From Metrics to Meaning: Insights from a Mixed-Methods Field Experiment on Retail Robot Deployment

Authors: Sichao Song, Yuki Okafuji, Takuya Iwamoto, Jun Baba, Hiroshi Ishiguro · HRI 2026

Licences

Open Access
Paper: CC BY 4.0

Everything below draws on the full paper under its CC BY 4.0 licence; its figure is credited to the authors.

TL;DR

In a live bedding store, an entrance robot raised the share of passersby who stopped but cut clerk approaches, entries and purchases per stopper; staff interviews explain why.

Why it matters

Attention metrics alone would have made the robot look successful, while the conversion data and staff accounts showed it diverted customers from human service. The paper turns this into concrete deployment guidance on robot-to-clerk handoffs and entrance staging.

Problem

Retail robot studies mostly report short-term, customer-side metrics such as stopping and dwell time, with little input from the staff who must turn that attention into service and sales. Real store interactions are triadic (robot, clerk, customer), and it was not known whether staging a robot on an entrance fixture would carry attention through to approach, entry and purchase.

Proposed solution

The authors placed a Sota robot (GPT-4-family dialogue, Google Cloud speech recognition and synthesis) at the entrance of a Japanese bedding store for 12 days, alternating Baseline, Robot-only and Robot+Fixture (4 days each), with clerks free to act on their own judgment. They video-annotated the funnel from passerby to purchase for over 62,000 passersby, tested it with chi-square and Cochran–Armitage trend tests, and then interpreted the patterns through reflexive thematic analysis of interviews with six staff.

Key insights

Higher stopping did not carry through: clerk approach, store entry, assisted experience and purchase per stopper all fell from Baseline to Robot-only to Robot+Fixture.
Clerks held back to avoid interrupting robot–customer conversations, and response latency and speech-recognition errors made the right moment hard to read.
The robot mainly drew children: child-only groups made up 2.72% of stopping groups at Baseline, 28.96% with Robot-only and 40.24% with Robot+Fixture, and children rarely lead to high-priced bedding purchases.
The fixture raised visibility but created a separate micro-space at the threshold where curiosity was satisfied without entering the store.
Staff saw the robot as an attractor or 'PR support' and themselves as closers; the authors recommend observable handoff cues, a short staff-facing note of the conversation topic, and samples placed slightly inside the door.

Main results

Stopping per passerby: 0.73% Baseline, 1.50% Robot-only, 2.48% Robot+Fixture (χ²(2) = 201.55, p < .001; all pairwise differences p < .001), from 20,180, 21,938 and 20,035 passersby.
Clerk approach per stopper: 19.73%, 13.41% and 4.83% (χ²(2) = 34.54, p < .001); store entry per stopper: 31.97%, 20.12% and 9.46% (χ²(2) = 46.65, p < .001).
Assisted in-store experience per stopper: 18.37%, 6.10% and 4.02%; staff-led entry: 8.84%, 1.52% and 0.80% (both p < .001).
Purchase per stopper: 11.56%, 5.49% and 2.62% (χ²(2) = 19.67, p < .001; only Baseline versus Robot+Fixture significant after adjustment); purchase per passerby did not differ: 17, 18 and 13 purchases (p = .744).
Sample touch per stopper was highest for Robot-only (63.72%), against 47.62% at Baseline and 46.68% with the fixture.
Coded interview utterances: attraction 30, fixture design and effect 29, role positioning 21, timing and priority 18, conversion barriers 8.

Limitations

The robot was deployed only at the entrance, so staff could not compare other placements such as inside the store.
Twelve days may not show long-term effects: novelty-driven stopping could fall and clerk handoffs could improve with longer exposure.
Results may depend on the childlike appearance of Sota; other robot types such as androids or virtual agents were not tested.
One store, one product category (high-priced bedding) and six interviewed staff.
The Figure 1 caption calls purchase per stopper 'roughly flat', while Table 1 and Section 3.1 report a drop from 11.56% to 2.62%.

Field impact

Honourable mention for the HRI 2026 Best Paper Award (User Studies track, full papers), with 2 citations. The paper does not state a release of data, code or fixture designs.

Abstract

We report a mixed-methods field experiment of a conversational service robot deployed under everyday staffing discretion in a live bedding store. Over 12 days we alternated three conditions–Baseline (no robot), Robot-only, and Robot+Fixture–and video-annotated the service funnel from passersby to purchase. An explanatory sequential design then used six post-experiment staff interviews to interpret the quantitative patterns. Quantitatively, the robot increased stopping per passerby (highest with the fixture), yet clerk-led downstream steps per stopper–clerk approach, store entry, assisted experience, and purchase–decreased. Interviews explained this divergence: clerks avoided interrupting ongoing robot-customer talk, struggled with ambiguous timing amid conversational latency, and noted child-centered attraction that often satisfied curiosity at the doorway. The fixture amplified visibility but also anchored encounters at the threshold, creating a well-defined micro-space where needs could “close” without moving inside. We synthesize these strands into an integrative account from the initial show of interest on the part of a customer to their entering the store and derive actionable guidance. The results advance the understanding of interactions between customers, staff members, and the robot and offer practical recommendations for deploying service robots in high-touch retail.

Open Source Robotics