Asking bigger questions of your Maximo data, and AI Tester in early access

A plausible answer is easy to come by now. One you can act on, built on all the records that bear on the question rather than the first screenful, is still hard. Both of the things below are work on that gap.

Interpretos: widening how much Maximo data one question can reach

A reliability engineer wants the failure rate of a family of assets set against PM compliance over the last eighteen months, before Thursday's review. A planner has to tell the plant tomorrow which jobs are not happening, so he wants every open work order blocked on parts for more than fourteen days, by craft. A maintenance head wants the ten assets draining labour hours this quarter, before the budget conversation. None of those is an exotic question. What makes them awkward is that a sound answer has to cross years of records, and an answer drawn from the first screenful looks exactly like an answer drawn from all of them.

The difficulty is worth stating plainly. An AI model can only hold so much text in front of it at once. That limit is its context window, and it is fixed. A Maximo estate is not: work orders, assets and history pile up for years, and one question can touch far more of that than any window will hold. You cannot simply hand the model everything. Summarising a sample is worse than it looks: a summary of a sample reads like a finding about the estate and is not one. And whatever is done to get past the limit, the answer still has to be checkable against the live system, or it is just a number somebody typed.

That is where the last few months have gone. What it bought is reach. A question can now draw on far more than the on-screen sample of records, and reach modules the standard interface does not expose, read-only. The answer still arrives as it always had to: verifiable against Maximo, showing its working and citing its sources where the data allows. The difference is the size of question worth asking at all.

AI Tester: in early access, and onboarding teams now

A test manager has a release going in at the end of the month and a list of checks nobody has the hours to get to. She knows exactly what needs proving. What she does not have is a person free to prove it and write down what happened.

The hard part is not reading her script. It is following it through an interface the machine has not seen before, deciding whether what happened counts as a pass, and producing evidence a colleague who was not watching can check. A run that reports a pass and cannot show why is worth nothing on a release call.

So AI Tester is released, in early access, and onboarding teams now rather than waiting for a queue to form. You describe the check in plain English, and what comes back is an evidence pack: what was tried, what happened, and what it means.

Reply to this email with the question your reporting has never answered, or the check nobody has the hours to run, and we will say which of the two would take it and what that would involve.