Current (est.)
~70%
The signals Uber rewards, where Patri stands today, and the 30/60/90 day plan to earn the rank that makes the Feb partnership pay for itself.
A restaurant can buy impressions with ads. It cannot buy rank. Rank is what determines whether a hungry customer searching "Indian near me" sees Patri at position 1 — where conversion is 4× higher — or position 12, where they don't see it at all. The marketing investment Uber is offering in the Feb proposal accelerates whatever rank Patri already has. If we are operationally weak, the ads waste themselves on a listing the algorithm refuses to push. Earn the rank first; the marketing then compounds it.
This is the simplified ranking equation we are now operating to:
The weights are not published — but the directional logic is. Below is each signal, the lever we control, and the date by which we own it.
The single most punitive signal. Decline an order → algorithm interprets it as "this restaurant is unreliable" → rank drops, the offer doesn't refill. Uber's Feb proposal sets 98% as the floor. We hold ourselves to 99%.
How often the listing is actually open during published hours. Every "store closed" minute during peak is a rank hit. Common cause: staff forget to switch back on after a tablet restart.
This is the most underrated signal. Set 18 min, food ready in 18 min → courier waits zero, customer's tracker is honest, rating rises. Set 18, deliver in 28 → courier dwells, customer panics, rating tanks twice (food and delivery time).
Below 4.5 the algorithm de-ranks aggressively. 4.6–4.7 is the West London average. 4.8+ is the Top Eats threshold. We sit at an estimated 4.6 — we close the gap with packaging + reply discipline, not promo discounts.
What share of accepted orders make it to the customer. Affected by cancellations after acceptance, missing items, and refunds. The Feb proposal does not name this signal directly but Uber tracks it relentlessly.
How many customers save the restaurant to favourites or return within 30 days. The single best signal that the kitchen is making someone genuinely happy. Uber uses it to weight relevance in personalised feeds.
Photo on every item. Modifiers configured. Dietary tags (vegan, halal, gluten-free). Description ≥10 words. The algorithm filters by these tags before it sorts by rank — incomplete menus are invisible to half the demand.
Uber One subscribers (40% of customer base in the Feb proposal) see Eats Pass-eligible restaurants more often in feed. Joining costs Patri only the small discounted-delivery exposure — and Uber's CRM places Pass-eligible restaurants in front of higher-LTV customers.
The destructive twin of acceptance rate. A cancellation after the customer has paid is worse than a decline at the start; it nukes both rating and rank.
Each signal individually moves rank by single-digit percentages. The compounding effect is what matters: a listing that is +5pt on every one of the nine signals doesn't move +45pt — it moves to a different ranking band, because Uber's algorithm scores in a non-linear way. Move from "middling" to "trusted" and the listing is shown to a different customer pool: higher-LTV, higher-rating-history, more Uber One subscribers. The marketing then compounds against that pool.
This is the Acton Deliveroo lesson again, restated for Uber: when the operational discipline catches up, the algorithm does the marketing for free. Marcus's line — "the obstacle is the way" — is literally what an algorithm reward curve looks like.
Signals moved: Online · Acceptance · Cancellation · Eats Pass
Signals moved: Menu · Save Rate · Customer Rating · Prep Accuracy
Signals moved: Volume · Rank · the badge itself
| Signal | Last week | Target | Owner |
|---|---|---|---|
| Acceptance rate (7d) | — | ≥99% | Asim |
| Online ratio (7d) | — | ≥99% | Asim |
| Prep time accuracy | — | ≥95% | Kitchen |
| Customer rating | — | ≥4.80 | Puneet |
| Cancellation rate (7d) | — | <1% | Asim |
| Menu completeness | — | 100% | Asim |
| Orders / 7d (Hammersmith) | — | ≥80 | Puneet |
| Reviews replied / unreplied | — | all replied | Puneet |
One A4 page, eight rows, Monday 9am. If any number is red two weeks running, it gets its own meeting. Everything else stays in flow.