AI at trade shows, in service of the visitor
By AlexandreCTPO and co-founder, AI, Product and Tech
A trade show visitor asks their question the way they do on their phone, instead of searching through a printed programme. Here is what changes when a conversational agent answers from the event's own content, draws the route to the right place, and knows when to hand off to a human.
The visitor asks a question, not a stand number
At a trade show, a fair, a conference or a festival, visitors arrive with the same habit they use on their phone the rest of the day. They do not unfold a paper map or dig through a tiered menu; they type or say what they are looking for, in plain language. Where is this stand, what time does that session start, where are the restrooms, where can I eat. The habit comes from everyday apps, but it shows up on event grounds without asking permission.
That is exactly the starting point of EventKab, our agent for trade shows: a chat placed on the event website or reached through a QR code, which takes the question as it is asked and answers it from the content the organiser prepared.
This shift is not limited to younger visitors or to the most technology-minded events. A family visiting a public fair, a busy decision maker at a professional trade show, an attendee juggling two conference sessions, or a festival goer looking for a water point outdoors: all of them ask a question before looking for a document. The printed programme, the wall display or the map folded in four have not disappeared, but they are no longer the first reflex. Designing the visitor experience around the question rather than around a document means reordering priorities: the content comes first, the interface that shows it comes second.
The conversational agent answers from your content, and names its source
The core of the system is a conversational agent. It does not answer from general knowledge picked up on the web: it answers from the venue map, the session programme, the exhibitor list and the practical information of the event at hand. Every answer states which document it came from, so both the visitor and the organiser can check where the information is drawn from.
This constraint changes what the tool is. A sourced agent cannot make up a time slot or a location it does not know: if it finds nothing in its base, it says so. That is an engineering choice, not an interface detail, and it is the one we hold to in our AI agent methods for any assistant exposed to the public.
The agent also answers in the language of the question, with no configuration needed on the organiser's side. A visitor writing in Spanish or English gets an answer in the same language, drawn from the same content base. For an international audience as much as a locally multilingual one, this avoids preparing a separate version per language.
The full detail, with real exchange examples shown on the site, is on the page dedicated to the chat agent.
This content base is nothing like a fixed document written once for opening day. It gathers the venue map, the session programme, each exhibitor's profile and the event's practical FAQ, and it can be corrected right up to opening and throughout the event. A room change announced the same morning, a moved activity, catering that closes earlier: the update flows into the agent's answers with nothing to redeploy. That is what sets an agent answering from a living corpus apart from a simple question-and-answer bot frozen at launch.
When the answer is a place, the 3D map takes over
A large share of the questions asked at an event call for an answer that is a place: a stand, a room, a stage, a water point. For these questions, text alone still forces visitors to find their bearings on a paper map folded in four, compare one numbering system with another, then ask their way at the reception desk.
The interactive 3D map removes that step. Halls, stages and spaces are shown in volume, navigable with a finger from the visitor's phone. When the agent identifies that the answer is a location, it displays this map, highlights the point searched for, and draws the route from the visitor's position, through the aisles rather than in a straight line. The route recalculates if the visitor takes a wrong turn.
This map is built from whatever the organiser already has: a PDF, an image, a layout file supplied by the stand contractor, or failing that the venue's own floor plan, which any convention centre or hosting site already holds. Existing stand numbering is carried over as is, instead of creating a second reference nobody would maintain. If a location changes the day before opening, it moves once in the tool, and the map as well as the agent's answers follow, with no reprinting and no redeployment.
All of this stays inside the same conversation: the visitor never leaves the chat to check a separate map. The answer, the point's detail card and the map zoomed on the route appear in the same place.
This principle works just as well indoors as outdoors. A conference with rooms spread across several floors, a festival with several stages, scattered entrances and water points: the logic stays the same, only the map's content changes. Exhibitors themselves can reuse the same tag on their own page, pointed at their spot, with the route drawn from the site's entrance, which extends the tool beyond the organiser's own website.
A personalised visit, before arrival
A visitor does not need to be on site to start preparing their visit. By stating their role or the topic they are interested in, they receive suggestions for relevant exhibitors or sessions, with the reason behind each suggestion. They can then keep chatting to refine this visit or ask another question.
This personalisation answers a real problem at large events: a visitor who does not know the full exhibitor list or the programme in detail does not know what they are missing. By starting from a need or an interest rather than a name, the agent widens what the visitor discovers, including the least visible locations.
