The numbersBriefEpisode mapTakeawaysChartsQuotesClipsGlossaryGo nextTranscript
Episode 132 · Enterprise · 58 min

The inventory that dies at the gate

Sabre has been travel's invisible plumbing for sixty years, with close to 2,000 US agencies hanging off a single mainframe before the internet existed, and has just moved 40,000 servers out of 19 of its own data centres into Google Cloud without an agency noticing. The claim from its India head: travel is a harder retail problem than e-commerce because a seat is worth nothing the second the plane leaves the gate, and the only fix is to turn a trip into one order instead of nine bookings.

SaaS & enterpriseData & digitisationAI & machine learning
S
Sriram Gopalswamy
Managing Director, Sabre India · with Vishal Krishna
The inventory that dies at the gate · episode thumbnail
58:12
Said in this episode
▶ 20:17
19
Data centres shut down
Owned and managed data centres closed across the globe in about four years as Sabre exited its own estate for public cloud.
▶ 20:33
40,000
Servers moved to Google Cloud
Along with roughly 400,000 CPUs and 50 petabytes of data; a further instance count quoted on air is garbled in the captions and is left out here.
▶ 19:47
90-95%
Sabre compute now on public cloud
All of it on GCP, with a mainframe remnant still carrying airline accounting, billing and reconciliation integrations.
▶ 8:57
70-80
Airlines that have adopted NDC
Out of roughly 450 to 500 airlines worldwide. Adoption slowed because carriers are still working through post-Covid operational recovery.
▶ 38:25
90%
Hotel customer care handled by AI
The SynXis hospitality AI fields around nine in ten customer-care interactions, per the guest, without a human specialist pulling documents.
▶ 3:02
2,000+
Agencies on one mainframe, pre-internet
The basis of Sabre's claim to have been 'the worldwide web before the worldwide web existed': US agencies connected to a single mainframe roughly 60 years ago.
The brief

The argument in sixty seconds

Sriram's claim is that travel is the hardest retailing problem in technology, and that the difficulty is structural rather than cultural. An unsold t-shirt can be discounted or given away with a pair of joggers; a kirana can break a 2 kg pack of cornflakes into 20 g sachets. But once the plane pushes back from the gate, the seat is worth nothing forever, which is why a global distribution system exists at all, holding airline inventory and pushing it out to consumer apps, modern OTAs, the street-corner agency and the airline's own site, while quietly filing passenger data with the TSA on the way. Sabre, he says, was the worldwide web before the worldwide web: close to 2,000 agencies on one mainframe, pre-internet. His argument for why the plumbing is being rebuilt now is that the customer who gets groceries in ten minutes will not forgive a booking flow assembled from EDIFACT payloads, so the industry is migrating to NDC, where a ticket becomes an order with rich content attached and the airline finally knows who is sitting in 14A. Only 70 to 80 of roughly 450 to 500 airlines have made the move. Sabre's own bet was infrastructural: 19 data centres shut in four years, 40,000 servers, 400,000 CPUs and 50 petabytes moved to Google Cloud with no customer disruption, which he likens to repairing a flat tyre while the car is still moving. The payoff is an AI layer per stakeholder: fare experimentation, network planning, a hospitality bot already handling 90% of customer care. The stakes: when a trip becomes one order, the parking slot on IPL night, the two-hour hotel room and the carbon number all become sellable, and a sixty-year-old company gets to be the retail engine underneath everyone else.

Worth your time if you are

Engineers who assume legacy systems mean boring work
Product leads unbundling a booking into sellable parts
Travel and hospitality operators weighing up NDC
Anyone running a mainframe-to-cloud migration in production
Students choosing a first job beyond the AI hype
Episode map

Where the conversation travels

Every block is a chapter, coloured by what it's about. Click any of it to jump straight to that minute on YouTube.

