How to pilot a loyalty program before national rollout

When the pilot is proposed, every sales head instinctively nominates his best district — the one with the energetic ASM, the loyal dealers, the counters that answer the phone. It is an understandable instinct and it quietly ruins the pilot, because every number that comes out is flattered: enrolment is easier, activation is faster, fraud is politer. The national forecast built on that pilot then over-promises by 30–50%, and the program spends year one explaining the gap.

The honest design uses two or three districts in a deliberate mix: one strong (to see the ceiling), one average (to see the likely national reality), and ideally one difficult — weak share, a dominant competitor, patchier field coverage (to see the floor). Match them on the structural variables that drive trade behaviour: urban/rural mix, dealer density, competitor intensity, and category seasonality. A cement pilot in peak monsoon or a lighting pilot in the dead post-Diwali quarter measures the season, not the scheme — check the windows against the festive trade calendar. And pick districts served by different distributors: a single distributor's enthusiasm (or sabotage) is a variable you must not let masquerade as a program result.

Trade purchase data is noisy — festivals, local projects, copper moves and dealer credit cycles swing counter offtake 20–30% month to month. To read a genuine 8–10% lift through that noise, you need scale: as a working rule, at least 300–500 enrolled counters (or 800–1,000 influencers, whose individual behaviour is lumpier) across the pilot districts. Below that, the pilot review becomes an argument about anecdotes.

The control group is what converts numbers into evidence. Build a matched set of counters — similar pre-pilot purchase volume, outlet type and geography — that stays out of the program, and measure both groups identically through dealer secondary data or field audits. Two practical rules: put controls in an adjacent similar district, not the same town (counters talk; nothing contaminates a control like a neighbour showing off UPI credits), and freeze the matching before launch — selecting controls after you have seen results is the polite name for cheating. The growth gap between enrolled and control counters is your incrementality; the pilot group's own growth alone is marketing. This is the same matched-control method the program will need forever, as covered in the KPI guide — the pilot is where you build the muscle.

One compliance note that pilots forget: Section 194R applies from the first rupee of benefit. Capture PAN at pilot enrolment and run the real TDS machinery — 10% once cumulative benefits cross ₹20,000 per PAN in the financial year — because the pilot is precisely where you want to discover that the TDS pipeline, not just the payout rail, works end to end.

The single highest-leverage act in pilot design is writing the pass/fail thresholds down before launch, signed by the sales head and finance. Typical gates for a counter program:

• Adoption: 30-day activation above 50%; steady-state monthly actives above 35% of enrolled.
• Impact: incremental lift vs control of at least 5–8% of secondary volume, with premium-mix movement in the intended direction.
• Operations: payout success above 98%; median helpdesk resolution under 48 hours; confirmed fraud under 1–2% of reward spend.
• Economics: pilot unit costs projecting to a national payback of 9–15 months at the budget band in our budgeting guide.

Then the review has only three outcomes: scale (gates passed — roll out in district waves, using later waves as fresh temporary controls), iterate (mechanics work, one gate missed — fix the specific failure and extend the pilot 4–6 weeks; most commonly the reward level or the onboarding flow), or stop (structural failure — the audience does not value the reward, or the channel structure defeats verification). Stopping after a ₹40 lakh pilot is a success of the method: the same failure at national scale costs crores and years of trade goodwill. What pre-agreed gates prevent is the fourth, most common outcome — the ambiguous pilot that scales anyway because momentum and sunk pride demand it.

• The over-supported pilot. Head office visits weekly, the platform vendor's founder personally unblocks payouts, the ASM does nothing but the pilot. None of this exists in district 247. Ration the support to what the national model will actually provide — one training session, standard helpdesk, normal beat frequency — or run the last four weeks with support deliberately withdrawn and watch what happens.
• Dealer hand-holding that won't scale. In pilots, dealers often personally walk retailers through scans because the RSM asked them to. At national scale that favour evaporates. Design the pilot so enrolment and scanning survive dealer indifference — because nationally, indifference is the median.
• The launch-bonus mirage. Extra-generous pilot-only rewards ("double points for pilot districts") produce adoption the national economics cannot repeat. Pilot the real scheme; if you must add launch energy, use time-boxed boosters you can also afford nationally.
• Measuring the festive quarter. A pilot spanning Diwali will show lift that is seasonality wearing a program badge. Either avoid the window or lean harder on the control group, which experiences the same festival.
• Cherry-picked reporting. Decide the report template — every gate, both groups, all districts — before launch, and publish the difficult district's numbers with the same prominence as the strong one's.