Four hundred attendees. Sixty exhibitors. Two days. And at the morning briefing before the event opens, you check the meeting dashboard and find that more than half of all meeting requests are still unanswered. The traditional fix (chasing people by email) does not scale. The modern answer is AI matchmaking.
In 2026, AI matchmaking is the most common AI feature in event-tech platforms, with 35% of meeting professionals naming it as their top planned AI use case (Amex GBT, 2026). Clarion Events achieved a 44% year-on-year increase in meetings using it. The results are real, but so are two serious blind spots that almost nobody in the market discusses openly.
The first: AI matchmaking is actively counterproductive in event formats where organiser control over meeting assignments is a commercial and contractual obligation. The second: the most common form of “AI matchmaking” in 2026 is an event coordinator uploading an attendee spreadsheet to ChatGPT, a practice with significant GDPR exposure.
This guide explains how AI matchmaking actually works under the hood, what the data shows it genuinely improves, and when keeping control yourself is the better decision.
What Is AI Matchmaking at Events?
AI matchmaking at events is an algorithm-driven recommendation system that analyses attendee profile data and real-time behavioural signals to surface high-probability, value-aligned 1:1 connections between buyers, sellers, founders, investors or peers. In 2026, it is the most common AI feature across event-tech platforms, with 35% of meeting professionals planning to deploy it (Amex GBT Global Meetings & Events Forecast, 2026).
Two important distinctions before going further:
- AI matchmaking is a feature, not an event format. It runs inside your event management platform: it does not replace the event, the programme, or the organiser’s judgement. Treating it as a standalone solution is the most common misconfiguration.
- AI matchmaking amplifies what is already in the data. Thin attendee profiles generate thin matches. The algorithm cannot fabricate relevance from empty fields. Profile data quality is the single most important determinant of match quality.
Three Generations of AI Matchmaking Technology
Understanding which generation of AI a platform actually uses, rather than accepting “AI-powered” at face value, is one of the most useful questions you can ask in a vendor evaluation. There are three distinct generations currently in production.
Technology
Three Generations of AI Matchmaking
Generation 1: Rule-Based (Tag Intersection)
The oldest and most widely deployed form. The algorithm matches attendees by finding shared tags between two profiles: if Buyer A selects “Luxury Hotels” and Seller B also selects “Luxury Hotels”, they are paired. Fast, transparent, and entirely predictable, but brittle. A “Head of Procurement” and a “Purchasing Director” may never share a tag, and the algorithm misses them entirely. Rule-based matching is still the default in many mid-market event platforms and works well for small, well-tagged audiences.
Generation 2: Embedding-Based (Semantic Similarity)
More sophisticated platforms encode profile text, job titles, interest descriptions, session summaries, into mathematical vectors using a model such as Sentence-BERT, then calculate cosine similarity scores across the attendee pool. This captures semantic meaning beyond keyword overlap. “Procurement Director” and “Head of Purchasing” score as closely related because the model understands language, not just matching strings.
Significantly better match quality, but with two known limitations: the cold-start problem (profiles submitted less than 48 hours before the event have sparse data, which degrades quality for late registrants) and higher computational overhead at scale.
Generation 3: Agentic Matchmaking
The emerging frontier. LLM-based agent loops that update match rankings continuously based on in-event behaviour: which meeting requests were accepted, which were declined, how long an attendee browsed at a particular exhibitor profile, which session they attended. The system learns as the event unfolds, meaning matches on Day 2 are better than matches on Day 1. Bridged.media (2026) describes this as “agentic matchmaking”: adaptive, multi-signal, and increasingly available in Tier-1 platforms. Still newer and harder to audit, but directionally where the industry is headed.
When evaluating vendors, ask specifically: “Which generation does your matching engine use, and can you show me how it handles two attendees who describe the same interest in different words?”
What AI Matchmaking Actually Improves, and What the Data Shows
The outcome data for AI matchmaking, in the right format with the right configuration, is strong.
- Clarion Events (Grip, 2024–2025): 44% year-on-year increase in in-person meetings across their Gift & Souvenir trade show portfolio using AI matchmaking. At the Smoky Mountain Gift Show, 195 meetings were achieved against a target of 135, with 100% buyer engagement.
- Swapcard (2026): AI recommendations double match acceptance rates at Trade Show Organisation+ events. Tier-1 connection acceptance rates reach up to 100% in some Association+ event cohorts.
- The unanswered request problem: 65–75% of attendee-to-exhibitor meeting requests at trade shows go unanswered without AI-assisted outreach; the gap drops to 25–35% with AI matchmaking active.
- Market scale: AI-powered event matchmaking is a $9.06 billion segment growing at 6.7% CAGR (Cognizant, 2024). This will not be an optional feature in three years.
