Hey {{first_name}}

For all the hype we hear about AI in health, apparently it also has a 10% chance of destroying humanity, so last week I chaired a panel on the “dark side” of AI at BMJ Future Health in London. And then for a bit of a palate cleanser I also spent two hours building Lego ducks? Check out From our Desk for the recap!

This issue is all about Evidence Ops, the part of the Evidence Stack that continually keeps your evidence current. This week's deep dive is about how to prepare for your next conference, in the style of Mission:Impossible.

Previous edition: The Narrative

One of my earliest memories is watching Mission: Impossible when I was about eight. The original series, from the 1960s.

I loved one trope above the rest. Someone walks into a dangerous situation with an earpiece in, and back at base a colleague with all the intel tells them where to go. There's a KGB agent at your two o'clock. They're moving in from the basement. Tom Cruise still wears one in the films, and it keeps him a step ahead.

I would love to say I'm Tom Cruise in this arrangement. But in reality I'm the nerdy British guy at the computer, played by Simon Pegg. Slightly less appealing but more realistic.

That earpiece became the idea for a tool that helps people at conferences.

AI has made it hard to trust that what you're reading or watching came from a person. A conference is where you can be 100% sure. You breathe the same air, eat the same food and complain about the same coffee as your clients, future prospects, business partners, investors, even your frenemies and competitors.

Organisers have noticed. Some now sell tickets on the strength of arranged face to face meetings, like if Tindr and LinkedIn had a baby. For example, a recent development is a so-called ‘Market Connect session’: once you know who is going, you send someone an invitation. For around $1,000 a meeting, a vendor is guaranteed 15 minutes with someone who might buy. Buyers can get their ticket paid for, sometimes with a stipend toward travel and hotel, if they agree to take eight or so meetings.

Now consider the startup that sends its founder for three or four days. Every minute from the pre-conference breakfast to late-night cocktails and karaoke becomes precious. Across our clients we have noticed a growing share of marketing budget going to in-person events: tickets, extra staff, a booth, branded giveaway items, a side event, paid meetings.

So to make best use of those resources, each quarter we review a client's goals for the year against the conference calendar, both scientific and trade events. We look at where they could submit their best data, where they could pitch to speak, where they should attend or sponsor, and where a press release or side event would make a bigger splash.

The first question clients ask is who they should speak to while they're there. When a conference is months away, you don't have the information to answer that. Conference Command is how we get it.

It checks the speaker schedule, the sponsors, the floor plan and social media on a regular basis, looking for anyone announcing they will be there. From that we build a map of who you should meet, based on your goals and who you're sending. The map gets richer as organisers publish more. A year out, with no venue chosen, we might guess: conferences like this tend to flip between coasts, last year was West, so probably East, probably early December. Closer to the date, we check the final programme daily.

We also read the LinkedIn pages of your competitors, partners and prospects. Where are they going, what are they presenting, who is hosting an event, and above all what are their key messages? A buyer posting that they're heading to Conference X to meet people with dermatology solutions is a lead. A competitor teasing next week's results of an AI dermatology paper on 1,000 patients at 95% accuracy is a different kind of news. If your booth says 87%, you will look inferior. Better to learn that before the number goes to print.

On the day itself, the brief gets very specific. At a scientific conference we often know from the agenda who is presenting a poster. The brief can say that Dr Chen will be in the exhibit hall, aisle 12, poster 36, at ten past ten, with a map, as well as who to see before and who to see afterwards. If three of you are going, we split the floor into a West Loop, an East Loop and a North Loop, so everyone asks their questions and gets the intel. What you learn goes into a joint Slack channel, then into a retro.

The plan also covers the logistics.

It tracks deadlines, such as when the early-bird rate ends and when Market Connect bookings open, because the early bird gets the worm. It shows how many of your targets are speaking and checks the attendee list against your CRM, since doing more business with existing clients is easier than finding new ones. If your geneticist can leave the lab for one day, it can tell you that Wednesday is genetics day and to send her then. It sets out who is responsible for what, the date to book tickets and the last date to cancel without a penalty.

Each person gets their own sessions and talking points, drawn from our claims bank of vetted evidence, so when someone asks how you differ from a competitor, the answer has already been agreed. Every target is tagged: worked with before, or prospect. We add their LinkedIn profile, because a name and a title don't help you connect and LinkedIn is one of the best ways we have found to do it.

The floor map shows your booth and your prospects clustered around it. For one client it also showed a direct competitor opposite their booth, which we spotted a month out. A competitive radar tracks direct and adjacent competitors and what they say in public. If a competitor like Epic is standing on stage at 10 AM on day one, on a panel with one of your prospects, you need to know how you differ from them a month ahead.

We also make show cards in your branding for everyone on your team who is going. Working with your business development team, we add a QR code that points to your booth, or into a system like your CRM, so only the people you want to meet get your time. They go on LinkedIn on a schedule: six weeks out, then two weeks out, with a countdown. There are templates for the day itself and a retro carousel for afterwards.

One company president told me she had planned to show up and wing it. After seeing the plan, she was 100% confident she would use her time better. A scientist who dislikes crowds and had no wish to "just network" got a different plan: find the answers to three questions, at these three sessions, and write them down. He went in to the event testing hypotheses like a scientist. That mattered, because he has to be able to say with confidence whether his technology is at the cutting edge.

Conference Command sits inside Evidence Ops, our hands-on conference support, and it is live with several clients. We are planning the next 12 months of conferences now. If you're going to planning conferences for 2027, book a call and we would love to give you a preview.

— Paul

On Friday 30 October I chaired a panel at BMJ Future Health in London on the dark side of AI. Saira Ghafur of Imperial College London, Tauseef Mehrali of Flo Health and Prof Susan Shelmerdine of the Royal College of Radiologists joined me. After a day and a half on how much good AI could do, we spent 40 minutes on what could go wrong. My key takeaways are in this LinkedIn recap.

Cybersecurity came first. If AI diagnoses, triages and reads scans, a hospital held to ransom, as in season two of The Pitt, has far more to lose when the systems go down.

Then deskilling. When the computers failed for two hours in Prof Shelmerdine's radiology practice, she marked scans by hand. The junior doctors didn't know how.

Then regulation. General-purpose tools like ChatGPT give medical advice without being medical devices. On 27 October, a family filed a lawsuit alleging that ChatGPT helped their daughter, who had epilepsy, reduce her medication and advised her not to tell her doctor. She died of a seizure.

I told the journal editors in the room that we scientists have been discovering that peer-reviewed numbers of AI health models have proved weaker than the initial press releases. For example, Microsoft said its diagnostic AI was four times as accurate as doctors. In the peer-reviewed version in Science, it was only twice as good. Marketing claims run ahead of the evidence, so please measure what you said you would measure!

The Lego session was my favourite. Nineteen of us were given two minutes and the same six Lego bricks to make a duck. The result? We built nineteen different ducks (one person had never used Legos before). If nineteen of us read "make a duck" nineteen ways, imagine how differently we read "integrate AI into your clinical workflow."

Without a clear vision and shared implementation plan, AI may not be what it quacks up to be.

Wednesday 2 December, back to Boston!

Dr Eliane Boucher, who leads behavioural health at ProofStack, and I are co-hosting the inaugural meeting of the Boston chapter of the Hemingway Report community. Chapters already run in London, San Francisco and Melbourne, and the community has a Slack group. If you work in digital mental health, please join and come along!