NPS & Customer Loyalty Loops

Design the survey before you trust any number

Starbucks' easiest instrument is the receipt or in-app survey, and that is exactly the sample you must not rely on here, because it only reaches people who still bought something. The reported problem is transactions falling, which means the customer who matters most is the one who stopped coming, and a receipt survey structurally cannot see her. So run two instruments: a relational quarterly panel of US coffee buyers that deliberately includes lapsed Starbucks customers and Dutch Bros, Dunkin and independent-cafe regulars, and a transactional survey fired within an hour of the visit and tagged by channel, mobile order pickup, drive-thru or in-cafe. Every response carries the open-ended why and a permission-to-contact flag, because the verbatim is the only part of this that is diagnostic.

Compute the score and read the whole distribution

Illustratively the relational panel comes back at 40 percent Promoters, 38 percent Passives, 22 percent Detractors, so NPS is 40 minus 22 equals plus 18, down from roughly plus 42 two years earlier when the split was closer to 52 / 38 / 10. The strategic fact is not the net number, it is that 38 percent of the base is Passive. In a habit business bought three or four times a week, a Passive is a customer whose morning routine is one bad queue away from belonging to a competitor that is cheaper and faster. Critically, the behavioural data corroborates the survey: comps are negative on transactions, not ticket, so this is not a survey artefact, and NPS is being used only to explain a decline that the P&L has already confirmed.

Segment before diagnosing, and cut by channel first

Cut the transactional score three ways. By channel: drive-thru and mobile-order pickup together account for the large majority of US company-operated orders, and the Detractor mass sits almost entirely in mobile-order pickup at high-volume urban stores between roughly 7 and 9 am, where twenty people crowd a handoff counter with no seat, no name and no barista contact. By daypart: the afternoon cold-beverage occasion scores materially worse than the morning hot-espresso occasion. By geography: China must be scored and interpreted separately, because Luckin reset the local price and speed benchmark, so a US throughput fix does not transfer. That segmentation already kills the lazy hypothesis that the brand has gone stale, and replaces it with a specific one about peak-hour throughput in a specific channel.

Push the Detractor verbatims to a mechanism

Coded comments cluster into three themes: wait time versus the promised time is the largest, value or price is second, and impersonal experience is third. Five Whys on the wait theme: the app promises four minutes and delivers nine, because mobile orders arrive in bursts that are completely uncoupled from what the store can produce, because the app quotes from a store average rather than real-time queue load, and because the drink mix has shifted to cold, customised, multi-step builds such as shaken espressos, refreshers and cold-foam drinks on a store layout and equipment set designed for hot espresso. Baristas then sequence by arrival order, so a fifteen-second drip coffee sits behind four hand-shaken builds. Cross-check against operations using the p90 order-to-handoff time rather than the average, throughput per labour hour at peak, and partner NPS, which had been falling and preceded the customer signal by several quarters.

Attach money to a band change

Illustrative unit economics: a Rewards member visits roughly three to four times a month at a six to seven dollar ticket, so one incremental visit a month is about 75 to 85 dollars of annual revenue per customer. If 20 percent of the roughly 34 million 90-day actives are Detractors, that is about 6.8 million people; converting half of them at 80 dollars each is on the order of 270 million dollars of annual revenue, and that ignores the larger prize entirely. The larger prize is lapse prevention: a daily customer who quietly drops to weekly takes about 1,500 dollars of annual spend with her, and she never fills in a survey to tell you. Note also that the value leg here is not paid-media saving, as it would be for a D2C brand, but frequency and habit retention, so size the prize on visits per member per quarter, not on referrals.

Run both loops, and recognise that the public plan is the outer loop

Inner loop at 34 million scale cannot be phone calls, so make the app the recovery channel: any order that exceeds its promised handoff time by more than a set threshold triggers an automatic apology and a star credit without the customer asking, and Detractor verbatims with contact permission route to the named store manager for personal follow-up. Track recovery rate, the share of recovered Detractors scoring 7 or above at next contact. The outer loop is exactly what Back to Starbucks does: cutting around 30 percent of the menu removes the slow, low-volume, complex builds that clog the peak; the four-minute target plus a sequencing algorithm that interleaves mobile and in-store orders attacks the throughput mechanism; and condiment bars, ceramic mugs and handwritten names attack the impersonal theme. Assign each theme an owner: menu to product, sequencing to digital, labour hours and deployment to store operations, each with a fix date.

Judge it on behaviour, not on the score

Set the score target if you must, but measure success on transaction comps specifically rather than blended comps, because a ticket-led comp is a price increase masking a traffic problem. Track p90 order-to-handoff at peak by store decile, 90-day member reactivation, visit frequency by member cohort, and partner turnover, since the same throughput fix should show up in both employee and customer measures. Add the earned growth read: what share of growth comes from existing customers buying more versus new customers arriving without a discount, which is far harder to game than a survey. Guard the data explicitly, because the moment store bonuses attach to the score you will get please rate us ten prompts at the handoff plane and quietly suppressed receipts.

Split the score by journey stage before anything else

Vidyut runs one annual relational SMS survey with a 9 percent response rate, which is thin and skews to the angry, so treat it as directional and immediately add transactional surveys at three defined moments: 7 days after delivery, 24 hours after every workshop visit, and at the 12-month ownership mark. The split is the whole insight. Delivery-stage NPS is plus 61, essentially unchanged over two years, so people still love the vehicle when they take it home. Ownership NPS collapses to plus 6, and the collapse is concentrated between month 6 and month 9, which is exactly when the first paid service falls due. That single cut tells you this is not a product problem and not a pricing problem, and it saves you from building the entire case around a feature roadmap.