These results are achievable, but they require rich profile data, correct configuration of matching constraints, and (for hosted buyer formats) integration of commercial obligations. The numbers above come from well-configured deployments. Poorly configured AI matchmaking produces poor results, just faster.
KPI benchmarks to plan against for your next event:
Benchmarks
AI Matchmaking: KPI Benchmarks
| Metric | Realistic target | Strong result |
|---|---|---|
| Match acceptance rate | 40–60% | ≥ 60% |
| Meetings per attendee | ≥ 2 | ≥ 4 |
| Kept-meeting rate | ≥ 80% | ≥ 90% |
| Useful-meeting rate (post-event survey) | ≥ 65% | ≥ 75% |
| Time to first accepted meeting | under 24 hours | under 12 hours |
When AI Matchmaking Is the Wrong Tool
Here is what vendor blogs will not tell you: for certain event formats, AI-driven meeting allocation is not a feature: it is a liability. The core assumption in AI matchmaking is that the algorithm knows better than the organiser who should meet whom. In many large, open-registration events, that assumption holds. In the following formats, it does not.
Hosted Buyer Programmes with Sponsor Guarantees
Hosted buyer events run on explicit commercial promises. Sellers, hotels, venues, tour operators, destination organisations, pay to participate and receive a guaranteed number of qualified buyer meetings in return. Buyers sign contractual commitments to attend. The Meetings Show’s publicly available Terms & Conditions (Clauses 4.7–4.8) make non-attendance grounds for exclusion from future editions. IMEX America expects 6–8 pre-scheduled meetings per hosted buyer per day, actively monitors attendance, and withdraws hosted buyer status for repeated no-shows.
InEvent states it plainly in their hosted buyer documentation: “You are making explicit promises: to exhibitors, ‘You will meet qualified buyers.’ To buyers, ‘You will meet relevant suppliers.’”
When an AI algorithm optimises around these obligations, because it has access only to profile similarity, not to the commercial agreement, it can undermine the guarantees that make the format financially viable. Exhibitors who paid for access did not get it. Buyers found the meetings irrelevant. This destroys the trust that is an organiser’s core product. For more on running a hosted buyer programme correctly, see our step-by-step guide to hosted buyer programmes at tourism trade shows.
Association Formats with Member Hierarchies
Industry associations contain long-standing relationships, political sensitivities, and strategic logic about who should meet whom at the annual conference. None of that is in the profile data. An algorithm that matches purely on interest similarity will miss those dynamics entirely, producing meetings that are technically “optimal” but contextually wrong for the community.
Small Events Under 150 Attendees
At this scale, an experienced organiser typically knows every participant personally: their context, their needs, the conversation that should happen between them and a specific attendee. “AI decides” here is not an efficiency gain; it is a downgrade. Curated manual matching with AI as a secondary validation layer usually outperforms full AI allocation.
The Decision Framework
| Situation | Recommended approach |
|---|---|
| Sponsored seller access with guaranteed meeting quotas | Control-First: manual or hybrid (AI suggests, organiser confirms) |
| Association format with member hierarchies | Control-First, AI as support layer only |
| Large open-registration event (500+, homogeneous goals) | AI-optimised, organiser sets boundary conditions |
| Hosted buyer programme with clear buyer/seller profiling | Hybrid: AI within pre-defined category constraints |
| Startup Demo Day or investor conference, no sponsor guarantees | Full AI matchmaking, optimal when profile data is rich |
The best implementations reflect this logic. DELIVER events combine AI suggestions for volume with dedicated account managers for VIP pass holders. Swapcard markets “full control over matching rules, participant groupings, and meeting limits” as a top differentiator, because organisers asked for it. For a detailed look at how AI matchmaking applies specifically to founder-investor formats, see our playbook on matching founders with the right VCs.
The Shadow AI Problem: When “AI Matchmaking” Means Uploading Your Attendee List to ChatGPT
There is a form of AI matchmaking that no platform vendor will mention in their case studies: an event coordinator exports the attendee list to Excel and pastes it into ChatGPT with the question “Who should meet whom?”
It is more common than most organisations admit. And in almost every case, it is unlawful under GDPR.
Why It Is a Legal Risk
Your attendee list, names, email addresses, job titles, company names, stated interests, is personal data under Article 4 GDPR. Processing it requires a lawful basis and, when a third party handles it, a Data Processing Agreement (DPA) under Article 28 GDPR.
