NPS verbatim analysis that names the aircraft, the seat and the team that fixes it
Every lost NPS point tied to its cause, the flights behind it and the team that fixes it.
See one case, screen by screen ↓NPS verbatim analysis, also called NPS comment analysis, reads the open comments behind Net Promoter Score survey answers to explain why the score moved. Airline NPS Verbatim Analysis does it for airlines: eight agents read every post-flight survey comment the night it arrives, match it to the flight, aircraft, seat and loyalty tier, rank the causes by the NPS points each one cost, and prove each in the airline's records. The insight analyst decides what goes to which owner.
How to choose AI for airlines: a buyer's guide to agentic AI solutions →
Updated
Most NPS verbatim analysis at an airline, like most open-ended survey analysis, is still a sample and a topic tree. A post-flight survey goes out 24–48 hours after arrival, tens of thousands of comments come back each month in many languages, and senior managers read a handful a day. The rest go through survey text analytics that tags them “seat” or “food” against a topic list analysts keep by hand, so a new problem goes uncounted until someone adds a rule. Joining a response to the operating tail, the seat or the catering load is a warehouse project most programs never finish.
So the monthly deck explains the score with a word cloud and a few quotes, and NPS driver analysis, where it is done at all, runs on touchpoint ratings rather than on the comments. It cannot say which aircraft, which seat, or that the technical log already has the write-up. A route-cabin score swings by ±8 to ±11 points on a few hundred responses, so teams chase noise or miss a real drop. Nobody puts the premium revenue at risk next to the cause, the owner hears about it by email, if at all, and nobody measures whether the fix moved the score.
Estimated split for one month's insight report at a network carrier's CX insight team, by hand and with the solution.
Eight agents read every comment, find and prove the causes, check each other and brief the owner; the analyst decides.
Corvane Airways read 61,284 post-flight survey comments in September. Overnight, the agents flag that business-cabin NPS on Chicago O'Hare–London Heathrow fell from +42 to +31. Here is how Talia Brennan, the insight analyst for long-haul, works it through with the agents, from the question to a cited brief in Maintenance's hands.
The agents read 2,148 new comments overnight. Talia's Today screen shows network NPS at +34, −2 against August on 186,412 surveys, 61,284 comments read in September, and 5 things waiting for her. At the top of what the agents found: business, ORD–LHR fell 11 points, n = 412, beside five older findings: three already sent to owners, one piece of good news and one to watch.
“Business, ORD–LHR fell 11 points · Seats that will not go flat on 2 aircraft · Found today 05:52, as late September surveys arrived.”
The change watcher tested 1,128 route, cabin and tier cells. A change is shown only when it is at least 4 points, outside the 95% range and on at least 50 responses. Five changes qualify; Economy on MIA–GRU, down 5, does not, and is labeled “watch, not a change”.
“A fall of 5 inside a range of ±14.5 at n = 233. The agents will look again when October closes.”
She asks in plain words, and five agents go to work in view. They read 412 responses and 371 comments in 4 languages with names masked, match every response to its flight and tail and 409 to a seat, group 87 detractor comments into causes with no fixed list, check each cause against the technical log, inventory, catering load plan, facilities and on-time data, and price the risk from Platinum and Gold members.
“Matched each response to flight, tail, seat and tier — all 412 to flight and tail; seat 409 (3 had no record locator).”
Four causes explain most of the fall. The largest: business seats that will not recline or go flat, 38% of detractor comments (33), −5.6 points, 94% confidence, proved by records. Then the first meal choice gone before the last rows (−2.3), a crowded temporary lounge at ORD (−1.5, possible) and late arrivals on CV 971 (−1.1). Each rung opens to the passengers' own words, with score, tier, flight, date, tail and seat — a Spanish comment beside its translation.
“Seat 3A would not go flat. Crew moved me twice and the second seat stuck halfway too.” — Platinum, score 2, CV 972, Sep 14, N812CV
The proof follows a comment to its reservation, flight, aircraft and technical-log entry. Business NPS by aircraft splits the route: N812CV −1 and N814CV +13, with 5 and 6 seats placarded INOP, against +45 to +49 on the other four 787-9s. Together the two tails score +6 against +47. The recline actuator, CS-ACT-2210, is back-ordered to Oct 28, and the placarded seats were sold and flown 23 times in September.
“Read-only. The agents never change a technical-log entry or a deferral.”
