In 2012, researchers ran a randomized controlled trial across 57 stores. Two interventions were tested against untreated controls.
Protective display fixtures reduced theft by 56%. Special handling procedures for high-loss product reduced it by 58%.
Fourteen years later, that is still the strongest positive controlled result in the retail loss prevention literature.
This brief has spent twenty editions arguing that this industry measures the wrong things. This one goes the other way: what does the published evidence actually support, what has it refuted, and why is the file this thin?
What has held up
Three findings survive scrutiny. Note how few there are.
Protective fixtures and high-loss handling procedures. Hayes, Downs and Blackwood, 2012, in the Journal of Experimental Criminology: 57 stores, random assignment, pre/post measurement, chi-square with odds ratios. Fixtures −56%, procedures −58%. The authors themselves flag the limitation: a small sample drawn from a single retail operation.
Fixing inventory records, which pays in revenue rather than loss. Rekik, Glock and Syntetos, for ECR Retail Loss: seven European retailers, 233,000 matched SKUs, a 12-week baseline, a stocktake at test stores only, 12 weeks after. 59.5% of audited SKUs had record discrepancies. Correcting them produced an average 5.98% sales increase — and 14% on high-discrepancy products. This is the rare intervention in our field with an upside number rather than an avoided-loss number.
Actively monitored CCTV, modestly. Piza, Welsh, Farrington and Thomas, 2019: a 40-year systematic review, 80 evaluations in the meta-analysis. Crime down about 13% overall in CCTV areas versus controls, roughly 15% where systems were actively monitored — and no significant effect for passive systems. Retail was not even a category; there were too few retail evaluations to form one.
What backfired
This is the part your vendors will not put in a deck.
Tagging does not have a proven effect. Sidebottom and colleagues, 2017, identified 50 studies on tagging and found only 8 with usable quantitative data. A meta-analysis was impossible — heterogeneity precluded it. None tested baseline equivalence between action and control groups. Of those eight: Beck and Palmer recorded a 251% increase in losses after switching from visible hard tags to invisible soft tags; Downs and colleagues found beige tags +252% while red tags ran −42%; a UK retailer recorded +134%. The College of Policing's formal verdict is that tagging "has not had a statistically significant effect."
Uniformed guards showed no measured effect on shoplifting in Farrington's 1993 five-store experiment — the only impact study of its kind, and confirmed by the one other review that looked.
Anti-theft wraps are mostly null. Hayes and colleagues, 2019: 56 stores, stratified random assignment, mixed-effects model. Significant for cordless drills. Not significant for weight-loss supplements, not for skincare, not for overall retail loss. The authors call it "conditionally efficacious."
CCTV gains decay fast. Beck and Willis studied 15 stores with stocktakes at 13 and 28 weeks. At three months, units stolen fell 28%. At six months the reduction "all but disappeared" — 64 to 63 units a week. Only the most expensive continuously-monitored systems held an effect; the cheaper installations had, in the authors' words, no prospect of ever recovering their installation cost through loss reduction.
A perspective from the field
I have approved every one of these controls at some point, and I never once asked for the trial design. I asked for references and a payback period, which is what the industry trained me to ask for.
The uncomfortable part is not that some of these do not work. It is that the ones that backfired did so by 251% and nobody in the room would have known. We are a discipline that demands evidence from our own investigators and accepts brochures from our suppliers. Those two standards cannot both be right.
Why the file is this thin
ECR Retail Loss did something unusual and graded its own field. Risk Amplification in Retail Stores, updated in May, scores every available study on an adapted Maryland Scientific Methods Scale.
The results are bleak and worth knowing before your next capital request.
EAS and hard tags: 10 impact studies, results ranging from 17% to 93%, all on weak methodology.
CCTV: 3 impact studies — against more than 40 studies of how it gets implemented.
Signage and campaigns: 8 studies, dating from 1972 to 2011. Rated very weak.
Security and sales staff, store design, shelf interventions: one impact study each. The shelf study is described by the reviewers as poorly designed.
ECR's own conclusion: the evidence base is "surprisingly limited for technologies in widespread use."
Clarke and Petrossian, in the US Department of Justice problem-oriented policing guide on shoplifting, put it more bluntly: "few of the common preventive practices have been evaluated." They note that retailers resist rigorous evaluation and that government has barely funded the area.
And the body best placed to fix this — the Loss Prevention Research Council — states on its own research page that members have access to the results. Non-members do not. The largest research operation in this discipline produces almost nothing your readers can independently verify.
Three practical moves for the next 90 days
Ask for the design, not the result. One question to every vendor: was there a control group, and how were controls selected? A supplier who cannot answer has given you a testimonial, not evidence. This costs nothing and will change what gets approved.
Put a sunset date on every control you run. The CCTV evidence is unambiguous that measured effects can be gone by week 28. If a control has never been re-measured since installation, you do not know whether you are funding a result or a memory.
Look at your inventory record accuracy before you buy another deterrent. Nearly six in ten audited SKUs carried a record discrepancy in the ECR study, and correcting them moved sales ~6%. It is the only intervention in this literature that pays you back on the top line, and it competes for budget against controls with far worse evidence.
Closing note
None of this argues for doing nothing. Two interventions have real controlled evidence behind them, and one of them makes money rather than saving it.
It argues for a different standard of proof — the one we already apply to our own casework and have never once applied to a purchase order.
If you have run a controlled test in your own estate, with real controls, I would like to see the design. Reply with anything you can share, anonymized always.
Forward this to one LP or AP leader who should be reading it.
— Gabriel
The LP Brief is a weekly intelligence read for senior loss prevention and asset protection leaders. Free. No vendor noise.
Not yet subscribed? thelpbrief.com