Kirana retailers run their inventory from memory. Reorders slip a week or two past their usual rhythm, festival demand shows up before the shelf is stocked, and higher-margin brands never get tried. All of that happens because nobody is sitting behind the app pointing it out. A big FMCG brand sends a field team to do exactly this in person. An independent kirana has no one doing it for them.
So v1 stays small and honest. It's three hand-written rules, each one explainable in a sentence, and a shop sees at most two cards a week. Nothing builds a cart for the retailer and nothing swaps their brand. A card just gives them a concrete reason to look, and they tap through to the product page and decide for themselves. The plan is to get this simple, debuggable version right first, and leave anything that behaves like a black box for a later version, once this one has earned some trust.
This is an independent concept mockup. It isn't affiliated with or endorsed by Kirana Club. The “KC” mark is a neutral placeholder, not their wordmark, and the product names, rates and margins are taken from publicly visible app screenshots just to make the concept concrete.
यह एक कॉन्सेप्ट डेमो है · concept demo. नया ‘मौके’ टैब पहले से खुला है। पूरी फ़ीड देखने के लिए ‘Feed’ पर टैप करें, या किसी भी कार्ड पर टैप करके देखें।
Each rule runs independently, once a week, per shop. Trigger in plain English, the hypothesis it's betting on, and the metric that would prove or kill it.
For a SKU this shop already buys, it fires when days_since_last_order > 1.2 × the shop's own historical median reorder gap for that SKU. It doesn't look at other shops at all. It only compares the shop against its own past rhythm.
Example: the median gap between Bourbon Biscuit 5/JAR orders is 14 days, today is day 18, so it fires.
Retailers reorder consumables on a fairly fixed personal rhythm. When a shop is about 20% past its own usual gap, chances are they just haven't gotten to it yet, not that they've stopped needing it. A reminder recovers that sale before a competitor's van gets there, or before a customer asks and the shelf is empty.
Tap-through rate on this card, and the share of shops that got it and then placed a matching order within 48 hours. Both are read against a held-out control group of shops with the same depletion pattern that saw nothing.
It fires 10 to 14 days before a festival that's mapped to this shop's region, and only for a product category the shop has ordered from before. It never pushes a category they've never touched. The festival, region and category mapping is a table the team keeps by hand, which is small enough for a 25-person team to maintain.
Example: Diwali is 12 days out and toffee is a category this shop buys, so it fires.
Owners know the festival is coming but are too busy running the shop day-to-day to plan for the spike, so they react once the shelf is already bare. A concrete lead-time nudge with a real number (“~18% this week”) beats a vague “stock up for Diwali”.
Order value in the festival category this cycle, compared with the same shop's value in that category during last year's equivalent window. Where there isn't year-over-year data yet, compare against a matched control group instead.
Fire when a brand's platform-wide order count is growing fast week-over-week, and its margin is at or above this shop's own average basket margin, and this shop has never ordered it. The margin bar is personal to the shop, not a flat threshold.
Example: Ankit Koffy trending up sharply, ₹88 margin vs this shop's ₹40 basket average → fires.
Retailers want to stock what's about to be asked for by name, and rarely have time to comparison-shop new brands across categories. A concrete, above-your-own-average margin number is a stronger reason to try something new than “trending” on its own.
New-brand trial rate, meaning the first-ever order from the suggested brand, and then the number that actually matters, the repeat-order rate on that brand over the next two to three weeks. A trial on its own doesn't prove the hypothesis. A repeat does.
The order is fixed rather than a learned score, so it stays easy to explain. Reorder Due comes first because it's a need that already exists, then Festival Stock-Up if the shop is inside its 14-day window, then Rising Brand only if a slot is still open.
हर शॉप को हफ़्ते में ज़्यादा से ज़्यादा 2 कार्ड मिलते हैं, चाहे तीनों rule एक साथ चलें। ताकि यह सुझाव एक और ऐसा banner न बन जाए जिसे कोई नहीं पढ़ता।
A v2 would add the parts left out on purpose here. That's AI-assembled multi-item carts, swapping a brand for a higher-margin one inside the card, and a brand-reliability guardrail so a margin-driven suggestion never sends a retailer to an unreliable supplier. None of it is worth building until v1 earns it.
Across a few weeks of running these three rules against a control group that gets no cards, the bar is:
If a rule shows no lift over the control group after a few weeks, that's a real answer and worth keeping. A quick, honest no is better than quietly running a card nobody acts on.