A second agent checks every answer before Talia relies on it. The fall of 11 is outside its ±10.9 range, z = 1.98, and the seat cause does not depend on it. Platinum members answer at 11.8% against 3.1% for non-members; 38 of 412 responses came from two disrupted CV 971 flights, and without them the fall is −9. The lounge rung has only 12 comments and a closure that ends Oct 30, so it is marked possible.
“Operations linker → cause finder: 4 of the 33 seat comments came from other aircraft with no technical-log entry. Kept as comments, not as proof.”
Talia narrows the question with one click. The cause finder re-runs it for Platinum and Gold only: +40 → +22, −18 on 187 responses, and seats are an even bigger share of their detractor comments at 44%. The survey-quality agent confirms the fall is outside its ±16.2 range on the smaller sample, and the revenue at risk stays $2.5M a year.
Ask back: “Is it only those two aircraft?” · “Exclude the two disrupted flights” · “When will the seats be fixed?” · “What is at risk if we don't fix it?”
Four hundred seventeen Platinum and Gold members flew business on N812CV and N814CV in September, spending $15,600 each a year with Corvane. Corvane's own history says members who score 0–6 cut next-year spend by 38%, so $2.5M a year is at risk, estimated. The five largest open causes come to $6.1M a year. Members are never contacted by the agents; a recovery list goes to Loyalty only if someone approves it.
417 members × $15,600 × 38% = $2.5M a year at risk (estimated).
The hand-off writer drafts the brief to Dale Hutchins in Cabin Interiors Engineering, cc Renee Castillo, Product Manager, Cabin: 21 detractor comments from Platinum and Gold members (44%), 17 write-ups, 11 seats placarded INOP, and the ask — replace recline actuators on 11 seats. The measure: business NPS on the two tails back to the other tails within 6 weeks of the fix. Reply due Fri, Oct 23, five business days. Talia reads it and clicks Approve and send.
“Attached: evidence pack (21 masked comments, 17 technical-log write-ups, inventory extract). Passenger names never leave this solution.”
Owners shows every hand-off: the seat brief sent, reply due Oct 23; the meal brief ready for Gwen Talbot in Inflight Catering; Loyalty Operations two days late on the Meridale miles, with a Nudge; DEN station's reply due today; and Digital's agreed fix for the upgrade list in release 7.14 on Oct 22. After a fix, the agents measure NPS on exactly the affected flights and propose closing or reopening the cause.
“The hand-off writer sends nothing until you approve the brief, and never closes a cause on its own.”
The same NPS drop, seen by the four people who carry the voice of the customer at the airline — what their month looked like, and what it looks like now.
Eight agents read every comment, find and prove the causes, check each other and brief the owner; the analyst decides.
Reads every survey comment the night it arrives in 11 languages, keeps the original beside the translation, splits it into statements, masks names and contact details, and routes complaint-like comments.
Joins each response to its flight, date, operating aircraft, seat after any reseat, cabin, fare and loyalty tier.
Groups statements into causes in plain words, attributes the NPS change to each, writes the cause ladder with the remainder kept as mixed or unexplained, and re-cuts it on request by tier, aircraft or flights.
Confirms or rejects each cause against the airline's records: technical log and deferrals, seat inventory, catering load plans and run-out reports, on-time data, facilities notices, app releases and loyalty suspense.
Second reader on every answer: the change against its 95% range, response-rate skew by tier, disrupted flights and language coverage, working from the ISO 20252:2019 disclosures.
Tests every route, cabin and tier cell each night against the thresholds and raises only real changes.
Prices each cause by the Platinum and Gold members it touched, their 12-month spend and the airline's own spend change after a 0–6 score, and shows the multiplication.
Drafts the cited brief to the owning team, tracks the reply clock, proposes a nudge when an owner is late, and measures NPS on the affected flights after the fix.
Decide, send and sign. The agents propose; a named person decides.
Ask from any screen, in plain words. Every answer cites its source, and anything it proposes waits for your approval.
Why did business-cabin NPS fall on ORD–LHR?
From +42 to +31 in September (n = 412; range ±10.9). The biggest cause: business seats that will not go flat on N812CV and N814CV, 38% of detractor comments and −5.6 points. The technical log has 17 seat write-ups and 11 seats placarded INOP on those two aircraft. Then the meal choice running out (−2.3) and the ORD temporary lounge (possible).
Which causes put the most premium revenue at risk?
Estimated, from Platinum and Gold members: seats on N812CV and N814CV $2.5M a year, Meridale miles $1.8M, upgrade list $1.1M, meal choice $0.4M, DEN boarding $0.3M. $6.1M in all.
