Generative AI and Productivity Speakers: myth vs reality (2026 white paper)

Chapter 1 — What separates a good speaker from an excellent one on Generative AI and Productivity?

The difference fits into one signal: either the speaker arrives with figures from a project he ran himself, or he arrives quoting somebody else's report.

A good one explains what a language model does and why adoption stalls. An excellent one tells the committee how many people opened the tool every week against the licences the client paid for, how many weeks the team needed to get from curiosity to daily use, and which process broke in the attempt. Per the Events Industry Council (2024), the global meetings industry generated 1.5 trillion dollars in direct spending and returned to its growth path; inside that figure, the keynote is the only line item an organiser can specify for quality by contract. At RadarSpeakers we rank speakers by the ORIGIN of the data, never by accumulated reputation.

Chapter 2 — The figure that exposes a catalog speaker

Ask for the percentage of customised content: the answer arrives in thirty seconds and is worth half an hour of interview. Someone who repeated one talk eleven times in a quarter rewrites 10% to 15% of the deck. Someone who runs discovery with area directors rewrites 40% to 60%, and those hours show up itemised in the fee instead of wrapped in prestige. At the 1,200-attendee congress the assignment went to the half-price finalist, the one offering two weeks of interviews, and nearly eight in ten people in the room, 78% of them, wrote down what they would do back at the office. The academic explained the technology better than anyone on the shortlist. He moved nothing, because the room had not come to understand the model: it had come to decide on Tuesday.

Chapter 3 — Budget by annual revenue band: the same speaker does not cost the same

A fee is never judged in absolute terms, only against the annual revenue of whoever signs, and each band behaves differently. Below 500 thousand dollars the keynote competes with payroll: if base hourly wages in service roles rose 4% to 14.20 USD an hour according to 7shifts (2024), the committee is choosing between one talk and several months of a role. Between 500 thousand and 1 million the first formal content budget appears, fees of 3,000 to 8,000 dollars, no room for complex riders. Past the million there is a programming committee, and asking for prior discovery stops being cheeky. Over 5 million the range opens to 15,000-40,000 dollars with a closed workshop for directors. And over 10 million nobody argues the fee: they argue territorial exclusivity and recording rights, which cost more than the talk.

Chapter 4 — The high end: when a name-brand speaker is what you are buying

Above 5 million in annual revenue, the corporate event is not buying content: it is buying the name on the banner. An international speaker in that band charges stage time plus a logistics block committees underestimate almost every time: premium-class flights, two hotel nights, an intermediary agent on 20% to 25% commission, and a blackout window that locks him out of that city for six months. That window is worth more than any discount you win. We got this wrong for years: we confused a high fee with customisation, and paid first-tier money for a catalogue talk more than once. Each additional rating star moves 5% to 9% of revenue according to Harvard Business School (Michael Luca); public reputation behaves the same way, it lifts the price and guarantees nothing about the decision.

Chapter 5 — What the organizer must demand before signing

Demand four things and most of the risk collapses on its own. A signed pre-event brief, where the speaker states which business questions he answers and with which figures. Deployment evidence: the project numbers with permission to mention them, never the client logo. Two verifiable references from organisers who booked him inside the past year, organisers rather than attendees, and one question worth putting to them: did he repeat a talk he had already given? And the fee on a single line, taxes, travel and recording rights itemised, because 40% of booking disputes are born in the rider and not in the fee. None of this creates friction when the booking is direct, with no intermediary commission: the person negotiating is the person walking on stage.

Chapter 6 — The wrong format ruins a good speaker

A 45-minute opening, a 25-minute panel and a 3-hour workshop are three different assignments, and the committee that hands one script to all three has already lost. Whoever opens the congress has to leave a debatable thesis behind, something that organises the two days after it. On a panel, the job is to contradict the other two speakers with data, not to present. In a workshop, 70% of the clock belongs to the attendees, who work on their own processes while the speaker corrects over their shoulder. Every RadarSpeakers profile states which formats an expert commands and which he turns down, because taking all three without touching the structure means selling the same hour three times. Ask how he rewrites when fifteen minutes get cut that morning.

Chapter 7 — What happens if the committee books fame instead of fit?

What happens is that the congress buys entertainment and books zero initiatives: satisfaction climbs, return never shows up, and the content budget gets cut the following year.

Picture 35,000 dollars paid to the best-known name on the circuit, no prior discovery. The talk lands cleanly, the room applauds, the exit survey returns 8.4 out of 10. Three months on, the committee hunts for return and finds not one initiative born in that hour, because satisfaction measures entertainment while the number that counts is declared action at 30 days. A year later the sponsor asks for justification, gets none, cuts 30%, and the congress drops a category. That whole sequence started in twenty minutes of committee time. The paradox: whoever satisfies a room most rarely unsettles it, and no process changes without discomfort.

Chapter 8 — KPIs at 3, 6 and 12 months to measure the keynote

Measure three horizons or measure nothing. At 3 months the number is what share of the room promised something at 30 days and then actually did it: against that 78% who promised, real completion of 25% to 35% already survives a sponsor's questions. At 6 months, count generative AI initiatives that cite the talk as their origin and work out cost per licence seat used, not contracted, because the gap between the two runs past 40% in corporate rollouts. And at 12 months the KPI belongs to the event: how many attendees came back, how many sponsors renewed. Documented savings from AI in operations, labour costs down 8% to 12% with forecast accuracy above 90% according to TimeForge (2025), set the order of magnitude your committee should demand before signing.

Chapter 9 — Five differences the committee can verify before signing

TECHNICAL DEPTH WITH TRANSLATION. What a model does, any competent generative AI speaker can explain; what changes on Monday in the workflow of the department listening, with the hours-released figure per process, only the excellent one explains. One informs. The other moves a budget line. WHERE THE NUMBERS COME FROM. Quoting McKinsey is free. We ask for something else: the figure the speaker pulled out of his own project, how many of them logged in weekly against the seats that were bought, how many weeks passed before daily use, and what broke on the way. TALK ARCHITECTURE BY FORMAT. An opening slot and a closed workshop need different scripts, and a panel takes neither.

Chapter 10 — Five differences the committee can verify before signing — in practice

The excellent speaker rebuilds the structure when the clock changes; the average one drops slides and trusts that nobody spots the seam. HANDLING CONFLICT IN THE ROOM. Job anxiety comes attached to this topic. When someone in row twelve asks whether AI will take their job, the average speaker answers with generic optimism; the excellent one separates task replacement from role replacement, names the source, and holds the discomfort without decorating it. SHELF LIFE OF THE MATERIAL. In a field where model capabilities turn over every quarter, one-year-old material is a liability on the books. Demand a last-revision date and a pre-event update clause in writing: the cheapest clause to negotiate, the one that protects the congress reputation most.