What problem does it solve?
A short haul turnaround runs on a tight schedule, and every minute is shared by several ground handling teams (fueling, catering, cleaning, baggage, boarding), working from different companies with their own systems, watched by airport and airline staff who cannot see the whole apron at once. When one subprocess runs late, whoever is watching it often finds out only once the knock on effect already shows up as a late pushback, a missed slot, or a gate conflict with the next aircraft.
Airports and airlines have tried to fix this with radio calls and manual observation for decades. The step change is computer vision trained to recognize turnaround milestones automatically from cameras on the apron, so every stand gets the same real time visibility a supervisor would have if they could watch every gate at once, and a prediction of the departure time that updates as the turnaround progresses.
How does it work?
- Watch the stand. Cameras on the gate or apron are analyzed with computer vision to detect turnaround milestones (chocks on, GPU connected, catering truck docked, cargo doors open and closed, boarding started, pushback).
- Predict the departure time. As soon as the aircraft is on stand, the system builds a live prediction of the off block time and updates it as each milestone happens or slips.
- Alert on deviation. When a subprocess is running behind the plan, an alert goes to the ramp agent, the ground handler or the airport's turnaround coordinator, with which process is late, not just that the flight is.
- Act on the alert. Staff use the extra minutes of warning to reallocate equipment, request help, or flag a likely gate conflict before the next aircraft arrives, rather than reacting after the delay has happened.
- Feed the record back. Every turnaround's milestones and any deviation become a record used for A-CDM (Airport Collaborative Decision Making) reporting, performance review with ground handlers, and to retrain the milestone detection.
- Audience
- Employee facing
- Autonomy
- Assist
- Adoption
- Early adopters
- Channels
- Internal tools
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Cost savings | Not pooled | at least USD 25 million | 1 | 1 vendor |
| Cost savings | Not pooled | CAD 47 million | 1 | 1 vendor |
Value drivers: Lower cost to serve, Speed and cycle time, Risk and loss reduction.
Indicative value
An airline or airport handling 200,000 turnarounds a year at equipped stands
USD 6 million to USD 80 million
Ground delay cost avoided per year
How this is calculated
Formula: turnarounds * costPerMinute * minutesSaved. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Turnarounds per year at equipped stands turnarounds, turnarounds per year | 200,000 | 200,000 | The reference airline or airport. |
| Cost of ground delay per minute of taxi in time saved costPerMinute, USD per minute | 60 | 100 | Editorial assumption blending fuel, crew and downstream schedule cost per minute of taxi in time saved. Replace with your own. |
| Average taxi in time saved per flight minutesSaved, minutes per flight | 0.5 | 4 | The two results on this page come from large hub airports, as reported by the vendor, on average 49 seconds (0.82 minutes) per flight at Seattle Tacoma Airport after almost a year of use, and almost 8 minutes per flight at Toronto Pearson. The low end sits below the Seattle Tacoma average to cover a smaller airport; the high end stays well below the Toronto Pearson figure. |
What it leaves out: Gross ground delay cost avoided only. It leaves out the camera and platform cost and the CO2 value of less time spent with engines or the auxiliary power unit running.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Greater Toronto Airports Authority
Canada · Travel and hospitality · 2024
Toronto Pearson International Airport, Canada's busiest with more than 40 airlines flying to over 160 destinations, sought full situational awareness of its turnaround process across all 106 gates to give a single source of truth for turnaround operations and feed the airport's Airport Collaborative Decision Making (A-CDM) process, and deployed Assaia's TurnaroundControl to do it. The system uses computer vision to track turnaround subprocesses such as fueling and catering against schedule, predicts departure times as soon as an aircraft arrives, and alerts operational staff to deviations so they can intervene before a delay compounds. Assaia's case study credits the deployment with a large reduction in average taxi in time and the resulting annual savings, and a shorter term reduction in ground delays; a Greater Toronto Airports Authority representative is quoted describing improved visibility and resilience across airline partners, ground handlers and service providers without repeating the numeric figures.