010:00–3:20 · Cold open: the Phoenix and the unicorns023:20–7:23 · What a GDS actually does037:23–11:25 · EDIFACT out, NDC in0411:25–14:13 · One order from cab to hotel0514:13–16:43 · The seat dies at the gate0616:43–22:20 · Nineteen data centres into Google Cloud0722:20–25:00 · Experimentation, and where the pings land0825:00–27:20 · A booking fee, never a licence0927:20–30:10 · What India's engineers actually own1030:10–33:40 · Personalised offers and the carbon question1133:40–37:10 · Is a sale success, or experience?1237:10–40:20 · Generative AI without the shiny penny1340:20–44:05 · Hyper Jump, and code built to last1444:05–48:00 · Travel is inherent; 250 new airports1548:00–51:15 · Hospitality and the unlimited SKU1651:15–58:12 · Detach, and the board-game interview0:0010:0020:0030:0040:0050:0058:12
SaaS & enterpriseData & digitisationAI & machine learningD2C & commerceGCCs & servicesHiring & talent
01Cold open: the Phoenix and the unicorns 0:00 Vishal frames the confluence of retail, tech and travel, and the guest supplies the line the episode is built on, Sabre as the Phoenix ready to take on the unicorns, before sketching a sixty-year-old system that had close to 2,000 US agencies on one mainframe before the internet existed. 02What a GDS actually does 3:20 GDS defined: Sabre holds airline inventory and distributes it to every channel (consumer apps, modern OTAs, the street-corner travel agent, even the airline's own website) while shipping passenger information to bodies like the TSA and hiding the complexity the way a car hides its fuel injection. 03EDIFACT out, NDC in 7:23 The old EDIFACT protocol shatters a booking into payloads across separate inventory, schedule and fare systems; NDC replaces it with richer content and a direct customer relationship, and only 70 to 80 of the world's roughly 450 to 500 airlines have adopted it so far. 04One order from cab to hotel 11:25 The end-state journey, from the cab to Kempegowda through Digi Yatra and facial recognition, the flight, the hotel, even flowers waiting in the room, all on one order, set against today's reality of separate bookings and a seat map that says 14A and tells you nothing else. 05The seat dies at the gate 14:13 Why travel's supply chain is harder than e-commerce: an unsold t-shirt can be bundled and a 2 kg cornflakes pack broken into sachets, but the seat and the hotel night vanish at departure, which puts operations research and demand prediction at the centre of what Sabre sells. 06Nineteen data centres into Google Cloud 16:43 Twenty years into India, part of the estate still runs on mainframe, but the customer who gets groceries in ten minutes set the pace: 19 data centres shut in four years, 40,000 servers, 400,000 CPUs and 50 petabytes moved to Google Cloud, 90-95% of compute now public cloud, with no service disruption. 07Experimentation, and where the pings land 22:20 AI arrives layer by layer: self-healing infrastructure as table stakes, then an experimentation capability that lets an airline test a $100 fare against a competitor's $150 using years of anonymised booking data. Every OTA, hotel and hosted carrier ping lands on the same rails. 08A booking fee, never a licence 25:00 Airport retailers could join the marketplace once orders replace tickets; meanwhile the model has always been a booking fee from airlines and commissions paid back to agencies rather than software licences, and Sabre says it is the second-largest consumer of Google Cloud training after Google itself. 09What India's engineers actually own 27:20 India's teams own entire verticals (airlines, OTAs and agencies, hospitality, corporate travel) plus the reliability and availability work, because an agency whose system is down for fifteen minutes loses the booking to the shop next door. 10Personalised offers and the carbon question 30:10 Anonymised data lets an airline price a slow Tuesday differently and tailor an offer to a family of four, while the newer traveller demand is sustainability: the carbon footprint of a route, and the three untouched 100-gram soap bars in every hotel bathroom. 11Is a sale success, or experience? 33:40 Today success is honestly a sale, he admits, but the target is experience and the loyalty that follows, including a proposal that frequent flyers be walked through the safety briefing properly once, certified, and never made to sit through it again. 12Generative AI without the shiny penny 37:10 More than a thousand developers now write code with generative AI inside a contained, compliance-checked environment; the SynXis hospitality AI already handles around 90% of customer care, with network planning and ancillary bundling next in line. 13Hyper Jump, and code built to last 40:20 Campus hiring runs through Hyper Jump, a paid six-month programme that turns graduates into travel-tech specialists, with Sabre co-designing curricula and warning students off the shiny penny, while the stack stays deliberately narrow, mostly Java with C++ and Go, written to survive fifteen or twenty years. 14Travel is inherent; 250 new airports 44:05 Travel is a millennium-old human need that will normalise whatever happens to globalisation, and India is building for it: a claimed 250 new airports within four to five years, plus highways and railways compressing Delhi to Jaipur to two hours, while twenty-odd big OTAs do the distributing and every app now wants to retail travel. 15Hospitality and the unlimited SKU 48:00 Hospitality is the fastest-growing line, with Sabre powering the world's largest hotel chain and running Taj Hotels end to end, property management as well as distribution. The frontier is the unlimited SKU: the spa, the bathrobe, a two-hour room, or 400 parking slots on an IPL night at the Chinnaswamy. 16Detach, and the board-game interview 51:15 The personal close: detachment as a leadership discipline against 800 billboards a day, a reading list that takes in Zen and the Art of Motorcycle Maintenance, Blue Ocean Strategy, The Long Tail and Tuesdays with Morrie, and board games, Mastermind to Catan, used as a hiring round because cooperative play reveals what a self-description cannot.
Takeaways

