Crew pairing optimization that shows what every dollar buys — every trip legal and cited
Every trip legal and cited, at the cost and crew quality of life you chose.
See one case, screen by screen ↓Crew pairing optimization builds an airline's monthly crew pairings — multi-day trips that start and end at a crew base — so every flight is covered by a legal crew at the lowest cost. Crew Pairing Optimization is airline crew pairing software with seven agents that build many solutions, check every pairing against Part 117 and the contract, replay them against two years of real delays and explain each trade-off, so the pairing analyst chooses and the Director of Crew Planning signs.
How to choose AI for airlines: a buyer's guide to agentic AI solutions →
Updated
Crew pairing optimization is how a network airline cuts its monthly schedule into pairings — multi-day trips that start and end at a crew base — so every flight has a legal crew at the lowest cost. Those pairings decide most of the crew bill (credit, rigs, hotels, deadheads), most of what pilots and flight attendants feel about their month (red-eyes, short rests, long sits, days away) and how the operation survives a bad day. Today a crew pairing optimizer runs overnight and gives one answer; analysts tune weights by trial and error and explain the result to the director and the union scheduling committee in spreadsheets.
The rules move faster than the tool. Part 117 tables, the flight attendant rest rule, the pilot contract, new letters of agreement and hotel contracts all shape a legal pairing, and a new letter of agreement can take months to encode. Cheap pairings with tight aircraft changes and minimum rests save money on paper and become reassignments and reserve call-outs on the day. And the director signs a total without seeing what was traded for it.
Estimated split of an analyst's working days on one monthly pairing problem, by hand with an overnight optimizer and with the solution.
Seven agents build, check, replay and explain the month's pairings; the analyst chooses and the director signs.
Corvane Airways must publish November's 787-9 pilot pairings — 2,386 flights from the JFK and LAX bases — to the bidding system by Wednesday 7 October, 17:00. It is Monday morning. Here is what happens next, in the working solution.
Overnight the agents read November's schedule, applied LOA 26-07, built 46 solutions and replayed each against two years of delays. Megan's Today screen shows the deadline — 2 days 9 hours left, committee review included — the agents' pick, solution 3 at $41.2M, $0.3M under the $41.5M budget, and 93.4% of pairings holding on replay against 90.1% for October's method. Below it: what needs her, and the pairing agents' overnight notes.
“November’s 787-9 pilot pairings are ready for you to choose.”
The schedule reader compared November's file from network planning with October's: 2,386 flights, 22,180 block hours, 31 changes — 14 new, 9 retimed, 5 canceled, 3 moved to the 777. CV 95 JFK–EZE now carries a 10:55 block time, but October's actual median over 34 flights was 10:20, so it drafted a question for network planning. A late Thanksgiving extra section, CV 2416/2417 on 29 November, arrived after the run started and waits for a re-run of that day.
“I never edit the schedule; I ask network planning.”
Every rule the builder and the checker apply carries the passage it comes from: Part 117, the flight attendant rest rule, the pilot contract, letters of agreement and hotel contracts — 17 in all. LOA 26-07, a 24-hour layover at NRT after a flight over 12 hours block, is in effect, approved by the director and acknowledged by the committee. The committee's new letter, LOA 26-08 — 75-minute aircraft changes at JFK — is drafted and waiting: 61 pairings in solution 3 have an aircraft change under 75 minutes at JFK, an estimated +$0.2M if applied to November.
A rule takes effect only after the director approves it; contract rules also need the committee's acknowledgment.
All 46 solutions sit on one chart, with the budget line, the quality-of-life floor of 70 and a ring on each point for the share of pairings that hold on replay. The agents recommend solution 3, the cheapest above both floors: $41.2M, quality of life 74, 1.8% deadhead, 6,412 hotel nights, 188 red-eyes followed by more flying. Point at solution 4 and the explainer says what it buys: crews lay over in Tokyo 12 more times instead of flying the Manila shuttle on arrival day, for +$0.3M.
“Solutions 1 and 2 are cheaper but under the 92% robustness floor (88.2% and 90.6%).”
Megan types “Fewer red-eyes mid-trip, stay under $41.4M”. Before it starts, the pairing builder states its reading: fewer pairings that fly on after a red-eye — 188 in solution 3 — with a cost cap of $41.4M, starting from solution 3, as run 5. It finds the pairings the request touches — 188 LAX–JFK–LAX pairings, 130 on CV 1442 and 58 on CV 1446 — and re-opens only those 30 days: 412 flights, 2,118 candidate pairings priced.
“The schedule’s 214 red-eye flights themselves cannot change.”
