We looked at six real, publicly documented cases — three that worked and three that went wrong. The difference was almost never the quality of the idea.
We looked at six real, publicly documented cases — three that worked and three that went wrong. The difference was almost never the quality of the idea.
Wednesday, July 15, 2026
Wednesday, July 15, 2026
There is a lot of advice available about selling into large healthcare organisations. Most of it is about the pitch. Almost none of it is about what happens in the buyer’s building after you leave the room, which is where the decision actually gets made.
So we went looking for cases with a public record — real programmes, documented by regulators, published studies, company disclosures and named partners — where we could see both the mechanism and the outcome. Six of them. Three worked. Three did not.
The idea quality was roughly even across both groups.
The three that worked shared three features
A measure that improved for the person who authorised the work. In an Indian public-health screening programme, an AI radiology triage tool was deployed into facilities where the constraint was specialist time. It did not replace anyone. It changed what the existing radiologist spent time on. The buyer was a government with a screening target, and the number that improved — cases detected — was the number they were already reporting.
A metric the buyer had already defined. A hospital PVC recycling programme in Australia and New Zealand ran for seventeen years and reached more than 280 facilities. Nobody had to be persuaded twice, because the measure was tonnes diverted from landfill and the hospitals were already counting it.
A deadline that existed independently of the project. A tender date, a completion date, a scoring change, a regulatory cycle. Something that would happen whether or not your pilot did.
There is a lot of advice available about selling into large healthcare organisations. Most of it is about the pitch. Almost none of it is about what happens in the buyer’s building after you leave the room, which is where the decision actually gets made.
So we went looking for cases with a public record — real programmes, documented by regulators, published studies, company disclosures and named partners — where we could see both the mechanism and the outcome. Six of them. Three worked. Three did not.
The idea quality was roughly even across both groups.
The three that worked shared three features
A measure that improved for the person who authorised the work. In an Indian public-health screening programme, an AI radiology triage tool was deployed into facilities where the constraint was specialist time. It did not replace anyone. It changed what the existing radiologist spent time on. The buyer was a government with a screening target, and the number that improved — cases detected — was the number they were already reporting.
A metric the buyer had already defined. A hospital PVC recycling programme in Australia and New Zealand ran for seventeen years and reached more than 280 facilities. Nobody had to be persuaded twice, because the measure was tonnes diverted from landfill and the hospitals were already counting it.
A deadline that existed independently of the project. A tender date, a completion date, a scoring change, a regulatory cycle. Something that would happen whether or not your pilot did.
Whose number moves, and who would notice if it stopped?
Amplifai analyst desk
Amplifai
Whose number moves, and who would notice if it stopped?
Amplifai analyst desk
Amplifai
The three that went wrong shared one
In each, the exposure was invisible to the party carrying it, right up until it was total.
A patient monitor was found to be sending plain-text patient data to a hard-coded external address, in its default configuration, on hospital networks. The vulnerability was not exotic. It was visible to anyone watching what the device did on the network — and almost nobody was watching.
A hurricane flooded a single manufacturing plant that produced roughly 60% of one country’s hospital IV fluid. The exposure was not the storm. It was that most of the institutions depending on that site did not know it was one site.
And in a market where a national procurement programme cut device prices by more than 90%, several suppliers modelled the cut as a permanent state of the world. It was not. Prices rose at both subsequent renewal rounds; the ceiling has been raised twice; the company that bid the floor price now sells at nearly double it. The rules were published the whole time.
The three that went wrong shared one
In each, the exposure was invisible to the party carrying it, right up until it was total.
A patient monitor was found to be sending plain-text patient data to a hard-coded external address, in its default configuration, on hospital networks. The vulnerability was not exotic. It was visible to anyone watching what the device did on the network — and almost nobody was watching.
A hurricane flooded a single manufacturing plant that produced roughly 60% of one country’s hospital IV fluid. The exposure was not the storm. It was that most of the institutions depending on that site did not know it was one site.
And in a market where a national procurement programme cut device prices by more than 90%, several suppliers modelled the cut as a permanent state of the world. It was not. Prices rose at both subsequent renewal rounds; the ceiling has been raised twice; the company that bid the floor price now sells at nearly double it. The rules were published the whole time.
The question that sorts them
Not “is this problem real?” — the filings already tell you that.
Whose number moves, and who would notice if it stopped?
If you cannot name the person inside the organisation whose own measure improves, and you cannot name what would fail visibly if your thing stopped working, you do not have a pilot. You have a demonstration. They look identical for about nine months.
The question that sorts them
Not “is this problem real?” — the filings already tell you that.
Whose number moves, and who would notice if it stopped?
If you cannot name the person inside the organisation whose own measure improves, and you cannot name what would fail visibly if your thing stopped working, you do not have a pilot. You have a demonstration. They look identical for about nine months.
Get the Latest News
Be the first to discover insider tips and the hidden gems of your favorite destinations.
Share This
Get the Latest News
Be the first to discover insider tips and the hidden gems of your favorite destinations.