- Cost savings: CAD 47 million, per year
"44% Reduction in average taxi-In time resulting in annual savings of CAD $47 million"
Claimed by: vendor
Alaska Airlines
United States · Travel and hospitality · 2022
Alaska Airlines is quoted in Assaia's customer testimonial carousel as using Assaia's platform to improve its aircraft turnaround process. The same case study uses Alaska's 71,000 flights out of Seattle Tacoma International Airport (SEA) in 2019 as a worked example for a projection: it combines an almost 10% taxi in time reduction that Assaia reports for Seattle Tacoma Airport itself, with a separate 4 minute average reduction in ground power unit (GPU) and air conditioning unit (ACU) connection time that Assaia reports from "another major airport in the US", which lets the aircraft's own auxiliary power unit (APU) be shut down sooner. Applying the FAA modelled operating cost per block hour to both results, Assaia projects USD 9.2 million a year in total operating costs eliminated for "a carrier such as Alaska Airlines" at that flight volume. This is a vendor modelled projection, not a measured Alaska Airlines outcome.
No outcome disclosed.
Port of Seattle
United States · Travel and hospitality · 2022
Seattle Tacoma International Airport (SEA), operated by the Port of Seattle, has been using Assaia's Predicted Off-Block Time (POBT) feature for almost a year, per Assaia's own case study on reducing kerosene costs. The system builds a continuous prediction of each aircraft's off block time from live turnaround data, which the case study credits with a reduction in average taxi in time at SEA. Assaia's case study translates that reduction into a per flight and airport wide operating cost saving for SEA, separately from the modelled Alaska Airlines projection recorded in its own evidence record for this use case.
- Cost savings: at least USD 25 million, per year
"an annual total saving of more than $25 million for all flights at SEA"
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Camera coverage of each equipped stand with a usable line of sight to the milestones being tracked
- Scheduled turnaround times and Airport Collaborative Decision Making (A-CDM) data per flight
- Agreement with ground handlers and airlines on which milestones are tracked and who is alerted
Systems to integrate
- Airport operational database (AODB) and A-CDM platform
- Gate and stand allocation system
- Ground handler and airline messaging or alerting channels
- Camera or video management infrastructure already on the apron
Complexity: Medium
Most of the effort is operational, not technical: getting camera coverage and lines of sight confirmed per stand, getting every ground handler and airline on the apron to trust and act on the same alerts, and agreeing who owns which milestone when three or four different companies work the same aircraft.
- 1
Start with the stands that cost the most when they slip
Equip the highest traffic or highest connection risk stands first, where a late turnaround has the most expensive knock on effect, rather than the whole airport at once.
- 2
Agree the milestones and who owns each one
Before switching on alerts, agree with every ground handler and airline which milestones are tracked and who is responsible for acting on a deviation in each one, or alerts get ignored as someone else's problem.
- 3
Predict, then alert, then automate the easy part
Ship the live departure time prediction first, since it is useful with no process change, then add deviation alerts, then automate anything low risk like gate conflict flags once staff trust the predictions.
- 4
Close the loop into A-CDM
Feed confirmed milestones into the airport's A-CDM process so the predicted off block time used for slot and gate planning reflects what is actually happening on the ramp.
- 5
Review with ground handlers, not just internally
Use the turnaround record in regular performance reviews with ground handling partners, so the data improves the service, not just the airline's or airport's own dashboard.
- 6
Extend stand by stand
Add camera coverage and alerts to more stands once the first group shows a clear reduction in ground delay, rather than a big bang rollout across the whole apron.
Guardrails
- Alerts name the specific subprocess and party responsible, not just "the flight is late", so accountability is clear
- Camera footage used for turnaround milestones is not repurposed to individually monitor or score ground staff without saying so
- Predictions and milestone detection are reviewed against actual outcomes regularly, and a stand is paused if detection accuracy drops
- The system supports A-CDM decisions; gate and slot decisions stay with airport operations and air traffic control
KPIs to instrument
- Average taxi in and turnaround time against the pre deployment baseline, per stand
- Ground delay minutes and their estimated cost, per stand and per ground handler
- Alert to action time, how quickly staff respond to a deviation alert
- Milestone detection accuracy against a manually reviewed sample
Human in the loop
Ramp agents, ground handlers and the airport's turnaround coordinator decide what to do with every alert; the system does not move equipment, reassign gates or hold an aircraft itself. Airport operations owns the A-CDM process the predictions feed into, and a joint review with ground handling partners checks the milestone data against what actually happened.
Common failure modes
- Alert fatigue
- Too many low value alerts and staff start ignoring all of them, including the ones that matter. Tune alert thresholds per stand and review false alert rates with the teams receiving them.
- Camera blind spots
- A stand's camera angle cannot reliably see a milestone (for example a cargo door on the far side), and the system reports a false deviation or misses a real one. Confirm line of sight per stand before turning on alerts for it, and flag stands with known coverage gaps.
