation has proved what it is selecting the platform to do.

There is a comforting idea underneath all this activity. Pick the right platform now, give the market a little time to settle, and a technical winner will eventually make the rest simpler. We have seen that happen in other technology markets. One approach becomes dominant, the awkward alternatives fade away and standardisation starts to look obvious.

AI is moving the other way.

Apple is expanding the AI that runs locally on its devices. OpenAI is developing specialised chips to run AI at provider scale. Both moves make sense for the very different work each company expects to serve. Their divergence is evidence of a market spreading out around different jobs, costs and operating needs, rather than closing in on one universal answer.

That changes the decision in front of a CFO, COO, Head of Technology or Head of Operations.

A vendor comparison measures the technology. It does not measure the amount of your organisation that has to change around it. That missing part includes the way work moves between people, the checks built into it, the information the platform needs, the skills your team must acquire, the suppliers you become dependent on and the cost that returns every month after the implementation team has gone home.

The platform is the small decision. The expensive one is how much of your business you bend around it.

Standardisation can be genuinely useful. One commercial agreement, fewer tools to support, common training and less duplication all have value. I have no argument with that. The value appears after you know which work deserves to be standardised, though, and that is where many organisations are getting ahead of themselves.

Choose a platform too early and its assumptions start leaking into the business. A process gets redesigned because that is how the tool works. People are trained around it. Connections are built. Controls are changed. A recurring cost becomes part of the operating budget.

Then the original use case underperforms, but the organisation has already made itself difficult to move. The platform decision survives because reversing it would be embarrassing and expensive, not because the work ever created enough value.

That is a nasty place to end up. You can become very efficient at work that was never worth reorganising the business around.

Now remove the promise that one universal winner is coming. Assume the market keeps producing better local options, larger provider services, specialist tools and different commercial models. A sensible organisation would stop trying to predict the final shape of the market. It would prove one valuable piece of work, learn what that work actually requires and make only the commitment needed to support it.

That is good news.

Fragmentation gives you choice. It lets sensitive work stay close, demanding work draw on larger services and narrow jobs use tools built for them. It also lets you benefit as prices fall and better options appear. An organisation that knows where AI creates value can change the technology underneath without rediscovering the business case every time.

This is not an argument to delay AI adoption. Waiting for the market to settle is pointless because it may never settle. Start now, but start with the work.

For the operational owner being pushed to select a platform, this means reversing the brief. Identify where delay, rework, manual handling or inconsistent decisions are costing the business. Prove that AI can materially improve one of those areas under real operating conditions. Then choose the smallest platform commitment that supports the result.

Do that and your platform choice becomes evidence-led. Skip it and you are asking the organisation to absorb a large change on the strength of a product demonstration.

Make the Work Earn the Platform

  • Name three pieces of work that consume real time, create avoidable rework or slow a customer outcome. Attach an operational owner and a measurable business result to each one.
  • Record the current position before introducing AI. Count the hours, elapsed time, exceptions, corrections, hand-offs and direct cost. Use the record as the baseline, not a general sense that the process is inefficient.
  • Choose one workflow with enough volume to measure and a consequence small enough to contain. Avoid starting with the most visible process in the business merely because it will attract attention.
  • Map every change the proof would force around the technology. Include the people, steps, information, checks, supplier dependence, training and recurring cost. Treat that total change as the real size of the commitment.
  • Set a 60-day proof period, a spending ceiling and a pass mark before the work begins. Define the improvement that would justify continuing and the result that would stop the effort.
  • Put real work and real users through the proof. Keep consequential decisions with a person until the evidence supports moving them. Capture the time saved after correction and rework, not before it.
  • Compare the outcome with the baseline at the end of the period. Stop the use case if the value does not clear the pass mark. Resist changing the measure to rescue an attractive tool.
  • Select the smallest platform, contract term and integration footprint that supports the proven work. Preserve access to the business information, process records and outputs needed to move later.
  • Repeat the exercise with the next workflow. Allow different work to use different technology until repeated evidence gives standardisation something solid to stand on.