Send the meal cause to Inflight Catering
I drafted the brief for Gwen Talbot: 18 detractor comments (21%), −2.3 points; load plan LP-ORD-J-0901 cut entrées from 34 to 30 per flight. Asked: restore 34 or open pre-order. Please review and approve; nothing is sent before you do.
Is the Platinum drop real or survey noise?
Real. Platinum NPS fell from +46 to +42 on 9,214 responses; the 95% range is ±2.3, so 4 points is outside it. 29% of Platinum detractor comments name the upgrade and standby list that app release 7.12 moved on Sep 3. Digital agreed a fix for Oct 22.
13 screens from the working solution, on its sample data. Pick one to see it large.
What counts as a real change, who owns each cause and how long they have to reply are settings, applied tonight and on the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| Minimum responses to show a score | 50 | Step up or down by 10; cells below show “too few responses” |
| Confidence for a real change | 95% | Step up or down one point; the change must fall outside this range |
| Smallest change to alert on | 4 NPS points | Step up or down one point; a preview names the changes that would drop out |
| Below this confidence a cause is “possible” | 75% | Step up or down by 5, for the survey-quality agent |
| Reply clock | 5 business days | Step up or down a day; the Head of CX is told after 2 more days |
| Owner routing | Cause family → owning team | Aircraft and cabin · Food and drink · Airport · Digital · Loyalty · Punctuality · Complaint — change each one |
| Mask passenger names and contact details in quotes | On | On · Off; for CCPA/CPRA and GDPR, quotes show tier, flight and seat, never a name |
| Keep comments and evidence | 25 months | Then deleted, with the audit trail kept |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample survey data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (a 2024 vendor case study of a large Latin American carrier's survey program — senior managers review 10 randomly selected comments or videos a day; Reuters, 2025 — about $8.2B from a co-brand card partner at a US network carrier). 14 CFR 259.7, 14 CFR 382.155, CCPA/CPRA, GDPR and ISO 20252:2019 are named as context only. NPS is a registered trademark of Bain & Company, Satmetrix and Fred Reichheld, used here descriptively. Corvane Airways, Meridale Air and the people named on this page are fictional — characters and sample data in the working solution.
NPS verbatim analysis, also called NPS comment analysis, reads the open comments behind Net Promoter Score survey answers to explain why the score moved. Airline NPS Verbatim Analysis does it for airlines: eight agents read every post-flight survey comment the night it arrives, match it to the flight, aircraft, seat and loyalty tier, rank the causes by the NPS points each one cost, and prove each in the airline's records. The insight analyst decides what goes to which owner.
NPS text analytics tags topics: “seat” or “food”. Here a cause is proved by a record: on the sample route, 33 seat comments lead to 17 technical-log write-ups and 11 seats placarded INOP on two aircraft. Nobody keeps a fixed topic tree; the agents suggest new causes in plain words and an analyst approves, renames or rejects them.
Every change is shown with its 95% range. The change watcher raises a change only when it is at least 4 points, outside the range and on at least 50 responses, and the survey-quality agent checks every answer for response-rate skew by tier and disrupted flights. A fall inside the range is labeled “watch, not a change”.
Yes. The flight matcher joins each response to its flight, date, operating aircraft, seat after any reseat, cabin, fare and tier from reservations, departure control and flight operations. A response with no matching record is marked unmatched, never guessed.
It can be. A comment that describes a specific problem is routed to Customer Relations untouched, and disability complaints under 14 CFR 382.155 too. The agents never answer a passenger.
The revenue-at-risk agent counts the Platinum and Gold members a cause touched, their 12-month spend and the airline's own spend change after a 0–6 score, and shows the multiplication. On the sample route: 417 members × $15,600 × 38% = $2.5M a year at risk, labeled estimated.
Yes. The analyst accepts, rewords or rejects every cause, with a reason code for a rejection, and approves every brief before it is sent. The Head of Customer Experience signs the monthly report. The agents never contact a passenger or member, never change an operational record and never close a cause.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your survey program, owner directory and a month of past comments — and it goes live once every quality gate has passed. A first-time setup connects the sources and ends with a test question the agents answer end to end on last month's data.
Stolen miles stopped before the award is flown, and honest members never locked out.
1-min demoMissing Miles Claims ProcessingMissing miles credited in minutes, each one tied to the flown coupon and its record.
1-min demoLoyalty Partner Billing ReconciliationEvery partner statement tied out line by line, every missed mile billed or credited.
1-min demoWe'll run Airline NPS Verbatim Analysis on a month of your own post-flight survey comments and the flights behind them.