Free and consumer-tier ChatGPT has no DPA. activeMind.legal, a German data protection firm with active GDPR practice, states explicitly that in the consumer tier, OpenAI acts as an independent controller, not a data processor, meaning the transfer of your attendee list to that tier is an unlawful international data transfer without additional safeguards. The same logic applies to consumer-tier Claude, Gemini, and any other general-purpose LLM accessed through a free web interface. ChatGPT Enterprise and the OpenAI API include a standardised DPA, but only when that agreement has been explicitly concluded and signed.
It Is Happening at Scale
Metomic’s 2026 AI Security Report found that 34.8% of all ChatGPT inputs in enterprise environments contain sensitive data, up from 11% in 2023. Forty-seven percent of organisations have no AI-specific security policies. The most commonly leaked data type: customer and employee personal data, which maps directly to attendee registration data.
Two real-world examples illustrate how this goes wrong. In Australia, a government contractor uploaded names, contact details, and health information of 3,000 flood-affected individuals to ChatGPT, without authorisation, without a DPA, without informing the data protection authority. In January 2026, a senior US cybersecurity official uploaded classified contracting documents to a public ChatGPT interface. If it happens in organisations with active security functions, it happens in event teams under deadline pressure.
The EU AI Act Adds a Disclosure Requirement from August 2026
Article 50 of the EU AI Act, covering transparency obligations for AI systems that interact with natural persons or prepare decisions affecting them, takes effect on 2 August 2026. Event matchmaking systems that generate recommendations determining which attendees meet which other attendees fall under the “limited risk” category. Attendees must be clearly informed that an AI system is generating their match suggestions.
If you are planning events from autumn 2026 onwards that use AI matchmaking, your registration flows and event app onboarding screens need a disclosure statement now. The EDPB and EAIA published draft implementing guidelines on 8 May 2026.
Five Safeguards to Implement Now
- Use only GDPR-compliant event platforms with an explicit EU-based Data Processing Agreement and EU data residency, not consumer AI tools.
- Before evaluating any AI matchmaking vendor, ask: where does attendee data reside? Which sub-processors handle it? Is model training disabled for your data?
- Establish an internal policy: no attendee personal data in consumer AI tools. Include this in onboarding for anyone with access to registration exports.
- If using enterprise LLM plans (ChatGPT Enterprise, Claude for Work): verify the DPA explicitly, confirm training opt-out, and confirm EU data residency before use.
- Pseudonymise as a last resort: if ad-hoc LLM input is unavoidable in a specific workflow, use IDs rather than names, emails, and job titles before pasting into any AI tool.
How to Choose a Matchmaking Platform: 10 Questions to Ask
Vendor demos focus on the interface. These ten questions reach the architecture and the legal infrastructure behind it.
- Which generation of AI does your matching engine use? Rule-based, embedding-based, or agentic? Can you demonstrate how it handles two attendees who describe the same interest in different words?
- Where is attendee data stored and processed? EU residency? Which sub-processors are listed in your DPA?
- Do you hold a Data Processing Agreement under GDPR Art. 28, and is the platform ISO 27001 certified? Ask for both documents, not a verbal assurance.
- Is attendee data used for model training? Explicit opt-out should be documented in the DPA.
- How does your platform handle the cold-start problem? What do late registrants see in terms of match quality?
- Can I set mandatory match constraints? Guaranteed exhibitor quotas, category locks, VIP-only meeting pools?
- What is your offline fallback if connectivity drops on-site? Critical for trade shows in venues with unstable Wi-Fi.
- How does the algorithm handle gaming by exhibitors? Over-tagging to appear in more searches is a known abuse pattern.
- What API is available for CRM or registration system integration? Avoid manual re-entry of profile data between systems.
- What KPIs do you report, and what are your benchmarks? Match acceptance rate, meetings per attendee, kept-meeting rate: you need numbers to hold the vendor accountable.
Solution. Converve’s matchmaking platform is built for hosted buyer programmes, tourism trade shows, and investor conferences: formats where organiser control over meeting assignments is non-negotiable. The core technology is a configurable meeting matrix: organisers define the rules (buyer/seller categories, mandatory match constraints, VIP quotas) and the platform allocates meetings accordingly. All questions above have documented answers: EU-hosted, GDPR-compliant, with explicit Data Processing Agreement, configurable mandatory match constraints, and a KPI dashboard that tracks acceptance rate and kept-meeting density in real time. Request a demo to see how it handles sponsor guarantees and GDPR-compliant data processing in a single workflow.
How to Implement AI Matchmaking: A 12-Week Timeline
The most common implementation error is starting matchmaking configuration too late. Profile collection and taxonomy design need to happen long before the first match is generated.
