Book the Boring One: How to Choose the Right AI Speaker

Searching for an AI keynote speaker 2026 audiences will actually act on? Book the one who talks about the dull projects. Your audience has already sat through the exciting version, and it did not survive contact with Monday morning.

The numbers back that up. MIT’s State of AI in Business 2025 research looked at enterprise AI spending of roughly $30 to $40 billion and found that 95% of generative AI pilots delivered no measurable impact on the P&L. The models were not the problem. Poor integration was, along with projects chosen because they looked good rather than because they mattered.

So the risk at your event is not that the speaker underwhelms. It is that the room is impressed, goes home, and does nothing.

What actually won at the AI Awards

I run the AI Awards in Ireland. In November 2025 we handed out 15 awards at the Anantara Marker Hotel in Dublin, the seventh year of the programme, drawn from nearly 200 applications.

Look at what won.

Laya Healthcare took the customer service award for ORCA, which automates outpatient claims for more than 700,000 members using AI vision, OCR and machine learning. It cuts fraud, speeds up reimbursements, gives real-time eligibility feedback and saves millions a year.

Read that again. The winner was claims processing. Paperwork.

St. Luke’s Radiation Oncology Network won in healthcare for AI-powered anatomical contouring, which halved manual planning time and cut variability in 68% of structures, with a human in the loop throughout.

Protex AI won in manufacturing for turning CCTV cameras that companies already owned into a proactive safety system, reducing incidents by 50 to 70%.

Not one of those is a moonshot. Each one sits inside a process that already existed, doing work that people were already doing by hand, measured against a number the organisation was already tracking. The full list of 2025 winners is here.

Now think about the projects you have seen in your own organisation that demoed beautifully and are switched off today. The technology was rarely the problem. Nobody owned the thing after launch. It sat beside the daily work instead of inside it. It was never tied to a number anyone was measured on.

AI does not usually fail in the model. It fails in the handover.

That is the talk your audience needs.

What the wins look like across sectors

None of this is futuristic. All of it is running today.

• Customer support and claims. AI triages, drafts replies, checks documents and surfaces the right policy or past case. People keep the hard ones.
• Internal knowledge search. An assistant over wikis, SOPs, shared drives and chat history. Staff ask a question and get an answer with a source.
• Document-heavy processes. Invoices, contracts, claims and forms read automatically, with the data pushed into the CRM or ERP. Manual keying goes, and so do the typos that cause disputes.
• Data cleaning. Deduplicating, standardising, categorising, enriching. Finance and operations stop tidying and start acting.
• Proposals and first drafts. Marsh McLennan won the large enterprise award for LenAI, an Irish-led assistant used by 70,000 staff across 130 countries that speeds up proposals and lifts win rate.
• Meeting and email summaries. Calls and threads turned into notes and actions people actually read.
• Workflow automation with judgement. Rules-based automation plus a model that handles the exceptions rules cannot cover.
• Safety and monitoring. Existing cameras and sensors made useful, as Protex AI showed.

The Two-Box Test

Every one of those passes the same test. It is the framework I give leadership teams, and it takes about a minute per idea.

1. Box one: is there repetitive knowledge work here? People reading, copying, checking, summarising or routing the same kind of thing over and over.
2. Box two: is there a number already being measured? Handling time, error rate, days to close, incidents, cost per case.

Both boxes ticked, start there. One box ticked, park it. Neither box ticked, it is a science project.

Run Laya’s claims process through it. Both boxes, obviously. Run a general purpose chatbot for the whole company through it. Neither.

That is the whole strategy. A company does not need a grand AI vision. It needs a short list of workflows that tick both boxes. I set out the full structure in my 5 P’s of AI Readiness framework, which is what I use with boards and leadership teams.

Audiences respond to this because it is honest. It respects what they already know, and it gives them something to start on Tuesday.

Six questions to ask before you book

Fee tells you almost nothing about fit. These questions do.

1. How many real AI implementations have you personally reviewed or advised on?
2. What will my audience be able to do on Monday that they cannot do today?
3. Which of my sector’s workflows will you use as examples?
4. What do you say about job losses, and how do you handle that question from the floor?
5. Will you run a pre-event call with me and two or three attendees?
6. Can the keynote extend into a workshop for the leadership team?

Question six matters most. A keynote changes how a room thinks. A workshop changes what it does.

AI keynote speaker 2026: five signals of the right booking

• Operator, not commentator. Ask what they have built, run, judged or advised on. Anyone can summarise the news.
• An evidence base. Named case examples with numbers, not anecdotes. Ask how many real projects they have seen up close.
• Range on depth. The same core message to a board and to an engineering team, at different depths, without losing either.
• Willing to say what does not work. If everything in the talk is upside, credibility drops.
• Real customisation. A pre-event call, a look at your sector, examples your audience recognises.

Sector knowledge matters more than most organisers expect. AI in a credit union is a different conversation to AI in a pharma plant. The data, the regulation and the risk appetite are all different.

Keynote or workshop

Format	Best for
Keynote (45 to 60 mins)	Shifting mindset across a large audience
Workshop (half or full day)	Leadership teams picking and scoping real use cases
Keynote plus workshop	Conferences with an executive track or a board session

More on choosing between them in my keynote versus workshop guide.

What I bring to the stage

My background is operational, not academic. I co-founded an AI and cybersecurity staffing firm, founded AI Ireland, and run the AI Awards. I have delivered over 300 keynotes and trained more than 10,000 professionals, for organisations including Microsoft, Oracle, SAP and PwC.

The talk is built on what those 1,000+ project applications show. Where AI paid back. Where it quietly burned budget. What separated the two.

Topics for 2026 events:

• Practical AI adoption and where the return actually sits
• AI readiness for boards and leadership teams
• AI agents and what they mean for how work is organised
• Building AI literacy and confidence across a workforce
• AI governance and the EU AI Act in plain English

Every session includes hands-on work where the format allows. People believe AI when they use it, not when they hear about it.

FAQs

What topics should an AI keynote cover in 2026? Practical implementation and ROI, AI agents and autonomy, human and AI collaboration, governance and risk, and sector-specific applications. Future-gazing has lost ground. Audiences want something they can apply.

Will you use real examples or generic ones? Real ones. The AI Awards gives me a fresh set of live Irish and international projects every year, with the numbers attached, and I pick the ones closest to your audience’s day job.

How far in advance should I book? Two to four months. Longer for spring and autumn conference season.

Will the talk be too technical for a mixed audience? It should not be. A good speaker pitches to the least technical person in the room while keeping the technical people interested. Ask directly how they handle that before you book.

Can the speaker also run a session for our board? Many can, and it is often the highest value part of the day. Boards ask different questions and need a private setting to ask them.

The bottom line

The project that won our customer service award last year was a claims processing engine. That tells you everything.

The AI work that pays back is quiet, repeatable and tied to numbers the business already tracks. That is also the best AI keynote speaker 2026 brief you can write. Not a prediction show. A clear picture of what is working, what is not, and where your audience should start.

Check available 2026 dates and talk through your event: markkellyai.com/contact

---

Mark Kelly is an AI keynote speaker, strategist and board advisor based in Ireland. He founded AI Ireland, hosts the AI Ireland Podcast, wrote the AI Unleashed series, and created the AI Awards.

---