Medha Thapliyal

Sheet 03 Writing Revised October 2026

Sheet 03/05. 11 March 2026. 3 minute read.

Consulting without selling

Most AI readiness tools for non-profits are written by vendors. Responsible advice often means recommending no AI at all, or a spreadsheet.

03/05 · Essay

A 2024 survey of non-profits found that six in ten had no AI policy of any kind, and that only one in twenty had one the surveyors considered adequate. The same year, a network of technology non-profits benchmarked nearly a thousand sector professionals across six dimensions of AI readiness and found that responsible-AI practice lagged well behind enthusiasm for tools. I do not doubt either finding. I see both every month.

What troubles me is not the gap. It is who is offering to close it. Most of the readiness assessments a public health organisation will encounter are written by companies that sell what the assessment recommends. The questions are tuned to produce a score that justifies a pilot. The pilot is the product.

What independent advice looks like

I have tried to write down what I think responsible advisory work on AI looks like for a non-profit or a health program, partly to hold myself to it.

Scope the problem before the model. The first deliverable is not a technology recommendation. It is a map of the decision or handoff the organisation wants to improve, with evidence that it is actually failing. In my experience at least half of these maps end with a recommendation to change a form, a supervision routine, or a referral protocol. That is a good outcome. It is also one that a vendor-written assessment is structurally unable to produce.

Put data sovereignty first. Where would the data go, who could see it, and what happens when the contract ends. The WHO’s guidance on large multimodal models is explicit about procurement, and the first international standard for AI procurement was published in 2025. Neither is long, and both are more useful than most vendor security pages.

Insist on a silent phase. Any tool that would influence a decision about a person should run for a period producing outputs that nobody acts on, reviewed by a panel that includes the people who would act on them. The Nairobi clinical copilot study did this, and its error review is the most credible part of the work.

Write the exit before the entry. Data export in a usable format, a model-version policy, and a plan for what the organisation does if the vendor disappears. If a vendor will not agree to these, that is the finding.

Build capacity, not dependence. The measure of a good engagement is that the organisation can evaluate the next proposal without me.

Recommending nothing

The hardest part of this work is the engagement that ends with a recommendation not to proceed. Funders are often more excited about AI than the programs they fund, and a scoping note that says “the problem is your referral register, and a spreadsheet would fix it” is not what anyone pictured. I have come to believe that this is the most valuable thing an independent adviser can say, and that the ability to say it is the only real difference between advice and sales.

A public health organisation that adopts no AI this year because the foundations are not ready has made a good decision. One that adopts a tool it cannot govern, on the strength of an assessment written by the tool’s maker, has not. The sector needs more of the first kind of decision, and more people who are paid in a way that lets them recommend it.

Sources

03 · Adjoining

Keep reading.

  1. May 2026 ABHA before algorithms India's bet is that identity, registries, and consent come before models. I think the consent manager, not the model, is the real AI policy.
  2. Sept 2026 The 0.9 percent Rapid qualitative analysis appears in less than one percent of implementation science studies. AI may finally make it common. The question is what we trade for the speed.
  3. Aug 2026 The red alert and the ASHA A clinical copilot in Nairobi cut diagnostic errors most when it interrupted. What would an interruption mean for a community health worker with no clinical backstop?