Implementation
AI Matchmaking Implementation Timeline
- T-12 weeks Platform & taxonomy Finalise the platform contract and agree the interest taxonomy with exhibitors and buyers
- T-8 weeks Registration opens Matchmaking configuration live from day one: tags, category constraints, mandatory match pools
- T-2 weeks First match review Run the first match generation and manually review the top 20 against your commercial logic
- T-day Live adaptation The agentic engine updates rankings; account managers verify VIP fulfilment
- T+1 week KPI review Collect survey data, map KPIs against benchmarks, document learnings for the next edition
- T-12 weeks, platform selection and taxonomy design: Finalise the platform contract. Work with exhibitors and buyers to agree on the interest taxonomy, the tags or categories they will use to describe themselves in the registration form. This step defines match quality for the entire event.
- T-8 weeks, registration opens: AI matchmaking configuration is live from day one of registration. Tags active, category constraints set, any mandatory match pools defined.
- T-4 weeks, profile quality review: Check the match pool for data completeness. Are profiles rich enough? Are late-registrant profiles complete? Send a targeted prompt to incomplete profiles.
- T-2 weeks, first match generation: Run the first full match generation and manually review the top 20 results. Do they reflect your commercial and curatorial logic? Adjust constraints if not.
- T-1 week, meeting scheduling window opens: Attendees start accepting and requesting meetings. Monitor acceptance rates daily. Intervene manually on any VIP or guaranteed-quota accounts that are underperforming.
- T-day, live adaptation: If using an agentic platform, the engine updates match rankings as in-event behaviour is recorded. Account managers verify VIP fulfilment throughout the day.
- T+1 week, post-event review: Collect post-event survey data. Map actual KPIs against your benchmarks. Document what worked and what did not for the next edition.
For a detailed look at structuring the pre-event matching process for startup conferences specifically, see our guide to the best matchmaking apps for startup conferences in 2026.
Conclusion
AI matchmaking in 2026 is mature enough to deploy confidently, but not mature enough to deploy blindly. The outcome data is real: Clarion’s 44%, Swapcard’s doubled acceptance rates, 11,000 meetings across 63,000 attendees in a single three-day event. These results are achievable in the right format with the right configuration.
The key is knowing when to lean in and when to step back. For large open-registration events with good profile data, AI matchmaking outperforms manual curation at every scale. For hosted buyer programmes with commercial obligations, for association formats with political logic, for small events where you know every face, organiser judgement is irreplaceable, and AI should assist, not decide.
And whatever your format: keep attendee data out of consumer AI tools. The GDPR risk is documented and the EU AI Act disclosure deadline is approaching fast. The tools to do this compliantly exist, they are called event-tech platforms with explicit Data Processing Agreements.
Frequently Asked Questions
How accurate is AI matchmaking at events in 2026?
Match acceptance rates of 40–60% are realistic targets with a well-configured deployment. Swapcard reports Tier-1 connection acceptance up to 100% on Association+ events. Clarion Events achieved 44% more meetings year-on-year using AI matchmaking across their trade show portfolio. Results depend heavily on profile data quality and the completeness of the matchmaking configuration.
Is AI matchmaking GDPR-compliant?
Yes, when profiles are pseudonymised where possible and the platform holds an explicit Data Processing Agreement under GDPR Art. 28, with EU data residency and no use of attendee data for model training. Consumer-tier LLMs (free ChatGPT, Gemini, Claude) do not satisfy these requirements. From 2 August 2026, the EU AI Act requires event organisers to inform attendees clearly that an AI system is generating their match suggestions.
Should hosted buyer programme organisers use AI matchmaking?
Not always. When sponsors or sellers have been guaranteed access to specific buyer profiles, AI-driven optimisation can undermine those commercial commitments. Hosted buyer formats typically require a hybrid model: the organiser defines mandatory match categories, and AI operates only within those constraints.
Is it legal to upload an attendee list into ChatGPT for matchmaking purposes?
Generally no, without additional safeguards. Attendee lists contain personal data under GDPR Art. 4. Uploading them to a consumer-tier LLM without a Data Processing Agreement constitutes an unlawful international data transfer under GDPR Art. 44 ff. Use only certified event-tech platforms with an EU-based DPA, EU data residency, and explicit opt-out from model training.
How long does it take to set up AI matchmaking for an event?
Plan T-12 weeks for platform selection and taxonomy design, T-8 weeks for registration with live matchmaking configuration, T-2 weeks for first match generation review, and live adaptation on event day. The taxonomy design phase is the one most frequently underestimated.
Can AI matchmaking work for small events under 200 attendees?
Yes, with caveats. AI matchmaking can surface relevant connections even in small cohorts. However, for events where the organiser has deep personal knowledge of every participant, manual curation with AI as a secondary validation tool often delivers better match quality than full AI-driven allocation. The format, not just the headcount, determines the right approach.