- Nobody owns the alert
- An alert reaches a general channel instead of the person who can act, and it is lost. Route each milestone's alert to the specific ground handler or team responsible, agreed in advance.
- Data used against ground handling partners without warning
- Turnaround data introduced into supplier reviews or contract disputes without the ground handler ever seeing it live damages the trust the whole system depends on. Share the same data with partners in near real time, not only after the fact.
What are the risks and rules?
EU AI Act
Depends on design
Predicting a departure time and alerting on a late turnaround subprocess is not listed in Annex III: it does not decide access to a service, creditworthiness or employment, and it is not a safety component of an aircraft or a product regulated under Regulation (EU) 2018/1139 that Annex I Section B lists. Annex III point 4(b) covers AI used to monitor and evaluate the performance and behavior of workers; a design that uses the same camera feeds to score individual ground staff, rather than to track turnaround subprocesses, would sit closer to that category, and GDPR applies to any footage that identifies staff.
Guidance
- EASA Artificial Intelligence Roadmap 2.0: A human-centric approach to AI in aviation (European Union Aviation Safety Agency, Europe). Sets EASA's human centric vision and action plan for AI in aviation, including conceptual guidance and rulemaking; the roadmap document itself is the source for the specifics, the landing page carries only a short summary.
Controls to put in place
- Purpose of camera based monitoring documented and communicated to ground handling staff and partners
- Alerts and predictions logged against actual outcomes, with regular accuracy review per stand
- Data sharing agreement with ground handlers on how turnaround data is used in performance reviews
- A stand's alerts are paused, not left running silently wrong, when detection accuracy drops below an agreed threshold
Frequently asked questions
- Does the AI move equipment or hold an aircraft?
- No. Across the deployments described on this page, the system predicts and alerts; ramp agents, ground handlers and the airport's turnaround coordinator decide what to do. Gate and slot decisions stay with airport operations and air traffic control.
- What results have airports and airlines reported?
- Assaia reports that Seattle Tacoma Airport saw an almost 10% reduction in taxi in time (on average 49 seconds per flight) after using its predicted off block times for about a year. The same case study then projects that a carrier such as Alaska Airlines, at Alaska's 2019 volume of 71,000 flights out of Seattle Tacoma, would see about USD 9.2 million a year in total operating costs eliminated, a modelled figure that combines the Seattle Tacoma taxi in result with a separate auxiliary power unit shutdown time result Assaia reports from an unnamed US airport, not a measured Alaska Airlines outcome. In a separate case study, Assaia reports a 44% reduction in average taxi in time at Toronto Pearson, with annual savings of CAD 47 million.
- Does this replace ground handling staff?
- The deployments on this page describe giving staff earlier visibility of a developing delay, not replacing the teams doing fueling, catering, baggage or boarding. The value comes from staff acting sooner on an alert, not from removing the ground handling work itself.
- Is this only useful at large hub airports?
- The public deployments on this page are large hub airports (Seattle Tacoma, Toronto Pearson), where high traffic and tight connections make a minute of turnaround time expensive. A smaller airport with less connection risk would see a smaller return, and cost effective camera coverage per stand is worth checking before committing to a wide rollout.
How to cite this page
Blits.ai AI Use Case Library, "AI for aircraft turnaround and ground handling optimization", last verified 30 September 2026, https://www.blits.ai/ai-use-cases/ground-handling-and-turnaround-optimization. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 30 September 2026: Published after review by an automated review workflow (independent skeptic review).
- 30 September 2026: Editorial fix after a second, adversarial review: removed the unsupported 'both mature deployments' claim from the minutesSaved note, added a Port of Seattle / Seattle Tacoma Airport evidence record so the Seattle Tacoma figure has its own public source instead of living only inside the Alaska Airlines record's note, downgraded the Toronto Pearson evidence stage from scaled to production (the 106 gate rollout is a stated requirement, not a confirmed completed one), corrected the Alaska Airlines evidence verification note's description of the $9.2M calculation, added a later Wayback capture with a named Alaska Airlines testimonial, softened 'usually computer vision' to 'often', replaced WhatsApp with email in blitsAi.howToBuild to stay within the platform feature inventory, made the Seattle Tacoma / Seattle-Tacoma spelling consistent, and reworded the FAQ to describe deployments generally rather than claim both are identical.
- 30 September 2026: Unpublished by an automated review workflow (independent skeptic review).
- 29 September 2026: First published