Reading these three effect groups in sequence rather than in isolation is the whole exercise, and it is worth naming why the economic figures on their own would tell a board the wrong story. A $4.2M annual saving against a $6.8M build cost reads, on a spreadsheet, like a payback inside two years. It does not read that way once the institutional and workforce effects sit next to it, because the $4.2M figure assumes the case-management integration is finished, the appeal pathway is regulated, and the caseworker transition is negotiated, none of which is true yet. This is precisely the discipline the technology impact assessment method I developed with the Social Cyber Institute is built to enforce: an economic projection that has not been checked against the institutional capacity to deliver it is not a projection, it is a hope.

The institutional effects group carries the item I would flag hardest in a live review: the enabling regulation defines an appeal pathway for a human caseworker's decision but not yet for an AI-assisted one. I spent a decade as a dean deciding which academic programmes an institution should commit real money to, and the accreditation reviewers I answered to did not accept "we'll write the policy once the programme is running" as a sequencing choice; they treated it as the item that determines whether the rest of the submission can be trusted. A claimant denied benefits by an AI-triaged decision, with no defined appeal route specific to that decision type, is not a hypothetical gap NESB can close later. It is the first thing a legal challenge or a parliamentary question will ask about, and it should be closed before the pilot's first denied claim, not after.

The workforce effects group is where I would spend the most time with NESB's own leadership, because the honest framing, 85 to 90 of 140 roles reshaped rather than eliminated, is a genuinely defensible position if it is communicated early, and a genuinely explosive one if a caseworker first hears about it from a leaked memo or a union newsletter. The gap named here, that neither staff nor union representatives have been briefed, is not a detail to fix quietly before the next board paper. It is the single highest-leverage action available to NESB this quarter, cheaper than the case-management integration and more consequential to whether the pilot survives its first six months politically intact.

Dr. Jayarethanam Pillai · AI Economist & Founder, Praxora Lab

The reason this table runs six months, then two years, then five, rather than any other split, is that each horizon is set at the point where a different kind of evidence becomes available, and naming what has to break at each stage is more useful to a board than a single confident forecast pretending to cover all three. The six-month horizon is bounded almost entirely by execution risk: the case-management integration either finishes on schedule or it does not, and everything else on this page waits on that one fact. I would tell NESB's steering committee plainly that the six-month column is not really a forecast, it is a conditional statement, and the condition is already visible on the scored assessment as the lowest dimension on the page.

The two-year horizon is where the story changes character, from an execution question to a negotiation question. Net headcount holding roughly stable depends on a redeployment plan that does not yet exist in negotiated form, and I want to be specific about what "breaks" means here: it does not mean the technology fails, it means the institution fails to convert a technically successful pilot into a politically sustainable one. I have watched this exact failure mode in national digitalisation programmes I reviewed at UNDP, where a pilot cleared every technical milestone and still stalled at scale-up because the labour question was left for "later," and later arrived as a dispute rather than a negotiation.

The five-year horizon is the one I would defend most carefully against a sceptical questioner, because it rests on an assumption that cannot be tested until the other two have already played out: that the model's accuracy on the two pilot regions' claim mix generalises to the full national caseload. This is not a minor technical caveat. Claim mixes vary by region for reasons tied to local labour markets, industry composition and demographic structure, and a model tuned on two regions can perform very differently on a national distribution it has not seen. The assumption-that-would-need-to-break column for this horizon is, in my view, the most likely of the three to actually break, and I would rather NESB's board hear that from me now than discover it in year four after the fraud-detection saving has already been booked into three budget cycles.

Dr. Jayarethanam Pillai · AI Economist & Founder, Praxora Lab

Stress-testing a brief before it reaches a board is not a courtesy pass for typos and tone; it is asking the four questions a genuinely sceptical reader, an opposition legislator, a union negotiator, a finance committee member who has seen a pilot fail before, will actually ask, and checking whether the brief survives them without retreating into reassurance. I have sat on the receiving end of exactly this kind of scrutiny for two decades of accreditation reviews, and the brief that survives is never the one with no weaknesses; it is the one that names its weaknesses before the reviewer finds them.

The first question, about the borrowed fraud-detection figure, is the one I would want asked in every session I run, because a benchmark transfer is the single most common way an otherwise rigorous economic case quietly becomes fiction. The answer here does the right thing: it does not defend the number, it commits to re-measuring it against NESB's own pilot data on a specific timeline. That commitment, dated and specific, is what separates a stress-tested brief from a hopeful one.

The second question, about the actual cost of reversing course, is where most first-draft briefs fail, because "reversible" gets used as a comfort word rather than a costed one. Naming the $1.1M exit cost does something specific: it turns reversibility from a rhetorical safety net into a number a finance committee can weigh against the downside of proceeding, which is the only way that comparison is actually useful to them.

The third question is the one I would not let NESB's steering committee soften, because "no one loses their job in phase one" is true and also incomplete if it omits that no one has been told yet. A board that hears the reassuring half of that answer without the honest half is being handed a brief that will not survive contact with the union's own communications the following week.

My recommendation on this page, proceeding conditioned on finishing the integration and briefing the union before launch rather than after, is not a hedge. It is the sequencing that determines whether this pilot is remembered, in five years, as the moment NESB modernised its claims process, or as the case study the next reform effort is measured against and found wanting.

Dr. Jayarethanam Pillai · AI Economist & Founder, Praxora Lab