How to Analyze Shopper Behaviour in Physical Stores

The behaviours worth measuring

• Arrival: when they come, alone or in groups, and to which zone first.
• Attention: which displays hold them and for how long.
• Engagement: whether an advisor reached them, and how long that took.
• Intent: what they said they came for, in their words, and how firm it was.
• Comparison: which rival product or store they named, and what price they had been quoted.
• Objection: what stopped them, and whether it was answered.
• Decision: bought, left, or left a number and a reason to call.
• Return: whether the follow-up happened and whether they came back.

Five sources and what each misses

A step-by-step method

1. Write down the three questions you want answered. For most Indian assisted-retail networks they are: why do walk-ins leave, what do they compare us to, and what did they want that we did not have.
2. Choose the source that can observe each. All three live in the conversation, so start there. Put consent signage at the entrance and counter, and capture on the advisor's phone with a disclosure line.
3. Run for two weeks in five to ten stores across two regions. Track coverage first: conversations captured as a share of walk-ins. Below half, fix advisor trust before reading anything else.
4. Read the objections and unmet requests by store, ranked by frequency, with the transcript line behind each. This is the first real picture of behaviour most networks have ever had.
5. Add movement where a question needs it. If the objection list is short but conversion is low, the problem may be before the conversation: wait time, unattended visits. That is where camera events, on the Borentis roadmap as ShopperDNA, would answer.
6. Cluster. Once there are a few hundred conversations, ShopperPersonas groups them into the customer types that actually walk in, each with its questions, objections and closers.
7. Act on one thing per store per fortnight, then re-read. Behaviour analysis that does not change a display, a script or a stock order is a report.

Reading the result

Read by store, not by network. A network average of objections hides the fact that one store hears price and another hears delivery time. Read by week, so a rising rival mention is a trend and not an anecdote. Read with coverage next to every number, so a quiet store's score is known to rest on twelve conversations, not two hundred.

Read non-buyers separately from buyers. Buyer behaviour is the story the till already tells. Non-buyer behaviour is where the recoverable revenue sits, and the guide on why walk-ins do not convert goes deeper on that.

Mistakes that waste the effort

• Starting with heatmaps. They are the easiest thing to buy and the hardest to act on, because a hot display tells you where people stood and not what they thought.
• Trusting the CRM's lost reason. It is chosen from a dropdown by the advisor after the customer has gone.
• Analysing a national sample once a quarter. Behaviour changes with the scheme, the launch and the rival's offer; the read has to be weekly.
• Identifying people. Facial recognition, age and gender estimates and emotion detection add legal risk and no decision a store manager can act on. Count events; hear consented conversations; identify nobody who did not ask to be called.
• Building dashboards before deciding who acts. Every behaviour metric should have an owner: the trainer, the buyer, the store manager, or the area manager.