Exhibition wayfinding analytics that show where attendees actually get lost

The information desk staff know exactly where people get lost. They will tell you, if you ask them on the Thursday, that everybody comes off the north escalator and turns the wrong way. Nobody writes it down, the desk staff change next edition, and the same escalator produces the same wrong turn for six years.

Exhibition wayfinding analytics turns that folklore into something you can act on before the next build, using data the wayfinding tool is already writing and almost nobody exports.

The three signals the tool already records

A digital wayfinding tool, whether it is a kiosk, an app map or a web floorplan, logs more about confusion than any sensor on the floor does.

It records the search string, which is the words the attendee reached for. It records the route request, an origin and a destination and a timestamp. It records whether that route was followed to its destination or dropped partway. And if you have zone reads on the floor, you can join a route request to what the person did next.

None of this requires new hardware. It requires an export, a place to put it, and somebody to look. The exports usually exist and sit unread because nobody has claimed the data, which makes this one of the cheapest things in the whole attendee analytics stack to start doing.

What does an abandoned route request actually mean?

Take a hall with a real number on it. Over three show days the tool logs 2,300 route requests with hall three as the destination. Of those, 943 end without the destination being reached, which is 41 per cent.

Forty-one per cent sounds alarming until you decompose it, and the decomposition is where the value is.

Abandonment splits into three groups by where it happened. Requests dropped within the first 20 metres, before the route's first turn, are the interesting group, because the person had barely started and gave up. Requests dropped after several correct turns usually mean the person got what they needed and stopped following instructions, which is fine. Requests that never move at all are frequently somebody browsing the map rather than travelling.

Suppose the 943 split as 612 before the first turn, 214 after several turns and 117 with no movement. The 612 is your number. That is 26.6 per cent of all requests to hall three failing immediately, and immediate failure has a physical location, because every one of those 612 requests carries the origin it was made from.

Plot the 612 origins and they will not be spread evenly. In my experience they pile up at two or three spots, and those spots are junctions where the instruction the tool gave conflicted with what the person could see. That conflict is fixable with a sign, a floor graphic or a steward, and it costs a great deal less than a floorplan change.

Search terms are a vocabulary problem in disguise

The search log is the part organisers underuse most, because it looks like noise.

Over the same three days, say the tool takes 6,400 searches containing 1,100 distinct strings. The top 15 strings account for 2,430 of them, 38 per cent. Read those 15 and you learn what your audience calls things.

Two failure types show up. The first is a term that returns nothing. Attendees type the product category, the informal name of a feature area, or a brand that exhibits under a different legal entity, and the tool returns an empty result. Every zero-result search is a mapping you could add to the index in ten minutes.

The second is subtler and more useful. Attendees search for a term that returns the wrong thing confidently. If 240 people search for a category name that your taxonomy assigns to one hall while most of the relevant exhibitors sit in another, the tool sends 240 people to the wrong place with complete conviction. That failure never appears as an abandoned route, because the route was followed successfully to a destination nobody wanted.

Lynch described the same problem in cities in 1960. He called the relevant quality of a place its legibility, "the ease with which its parts can be recognised", and built his account around five elements people use to construct a mental map: paths, edges, districts, nodes and landmarks. A hall named by its hall number is a district with no identity, so nobody can recognise it and nobody searches for it. A hall named for what it sells gives people something to hold on to, and the search log tells you which name your audience already uses.

That gives the search log a second job beyond fixing the index. Sort the distinct strings by whether they name a place, a product category, an exhibitor or a person. If the place names in your own signage barely appear in the top 100 searches while product categories dominate, your floor is organised on a scheme your audience does not think in, and every sign in the hall is answering a question nobody asked.

What does backtracking in a zone sequence tell you?

The third signal needs zone reads, so it is only available if you already have portals or scanners at zone boundaries.

Reduce each attendee's day to the ordered list of zones they were seen in. Then look for a specific shape: a zone that appears, is followed by a different zone, and then appears again within a short window. Somebody who reads as zone C, zone F, zone C inside six minutes did not decide to revisit zone C. They went the wrong way and came back.

Take 4,100 attendees with usable sequences. If 620 of them contain at least one of those tight return patterns, that is 15.1 per cent of the floor doubling back at least once. Rank the transitions those returns pass through and you get a list of junctions ordered by how often people got them wrong.

Sequence analysis has more to it than this one pattern, and the general treatment of zone sequences belongs to its own post. For this purpose you only need the return shape, and a six minute window on a show floor is short enough to exclude somebody who genuinely wanted a second look at a stand.

Mapping the signals onto decision points

Passini published his treatment of wayfinding in architecture in 1984, and it gives the framework for turning these three lists into a plan. He described wayfinding as spatial problem solving, made up of decision making and decision executing, and he put the analysis at the decision point: the specific place where a person has to choose and needs enough information to choose correctly.

That reframes the whole exercise. You are looking for decision points where the information available is inadequate for the decision required.

Take your three lists and lay them on the floorplan together. The 612 immediate abandonments give you origin coordinates. The zero-result and misrouting searches give you destination names. The 620 backtracking sequences give you junctions. Where all three land on the same spot, you have found a decision point that is failing, and you can go and stand on it and see why.

The remedy follows from what you find there. A junction with no sightline to its destination needs a sign at eye height at that junction. A junction where the sign exists and is behind a two-storey stand needs the stand moved or the sign moved. A hall whose name means nothing to anybody needs a name. Deciding what to instrument in the first place, and where, is a separate design question about people counting sensors and their placement.

Where this stops

Everything above measures the people who used the tool, and that is a minority of your audience with a specific character.

If 2,300 route requests come from a show with 9,840 arrivals, and if many of those requests came from the same people, your sample might be 15 or 20 per cent of the floor. Those users skew towards people who came alone, people who did not know the venue, and people comfortable with an app. The confident regular who has attended for nine years never opens it, and the group who follow one leader produce one request between five people. So the analysis finds the confusion experienced by first-time solo attendees, which is a real group worth serving and a partial view.

There is a second limit worth being blunt about. Confusion signals are silent about people who did not go somewhere at all. Somebody who could not work out how to reach hall three and gave up on the idea produces no abandoned route, because they never requested one. Those people show up as low traffic in a hall, and diagnosing a persistently quiet aisle needs a different method with different data.

The tool also cannot tell you whether the person eventually arrived by asking a human, which is the most common resolution and the one your information desk absorbs. Counting questions at the desk by topic, even roughly, for one hour on each day, gives you a cheap check on whether the digital picture matches the physical one.

The step for this week is one export. Pull every search string your wayfinding tool recorded at the last edition, sort by frequency, and read the top 50 with the zero-result ones marked. You will find several terms your own index does not know, and fixing those is a text change rather than a build change. If you want a second hour's work after that, join the route request timestamps to your arrival profile and see how many of them cluster in the first 45 minutes, which is when show opening surge management and wayfinding turn out to be the same problem.