Engineering Sound for the Retail Floor

Hakan Emren · Founder, notFM · August 2026 · Version 1.0 · 10 pages

A modern retail floor is an instrumented space. Lighting is specified in lux. HVAC is specified in air changes per hour. Planograms are tested and price is optimised continuously. Sound — the one variable that reaches every customer and every member of staff, continuously, for the entire trading day — is specified as “play something nice.”

This paper argues that background music in commercial space should be treated as an engineered system with a specification, a control loop and a verification method. It reviews what the published evidence does and does not establish, examines why a 1982 prescription does not transfer to a 2026 store, and describes a system built to close the gap — including the measurements that show it does what it claims.

What is inside

• 1The last unmanaged variable
• 2What the evidence does and does not establish
• 3The other audience: staff
• 4System design
• 5The control plane
• 6Verification
• 7What we do not claim
• 8Roadmap

A dissent, on the record

The best-known result in this field — Milliman’s 1982 supermarket experiment, usually summarised as slow music increases sales — is a single-store finding from a world without smartphones, price comparison or the ability to skip a track. Section 2 keeps it in the bibliography and states plainly why notFM does not repeat it as guidance, together with a practitioner’s dissent from thirty years in radio.

There is no transferable formula for what a given store should play. What a retailer needs is not a prescription but an instrument — one that can set tempo deliberately, deliver it reliably to every room, and prove afterwards what each room actually received.

What the paper measures

Section 6 states the selection engine’s behaviour as falsifiable predictions and tests them against 8,784 real broadcast events. The headline result: track selection is between 34 and 333 times more regular than random assignment. In a 261-track pool, every single track had been played either 8 or 9 times.

Section 7 is equally explicit about the limits. We have not run a controlled sales trial in our own client estate, and we do not claim one. What the paper demonstrates is delivery — that the system broadcasts exactly what it was instructed to broadcast, and can show the log. That is a precondition for any outcome claim, and it is the part the industry currently cannot demonstrate.

Questions about the specification or the verification method are welcome at hakan​[at]​notfm.com.