Delta Air Lines

Delta didn’t start scaling AI by buying tools.

They spent years modernising their backend first because legacy companies have decades of data that needs to get “AI-ready” before they implement AI tools and applications.

Starting in 2021, Delta partnered with IBM to migrate its environment onto a hybrid cloud built on Red Hat OpenShift, with AWS as its primary cloud provider.

The migration touched 1,300 applications and modernised 32 core systems responsible for roughly 90% of Delta’s IT costs. In 2023, Delta extended a five-year infrastructure partnership with Kyndryl to run core operations like crew rostering, scheduling, and the rebooking platform for cancelled flights.

We’ve seen a similar pattern in our previous case studies of AT&T, Bank of America, and Intel, where decades of accumulated network data only became a moat once the infrastructure caught up.

Without this groundwork, none of what follows would work.

Delta TechOps built APEX (Advanced Predictive Engine) to solve a problem every airline faces: engines fail without warning, and a single unscheduled removal can ground a plane for days.

APEX collects sensor data from nearly 900 aircraft throughout an engine’s life in real time. Machine learning models scan for patterns that precede failure, flagging specific components, “replace this part within 50 flight hours”, well before a breakdown happens.

Predictive material-demand accuracy rose from 60% to over 90%. Engines that once took 150 to 200 days to overhaul at outside vendors now take Delta TechOps under 90 days in-house. Followed by Eight-figure annual savings.

Delta extended the model outward too. The company formed the Aviation Digital Alliance with Airbus Skywise and later GE Digital, pooling fleet analytics among partners rather than solving the problem alone.

In 2024, Delta began testing AI-driven dynamic pricing built by Fetcherr, an Israeli startup whose large market model processes aggregated route-level demand to recommend fares.

By mid-2025, the system covered about 3% of domestic fares, up from roughly 1% in late 2024. Delta’s target is 20% of the network by the end of the year.

At Delta’s November 2024 investor day, Hauenstein called the system a super analyst working around the clock, one that improves as Delta feeds it more data and outcomes to learn from.

However, the slow rollout invited scrutiny. Lawmakers questioned whether AI pricing could mean individualised fares based on personal data. However, the company has been vocal about the new models and technologies using aggregated data and not personal.

The takeaway for enterprise leaders like you has little to do with pricing algorithms specifically.

Disciplined, gradual rollouts, paired with a willingness to explain the system publicly, build more trust than a fast launch ever could.

Which is kind of great, if you think about it, because today almost all brands sound the same, so a little transparency goes a long way.

At Hartsfield-Jackson Atlanta, the world’s busiest airport by passenger traffic, roughly 250 ramp agents move bags between aircraft every day.

Delta built Baggage AI in-house to solve a coordination problem at scale.

It works like a ride-sharing app. The model matches drivers to bags using real-time flight data, bag locations, and the time a passenger has to make their connection to the next flight.

The system re-optimises routes every two minutes, adjusting for gate changes and tight connections as they happen. Delta says the tool has improved baggage transfer success rates by as much as 20% at Atlanta, meaning thousands more bags make their connecting flights on time.

In 2026, Delta plans to expand Baggage AI to Detroit and Minneapolis-Saint Paul, add autonomous dispatching based on drivers’ GPS positions, and introduce bag scanning at transfer points to improve tracking.

It’s a constraint similar to the one that pushed UPS to build Dynamic ORION for logistics. The real world changes every two minutes, so a static plan made hours earlier stops being useful.

APEX, Fetcherr, and Baggage AI carry the strongest proof so far. But Delta is testing AI across customer experience and the workforce as well.

Delta Concierge, a generative AI assistant inside the Fly Delta app, is the airline’s biggest customer-facing bet. It’s still in limited beta.

Some impactful AI tools and use cases within Delta:

Delta Concierge’s roadmap is the one to watch. Planned features include passport and visa alerts before international trips, real-time airport wayfinding, and automatic rebooking with hotel or meal vouchers during disruptions. Delta has also said Concierge will eventually connect to Uber for ground transport and Joby Aviation’s electric air taxis, stitching together a single AI-guided itinerary from booking to landing.

On pricing, Delta’s stated target is 20% of its global network on Fetcherr's system by the end of 2025, up from roughly 3% mid-year. Expect Delta to keep talking about pace and transparency as much as the technology itself, given the scrutiny the rollout has already drawn.

Baggage AI’s expansion to Detroit and Minneapolis-Saint Paul will be the first real test of whether Atlanta’s results travel to hubs with different layouts and volumes.

Fix the foundation before the feature. Delta’s cloud migration and its partnership with Kyndryl weren’t AI projects. They were the prerequisites that made AI projects possible. Without them, APEX and Fetcherr would have nothing clean to run on.

Slow is a feature in the right applications. Take our time approach with Fetcherr protected Delta from the kind of rushed rollout that invites bigger problems than the one it was meant to solve.

Let one program prove the model before scaling it. APEX took years to mature inside Delta TechOps before the airline applied the same predictive logic to baggage and pricing. Proof in one domain built the case for the next.

Sources