The builder never certifies its own work. The legality and fatigue checker re-tests all 1,149 pairings — 0 illegal, 3 below alertness 65 reworked in the same run — and the operation replayer flies the new solution over 730 days of delays: 93.6% hold, up from 93.4%, with 74 fragile pairings instead of 76. Solution 8 lands on the frontier at $41.3M, quality of life 77, 142 red-eyes mid-trip — still within budget and above both floors.
“Solution 8: −46 red-eyes mid-trip for +$0.1M.”
In solution 3 a LAX crew flew the CV 1442 red-eye to JFK, rested 10:05 — 5 minutes above the 10-hour minimum — then flew CV 1449 back; when CV 1442 landed after 06:45 the next duty became illegal, and in September that happened 14 times. In solution 8 a JFK crew flies CV 1441 out, lays over 23:50 in LAX and flies the CV 1442 red-eye home. Every duty shows its Part 117 limit and basis, every check is cited, lowest alertness rises from 66 to 74 and the trip holds on 97% of replay days.
Each check names its rule: “Flight duty Mon 16 Nov: 6:30 of 9:00”, with its basis, Table B · 1 segment · report 00:00–03:59 (JFK time).
Megan chooses solution 8 and submits it with her note. The approval pack sets it against its neighbors on the frontier and October's method — cost, quality of life, hold on replay, red-eyes mid-trip, deadhead, hotel nights — and says what each neighbor gives up: solution 3 saves $0.1M but has 46 more red-eyes mid-trip, solution 4 costs $0.2M more for quality of life 78. 1,149 pairings are re-tested and LOA 26-07 is in effect. The publish agent shares it with the Scheduling Committee and the director, and Capt. Kessler comments in the same thread.
“The red-eye change is what we asked for in September. No contract questions from the committee. Review complete.”
Megan cannot approve her own solution, so it goes to Daniel. He picks a reason code — A1 Best balance, A2 Lowest cost above the floors, A3 Contract commitment or A4 Operational request — adds a note for the record if he wants, and signs. Approval lets the publish agent re-verify and send the pairings to the bid system.
“I approve November’s 787-9 pilot pairings as Daniel Whitaker, Director of Crew Planning — electronic signature.”
The publish agent checks the signature and the completed committee review, re-verifies the 1,149 pairings against the latest schedule and publishes them to the bid system, with the hotel forecast sent to hotel booking — two days before the deadline; the bid opens Thursday 8 October, 17:00. Every step is on the audit trail. On the same November schedule, the month comes in $0.8M below October's method.
“1,149 pairings are in the bid system; the hotel forecast is with hotel booking. Every step is on the audit trail.”
The same November pairings, seen by the four people who carry them — what their month looked like, and what it looks like now.
Seven agents build, check, replay and explain the month's pairings; the analyst chooses and the director signs.
Reads each month's schedule file (SSIM) from network planning, compares it with last month, and asks network planning about doubtful flights with the evidence — such as a block time that differs from operations control actuals.
Turns Part 117, the flight attendant rest rule, the pilot contract, letters of agreement and hotel contracts into rules, each with its source passage, and estimates which pairings a change touches.
Builds the month's candidate solutions under the objectives, caps and floors, and re-runs on a plain-words request — saying how it reads the request and re-opening only what it touches.
Re-tests every pairing of every solution against Part 117, 14 CFR 121.467, the contract and the alertness threshold, independently of the builder — computing the limit, the value and the margin for every duty and rest.
Flies each frontier solution against 730 days of real delays and counts the pairings that would break, with the cause.
Writes one sentence per solution and per difference, compares months planned and operated, and answers questions with citations.
Re-verifies an approved solution against the latest schedule and sends it to the bid system, with the room-night forecast to hotel booking.
Chooses; the committee reviews; the director approves and signs. The agents propose; a named person decides.
Ask about any solution, pairing, rule or schedule change — or tell the agents what to change. Every answer cites its source, and nothing is published without the director's signature.
Why do the agents recommend solution 3 over the cheapest?
Solutions 1 and 2 cost $0.6M and $0.3M less, but only 88.2% and 90.6% of their pairings hold on 730 days of real delays — under the 92% floor. Solution 3 is the cheapest above both floors: $41.2M, quality of life 74, 93.4% hold.
What does solution 4 buy for $0.3M?
28 pairings differ: 12 crews lay over 24 hours in Tokyo instead of flying the Manila shuttle on arrival day, and a second crew flies the shuttle: +$0.21M credit, +$0.04M hotels, +$0.05M positioning, and about 4 fewer long duty days a week.
Is the CV 95 block time right?
October's actual median over 34 flights was 10:20; the November file says 10:55. I never change the schedule myself. I have drafted the question for network planning for you to check and send.