Compute the score and read the composition

Ownership survey: 32 percent Promoters, 42 percent Passives, 26 percent Detractors, giving plus 6, against a split of 61 / 30 / 9 five quarters ago. Promoters have fallen by 29 points while Detractors rose 17, so unlike a simple service-failure story this is a base that has both lost its advocates and gained critics. In a category where the buyer typically researches for six to eight weeks and consults two or three existing owners before purchase, losing Promoters is not a soft loss, it is a direct hit to the top of the funnel. The corroborating behavioural number is the one to lead with in the room: referral share of new bookings has fallen from 38 percent to 21 percent over the same period, which is the same story told by the accounting system rather than by a survey.

Segment by service channel, fleet vintage and geography

Cut ownership NPS by workshop type: 62 company-owned workshops, mostly in metros, score plus 38, while 210 franchised dealer workshops score minus 14, and those franchises serve 68 percent of the fleet. Cut by vintage: vehicles aged 12 to 24 months score worst, because they are out of the first free-service block and into paid parts. Cut by city tier: the worst cohort is tier-2 towns that sold heavily two years ago and sell modestly today. That last cut is the one that unlocks the case, because it says the problem correlates with the size and age of the installed fleet, not with current sales volume, which is a very specific and testable shape.

Push the verbatims to a mechanism with Five Whys

Coded Detractor comments: 54 percent are vehicle stuck at the workshop waiting for a part, 23 percent are range has dropped and nobody will tell me why, 13 percent are no response after I complained. Operational cross-check confirms it: first-visit-fix rate is 46 percent at franchise workshops against 88 percent at company-owned, and average vehicle-down time is 9 days against 2. Five Whys on the parts theme: parts are allocated to dealers monthly on a new-vehicle sales quota, not on fleet-on-road and its age mix, so a tier-2 dealer with 9,000 aging vehicles but 60 units a month of current sales gets starved of exactly the controllers, chargers and brake assemblies his fleet needs. He will not pre-stock at his own cost because parts margin is about 12 percent on 45-day payment terms, so a part is only ordered after the customer is already stranded, adding 7 days of transit from the central warehouse. Root cause is a parts planning rule, not technician attitude.

Size the prize on referral, not on service revenue

Annual service value per customer is small: about two paid services at Rs 1,100 with roughly 45 percent margin, plus accessories and AMC, call it Rs 1,400 of contribution. The money is in referral. A Promoter generates about 0.6 referred buyers a year while a Detractor actively deters, worth about minus 0.3, so a band change is worth roughly 0.45 net referred units a year after discounting for overlap. There are about 1.09 lakh Detractors in the base and roughly 88,000 of them sit under franchise workshops; converting half, 44,000 customers, yields about 20,000 incremental units a year, an 8 percent volume uplift on a 2.4 lakh run rate, worth about Rs 28 crore at Rs 14,000 contribution per vehicle. Add acquisition saving, since a referred buyer costs about Rs 900 against Rs 6,200 paid, so 20,000 referred units save roughly Rs 10.6 crore, and add about Rs 2.3 crore from recapturing service retention, which runs 41 percent for Detractors against 79 percent for Promoters. Total is on the order of Rs 41 crore a year against a fix cost of roughly Rs 22 crore one-time working capital plus Rs 9 crore recurring, so payback lands inside a year.

Close both loops, and trigger the inner one off operations rather than the survey

Inner loop: do not wait for a survey response, because the stranded customer is angriest on day four and the survey reaches him in month eleven. Any vehicle down more than 48 hours auto-creates a case, the area service manager calls within 24 hours, a loaner scooter and free pickup-drop kick in from day three, and the manager holds standing authority to approve goodwill part replacement up to Rs 5,000 without head-office sign-off. Target a 45 percent recovery rate. Outer loop: a weekly parts council ranks stocked-out SKUs by vehicles-down times contribution at risk, and the top cause goes to the supply chain owner with three named fixes, min-max stocking driven by a fleet-age model instead of the sales quota, a dead-stock buyback guarantee so dealers pre-stock without balance-sheet risk, and four regional parts hubs giving a 24-hour delivery radius. Feed employee NPS in alongside: franchise technicians are paid per job card closed, so a vehicle waiting on a part earns them nothing, and that incentive must move to first-visit-fix or the loop will be quietly sabotaged at the workshop floor.

Prove it on behaviour and set the governance rhythm

Set the score target at plus 35 ownership NPS in four quarters, but do not report it alone to the board. Report first-visit-fix rate moving from 46 to 80 percent, average vehicle-down days from 9 to 3, parts fill rate at franchise workshops, service retention at the 24-month mark, and the number that actually pays for the programme, referral share of new bookings recovering from 21 percent toward 35 percent. That referral share is the earned growth read here, it comes off the booking system rather than a survey, and it cannot be gamed by a dealer coaching customers to give tens. Run it as a monthly loop review chaired by the service head with the parts and dealer-management owners in the room, and audit the survey itself, because the moment dealer payouts attach to NPS you will see suppressed contact numbers and coached responses.