After our conversation on Friday and some time with MMGH's work, here is my read: what I took from it, the opportunity I see, a high-level approach, and a few things I'd value your input on before we meet. Deliberately a direction — not a detailed plan. The detail is exactly what we should design together in Zurich.
MMGH is a small, senior-expert boutique in Zurich that advises the institutions that shape, finance, develop and deliver global immunisation — WHO, Gavi, UNICEF, CEPI, the Gates Foundation — turning evidence into decisions: market studies, full-value-of-vaccine assessments, investment cases, demand forecasts, evaluations. What struck me is that it competes on depth, credibility and independence, not scale — and that a single structured way of thinking runs through everything it produces.
That way of thinking is the Seven Ws — a repeatable investigation of how a product is and can be used, across seven dimensions. It is, in effect, MMGH's signature. Tap each one:
MMGH's real product isn't a document — it's senior judgement: decades of vaccine-policy expertise held in a handful of experts' heads. That is the moat — and the hardest thing in the firm to reuse, because much of it still lives in the room rather than on the page, and has to be re-applied afresh on every deliverable.
The opportunity I see is to capture and structure that method so it can produce credible, fully-cited first drafts — faster — while a senior always validates and owns the result. What makes me think this fits MMGH specifically: the first prize is not "an AI." It's the codified method itself — an asset that de-risks the firm and outlives any individual, valuable even if we never shipped a single automated draft.
Finding the right precedent and the current WHO/Gavi sources — across past deliverables, decks and the source sites — is skilled, manual work: often days of senior and analyst time before a word is drafted.
It retrieves the right precedent and the current sources from a curated, dated knowledge base — in minutes, in the same source-priority order every time.
Each section written into a blank document from scratch, figures and sourcing assembled by hand — slow, exacting manual assembly, with every audit-critical distinction held in one expert's head from the first draft to the last.
It drafts each section against the Seven Ws with an inline source on every sentence, and builds a Sources table and a Gaps & assumptions list as it goes. The house red lines — never a monoclonal antibody rendered as a vaccine, never a high-income price standing in for a Gavi one — are built into every draft, automatically.
The senior personally re-checks every claim and reworks the analysis — so a large share of the firm's scarcest time goes into verification and assembly, alongside the judgement only they can bring.
An independent check confirms every audit-critical claim against its exact source; the senior receives a cited draft and spends their time on judgement, not assembly — and still owns the result.
Before we talk about how it's built, here's the posture I'd hold it to — these are non-negotiables, not features:
Use managed, off-the-shelf platforms and fill them with MMGH's own intelligence. No custom infrastructure to own or maintain.
Every figure traces to a dated, resolvable source. Nothing is asserted from memory — essential for a WHO/Gavi audience.
It never decides. It produces a first draft; a senior expert validates, edits and is accountable for the result. Always.
Client-confidential material stays walled off, by design — and anything MMGH feeds the method is never used to train anyone's model. What a pilot may use versus what stays siloed is your call.
I wouldn't propose automating everything at once. I'd prove the idea on one real deliverable, judge it against an honest bar agreed in advance, and scale only if it clears that bar. And proving it small costs a fraction of committing at scale — the upfront ask on your senior people is deliberately bounded to a single deliverable, a way of working alongside how the firm runs rather than a change to it.
A WHO MI4A-style market study on RSV infant-protection products is a natural first case: it can be built end-to-end from public, current sources — so nothing hangs on confidential material while we prove the approach. your choice of case in Zurich
And the bar would be honest and yours to set — we'd hold the draft to three high-level numbers before anyone calls it a success:
Clear the bar and we scale to the other deliverables; miss it and we've learned — cheaply, early — exactly where the method still needs a human.
This is where your view shapes everything. Jot anything down right here as you read — then send it back to me in one click. No need to answer all four; whatever you note makes our two hours in Zurich sharper.
My hope for those two hours in August is concrete: to land on one real deliverable of your choosing — ideally a live or recent engagement — agree an honest bar for it, and leave with a clear read on whether this is worth a small first pilot. If the direction resonates, come with a case in mind; your early reactions would make the session far sharper. And if it doesn't, that's just as useful to know now.