What would LOA 26-08 change?
From the December bid month, aircraft changes at JFK would need at least 75 minutes. In solution 3, 61 pairings have a shorter change at JFK, and changes under 60 minutes there are the biggest cause of breaks on replay; about +$0.2M if applied to November. It takes effect only after Daniel Whitaker approves and the committee acknowledges.
13 screens from the working solution, on its sample data. Pick one to see it large.
What the agents optimize and the floors they may not cross are settings per pairing problem, saved with a signature and on the audit trail. Save a new floor or budget and the trade-off explainer re-checks every frontier solution against it, saying which fall outside a limit and whether the pick changes; solutions are not re-run until you start a run.
| Setting | Default | Choose from |
|---|---|---|
| Budget | $41.5M · 787-9 pilots, November | Set per problem and month |
| Robustness floor | 92% of pairings hold on replay | Any share; solutions below it are outside the floor |
| Quality-of-life floor | 70 | A score from 0 to 100 |
| Fatigue: flag below | Alertness 65 | Flagged for a person |
| Fatigue: never publish below | Alertness 60 | Hard stop |
| Quality-of-life weights | Red-eyes followed by more flying 30 · Rests near the minimum 20 · Commutable report and release 20 · Sits over 3 hours 15 · Days away 15 | Sliders, total 100 — what the committee measures |
| Approval reason codes | A1 Best balance · A2 Lowest cost above the floors · A3 Contract commitment · A4 Operational request | B1 Over budget, approved trade — required to propose a solution above budget |
| Roles | Pairing analyst · Director · Scheduling manager · Committee representative · Administrator | The analyst cannot approve her own solution; the committee and the administrator cannot approve |
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 data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (salaries, wages and benefits for all employees of $16.0B of $51.6B operating expense at a large US network carrier, 2024 Form 10-K; DOT consent order, 2023, $140M). Regulations named: 14 CFR Part 117 (117.11, 117.13, 117.17, 117.23, 117.25) and 14 CFR 121.467. Corvane Airways and the people, pairings, letters of agreement and contract sections named on this page are fictional — characters and sample data in the working solution.
Crew pairing optimization builds an airline's monthly crew pairings — multi-day trips that start and end at a crew base — so every flight is covered by a legal crew at the lowest cost. Crew Pairing Optimization is airline crew pairing software with seven agents that build many solutions, check every pairing against Part 117 and the contract, replay them against two years of real delays and explain each trade-off, so the pairing analyst chooses and the Director of Crew Planning signs.
Pairing comes first: it is the trip-building step, and it builds the trips without naming who flies them. The approved pairings are then published to the preferential bidding system, where pilots and flight attendants bid for their month — in the demo the bid opens 24 hours after the publish deadline, closes six days later, and awards follow five days after that.
An independent legality and fatigue checker re-tests every pairing against the flight time limits (117.11), the flight duty period tables (117.13 Table B, 117.17 Table C), rest (117.25), cumulative limits (117.23), the flight attendant rest rule (121.467), the pilot contract and the alertness threshold. Each check shows the limit, the value and the rule it applied; the builder never certifies its own pairings.
Yes. The rules keeper drafts a new letter of agreement or hotel contract change into a rule with its source passage and estimates which pairings it touches — 61 for LOA 26-08 in the demo. It takes effect only after the director approves it, and contract rules also need the union committee's acknowledgment.
The operation replayer flies every frontier solution against 730 days of real delays and counts the pairings that would break, with the cause — tight aircraft changes, minimum rest after a late arrival, long-haul duties near their limit, late deadheads. Robustness is a floor you set, and it is ringed on every point of the frontier.
The schedule reader flags it, and only the affected day is re-run rather than the whole month. Doubtful flights, such as a block time longer than last month's actuals, go to network planning as a question with the evidence — the agents never edit the schedule.
Yes. The agents build, check, replay and explain; they never change the schedule or a rule, and never publish on their own. The analyst chooses and cannot approve her own solution, the union scheduling committee reviews, and the director approves with a reason code and an electronic signature before the publish agent re-verifies and sends anything to bidding.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your contract, letters of agreement, schedule files and delay history — and it goes live once every quality gate has passed. We will walk you through it on one of your own past bid months first.
Every broken pairing re-crewed legally before it costs a flight, crew told in 20 minutes.
1-min demoFlight and Duty Time ComplianceNot one illegal duty day published, and every limit shown with the rule that sets it.
1-min demoFatigue Risk Management SystemEvery fatigue report tied to its roster and the science the same day, fixed by the board.
1-min demoWe'll run Crew Pairing Optimization on one of your own past bid months — your schedule, your contract and your delays.