Reading Exhibit Analytics Carefully

What exhibit analytics are

Many digital exhibits can record what happens on them: when a session starts, which screens are opened, which buttons are pressed, how long a session lasts, which language was chosen, and when the system was idle. These records, collected over weeks or months, are exhibit analytics. They are a form of evidence that does not require anyone to stand in the gallery with a clipboard.

They are attractive because they are automatic and continuous, and the numbers look precise. That precision is where the risk lies. A figure with many digits can feel more trustworthy than it deserves.

What logs can tell you

Used carefully, logs are good for questions about what happened on the system:

Whether the exhibit is used at all. A station that logs very few sessions over a busy month deserves a closer look.
Patterns over time. Use may rise on weekends, drop in the afternoon or spike during school visits.
Which content is reached. If a deep section is almost never opened, perhaps it is hidden, or perhaps the route to it is unclear.
Where sessions end. Many sessions ending at the same screen may suggest a problem, or a natural stopping point.
Language and option choices. The share of sessions using a language switch or an accessibility setting can inform planning for multilingual visitors and for access features.
Technical health. Errors, crashes and restarts can be recorded, which supports remote monitoring and a maintenance plan.

What logs cannot tell you

Logs record actions on a device, not people or intentions. They cannot say:

Who used the exhibit. Unless you ask separately, you do not know ages, groups or backgrounds.
Why they did something. A long session might show fascination or confusion.
What they understood or felt. Taps are not learning. The wider question of learning outcomes needs other evidence.
What happened around the screen. Watching companions, reading a nearby label or talking about the exhibit leave no trace.
Why people did not use it. Logs only count people who touched it. Visitors who walked past are invisible, unless some other counter records them.

Define every number

Before using a figure, find out exactly what it counts. A “session” might begin on the first touch and end after a period of inactivity, such as one or two minutes, depending on how the system is set up. A visitor who pauses to talk may be counted as two sessions. Two people who share a turn, one after the other, may be counted as one. A restart may be logged as activity.

Questions to ask of any metric:

What event triggers it, and when does it stop?
What is the idle timeout, and what happens after it?
Can the same person be counted more than once?
Are accidental or test touches filtered out? Staff cleaning the screen, a child pressing everything, or a technician running a check can all create sessions.
Was the logging working the whole time? Gaps from power cuts or software updates distort totals.

If a definition is unclear, say so in your report. If the definition changes during the period, comparing before and after may be meaningless.

The trouble with small numbers

Small counts bounce around by chance. If a station logged 12 sessions one week and 9 the next, that is a drop of 3 sessions, and ordinary variation could explain it. Treating it as a trend risks inventing a story. The same applies to comparing two buttons that were pressed 7 times and 4 times.

Some guidance for small samples:

Report the raw counts, not only percentages. “3 of 8 sessions” is honest; “37.5 percent” suggests precision that is not there. (For the record, 3 divided by 8 is 0.375.)
Be careful comparing groups of very different sizes.
Look for patterns that hold over several weeks and several conditions before calling them real.
Avoid ranking items that differ by only a few counts.
Use plain language: “slightly more”, “about the same”, “too few to say”.

If there is no good way to collect enough data, say that the numbers are only indicative and look to other methods.

Context matters

A number in isolation says little. Compare it with what you would expect given the setting:

Visitor numbers. If the gallery had far fewer visitors that week, lower use may reflect that, not the exhibit. Visitor attendance counts help.
School and holiday schedules. Groups produce bursts of use that skew averages.
Location and layout. A station by the entrance and one in a corner will see different traffic. Lighting, screen glare and noise from other exhibits all affect use.
Changes in between. Did staff move a sign, or did a software update alter the menu? Note changes in a log.
Novelty. A new installation often gets heavy use at first, and then settles.

Pair numbers with watching and asking

Logs say what and how often; observation and conversation say how and why. A common pattern is to notice something odd in the data, such as many sessions ending after 10 seconds, then go to the gallery and watch. You may find that visitors are tapping the attract screen and leaving because they expect something else, or that a heavy door slam keeps triggering the idle mode. The data pointed to a question; observation answered it. Visitor observation and dwell time measurements can fill in such gaps.

Surveys and short interviews can show whether heavy use reflects satisfaction.

Privacy and responsible logging

Collect only what you need. Logs that identify individuals, such as those recording faces, device identifiers or free-text entries, carry privacy obligations. Decide before collecting what you will keep, for how long, who can see it, and whether visitors need to be told. Anonymous counts of taps and sessions are generally simpler. If in doubt, ask whoever handles data protection for the institution. Visitor privacy deserves its own planning, not an afterthought.

Presenting results

When sharing analytics with colleagues or funders:

State the period covered and how the data was collected.
Describe what each metric counts.
Show counts alongside any percentages.
Say what the data cannot answer.
Avoid claims like “visitors loved it” or “the exhibit failed” based only on logs.
Recommend a next step, such as a short observation session.

A modest, carefully worded conclusion is more useful, and more likely to hold up, than an impressive-sounding one that cannot be defended.