At a public fair, the same logic reads differently: the visitor does not arrive with an exhibitor's name in mind, but with a need put in their own words, such as insulating a roof or finding an activity for children. The agent translates that need into a list of relevant stands, then draws the route to the first one. Searching by need rather than by name changes which aisles fill up and when, to the benefit of the ones that would otherwise stay outside the main flow.
Operations: what your visitors are searching for, live
Every question asked to the agent is also information for the organiser. Operations data aggregates what visitors are searching for in real time: the most requested exhibitors, recurring questions, and above all the questions left unanswered, which signal a gap in the content provided or a need nobody anticipated.
This data serves two purposes. During the event, it lets you fix missing content or information that has become outdated, without waiting for the next day. After the event, it becomes concrete proof to show exhibitors and partners for renewal: who was searched for, and how often, rather than a general impression of attendance.
For an organiser, this turns a vague sense of footfall into a list of names and topics that can be discussed with an exhibitor one on one. A stand that was searched for repeatedly, a session that visitors kept asking about, a practical question that came up far more than expected: each of these becomes a starting point for a conversation about next year's edition, grounded in what actually happened rather than in a general impression.
Handing off to a human
An agent answering from a defined scope will inevitably meet questions outside that scope. The rule set in EventKab's design is simple: outside its content base, the agent does not fill the gap with a plausible answer. It tells the visitor and hands the conversation over to the organiser's reception team.
This choice follows directly from what we defend in our approach to responsible AI: an agent exposed to the public must know the limits of what it knows, and hand off rather than improvise an answer that sounds right without being right. At an event, a wrong time or location has an immediate cost for the visitor who follows it.
This handoff also helps the reception team itself: it no longer absorbs the same question at every time slot, and can focus on requests that genuinely need a human.
One tag to place, no application to download
Technical integration comes down to one line of code added to the event page, with no application to install and no account to create on the visitor's side. A visitor scanning a QR code on site or clicking a link on the website reaches the chat directly. This lightness of access matters as much as the quality of the answers: a tool that requires a download before the first answer loses a large share of the visitors who would have asked their question.
For an organiser who does not yet have a dedicated website for their event, the showcase site is built alongside the agent, so the tag has a page to sit on from the start.
Hosted in France, and GDPR
Exchanges with the agent go through a French language model and data hosted in France. The scope of collection, the legal basis for processing and the retention period are set in the contract before the event opens, in a logic of GDPR compliance rather than options switched on after the fact.
The content provided by the organiser, programme, exhibitor list, map, is neither used to train a third-party model nor reused on another event. This clarity about hosting and data use is, for many public organisers or local authorities, a condition of entry even before looking at the features.
What the organiser needs to prepare
The quality of the answers depends directly on what is supplied beforehand. Three conditions come up every time:
- Up to date content. The map, the programme, the exhibitor list and the practical information must be accurate at opening, and updated if they change during the event: a room moved, a time changed.
- A clear scope. Define what the agent should cover, and accept that it hands off to the reception team beyond that scope, rather than trying to make it carry everything from the first edition.
- A stated format and scale. A public fair, a professional trade show with exhibitors, a conference with parallel sessions and an outdoor festival with several stages do not share the same content or the same search priorities; the initial scoping takes this format into account.
The project carried out for EventKab illustrates this logic end to end: a knowledge base editable by the organiser, an agent that only answers from it, and settings tested before opening. The full detail of this project is described in our case study chat agent for trade shows.
For an organiser wondering whether their event needs this kind of tool, or where to start scoping it, our AI diagnostic lets you ask the question before committing to anything.
Sources
Frequently asked questions
Can the conversational agent make up an answer when it does not know?
No, that is the design rule set from the start: the agent only answers from the content supplied by the organiser and names its source. If it finds nothing in that base, it tells the visitor and hands the question over to the reception team instead of offering a plausible answer.
Does the visitor need an app to use the chat and the 3D map?
No. Integration is done through a tag placed on the event website, and the visitor reaches the chat by scanning a QR code or clicking a link, with nothing to install and no account to create.
Does the 3D map work if the organiser only has a paper map?
Yes. A PDF, an image or even a scanned paper map is enough as a starting point; existing stand numbering is carried over as is, and the venue's own floor plan serves as a base when no detailed layout exists yet.
Where is visitor and organiser data hosted?
In France, with a French language model. The scope of collection, the legal basis and the retention period are set in the contract before the event opens, in compliance with GDPR.