NISTA's Early Warning System is a genuinely sensible piece of engineering, and it will not fix the thing that is actually wrong with a fifth of Britain's biggest government programmes. That's not a knock on the tool. It's a limit built into what any early warning system, human or artificial, can ever see.
The tool appears in NISTA's Major Projects Annual Report 2025 to 2026, published 13 July. Read straight, the report tells a reassuring story. At the end of March 2026, the Government Major Projects Portfolio held 189 projects worth £924.2 billion in whole life cost. Of those, 29 were rated Green, 109 Amber, and 34 Red, an 18% failure-risk rate the report itself frames as the minority case. Eighteen projects moved from Amber to Green in the year; one jumped straight from Red to Green. Progress, on the numbers.
The new addition is the Early Warning System: as NISTA describes it, "a new digital AI tool" that "uses existing GMPP data to flag projects at risk of transitioning to a Red rating." It watches for movement across cost, schedule, milestones, risk, approvals, benefits and resourcing, and the report says it's "already proved helpful in forecasting the future health of GMPP projects," particularly the ones NISTA isn't actively watching day to day.
That's a real improvement over waiting for the next quarterly RAG snapshot. My problem with it is narrower and, I think, more important: every one of those six signals is downstream of the schedule the department submitted, and nothing in the tool's design checks whether that schedule was any good to begin with.
The AI Is Reading the Patient's Chart, Not the Patient
Think about what "schedule" and "milestones" actually mean as inputs here. They're whatever dates and durations a programme team entered into the reporting system that quarter. If those dates rest on logic that never really held, hard constraints doing the job that proper sequencing should do, risk allowance quietly folded into activity durations instead of shown as its own line, a critical path that was never genuinely resourced, the Early Warning System has no way to know. It can only watch those numbers move. A programme built on that kind of foundation doesn't necessarily show milestone slippage early. It can sit still, looking stable, right up until the point where reality catches up with it all at once, because nobody ever checked whether the plan was practicable in the first place.
This is precisely the territory an independent schedule review covers, and it's a different question from "has this number moved since last quarter." A DCMA-style check on the underlying programme, negative float, broken logic, out-of-sequence progress, excessive hard constraints, tells you whether the schedule can be trusted at all. We've written about what that actually involves in our DCMA 14-point check triage guide. None of those defects show up as a "risk" or "resourcing" signal in NISTA's six categories. They show up as a schedule that looks fine until it doesn't.
The Honest Counter-Argument
I want to be fair to the tool, because the strongest case for it doesn't need a perfect baseline at all. Watching relative movement is still useful even on an imperfect schedule. A programme whose milestones keep sliding quarter after quarter, whose cost keeps growing against its own baseline, whose approvals keep stacking up unresolved, is telling you something worth acting on regardless of whether that baseline was independently assured on day one. You don't need to know a schedule was perfect to notice it's getting worse relative to itself. That's a genuinely useful, genuinely different signal from a one-off health check, and NISTA is right to want it running continuously across a portfolio too big for anyone to read by hand every quarter.
But relative movement is a trailing indicator by construction. It can only fire once something has already started to slip. A schedule that was unsound from the day it was baselined, and simply hasn't been tested by reality yet because nobody has forced the issue, moves nowhere on any of the six signals until the moment it breaks. That's exactly the profile of a programme built on hard constraints instead of logic: nothing to flag, right up until there's a great deal to flag at once. Trend-watching and baseline assurance aren't substitutes for each other. They catch different failures, at different points in a programme's life.
What I'd Actually Want NISTA to Add Next
An early warning system watching the output of a schedule is not the same thing as an assurance process checking the schedule itself, and the report doesn't claim otherwise, it's careful to describe the tool as additional, not a replacement for existing scrutiny. My honest view is that the more valuable reform sits earlier in the pipeline than a smarter dashboard: some form of independent schedule quality check before a programme's very first RAG rating is set, so the Early Warning System is trained on data that was sound to start with, not just watched closely afterwards.
That's not just a public-sector point. Any organisation running programme controls on Primavera P6 or a comparable tool has the same exposure at a smaller scale: a schedule can pass every internal review and still be quietly unsound, because the people checking it are often the people who built it. If schedule quality genuinely concerns you, run the diagnostic yourself before anyone else's dashboard does it for you. Our free schedule health check applies the same DCMA-style logic in minutes, on your own .xer or .xml, with nothing uploaded, stored or shared.
For a fuller independent read, including logic quality and critical path integrity beyond what an automated check alone can see, that's the territory our Independent Schedule Assurance Review covers. Either way, the point stands: catching deterioration faster is progress. Catching it before it was ever baselined into the plan is the actual fix.