Ideas to carry out of this hour

01

Travel's inventory is the most perishable retail there is

The reason travel distribution looks over-engineered next to e-commerce is that the goods evaporate. A t-shirt that isn't moving can be discounted, bundled with joggers or given away buy-one-get-one; a kirana can turn a 2 kg cornflakes pack into a 20 g sachet and keep selling. Once the aircraft leaves the gate, or the night passes at the hotel, that unit of inventory is gone and unrecoverable. Everything else follows from that single asymmetry: the operations research, the demand prediction, the willingness to reprice a Tuesday.

▶ 14:29▶ 14:58▶ 15:29
02

The customer forcing this change is a grocery app

Asked who is pushing airlines, hotels and agencies to modernise, he does not say a competitor or a regulator. He says the end traveller. Someone who receives groceries in seven or ten minutes experiences a travel booking as a different century, and that gap is the product opportunity. Covid then made digital transformation an executive priority rather than an IT one, which is what finally moved a famously conservative industry.

▶ 17:58▶ 18:14
03

NDC turns a ticket into an order, slowly

Under EDIFACT, a booking is decomposed into payloads that hit separate systems for inventory, schedules and fares, and the airline you fly may not know your name, age or phone number because an OTA sold the seat. NDC replaces that with an order and a direct relationship, which is what makes personalisation technically possible: extra baggage for a family of four, a vegetarian meal, a rebooking message when the Dubai leg slips four hours. Adoption rather than ambition is the constraint: 70 to 80 airlines out of roughly 450 to 500 have moved, and post-Covid operational recovery is eating the runway.

▶ 8:41▶ 8:57▶ 9:12
04

You can re-plumb a global system without customers noticing

Sabre shut 19 owned-and-managed data centres in about four years and moved 40,000 servers, roughly 400,000 CPUs and 50 petabytes into Google Cloud, taking compute to 90-95% public cloud while a mainframe remnant still carries accounting, billing and reconciliation integrations. The stated customer impact was none. His image is a car whose tyre goes flat and gets repaired while the car keeps moving. He credits the skilling programme as much as the architecture, and claims Sabre is the second-largest consumer of Google Cloud training after Google.

▶ 20:17▶ 21:46▶ 26:20
05

AI is a layer per stakeholder, not a demo

The interesting AI in this business is not the chatbot on the front page. It is fare experimentation, where an airline runs models against years of anonymised booking history to decide whether a $100 seat should be $115 on a slow Tuesday or cut when a competitor adds capacity; it is network planning that picks the departure time maximising onward connections; it is a hospitality assistant already handling about 90% of customer care. Over a thousand developers use generative AI to write code, inside a contained environment built for the industry's compliance load.

▶ 23:03▶ 38:25▶ 38:54
06

Everything around the seat is an SKU nobody is selling

Book a flight and you are shown 14A, 14B, 15C and a price. Not legroom, not the seat shape, not the entertainment, while Amazon will tell you which garden your coffee beans came from. Book a hotel and you get a room, with the spa, the bar, the Wi-Fi and the bathrobe stranded in silos. His sharpest example is a hotel beside the Chinnaswamy Stadium on an IPL night with 400 parking slots it can only sell to room guests. Sell the parking from 2pm to 11pm and the restaurant and bar revenue follows. The unlimited-SKU model, including a two-hour room, is what an order-based world unlocks.

▶ 13:27▶ 39:25▶ 50:15
07

India's centre owns availability, which is the hardest promise

The India organisation is not staffed as support: it runs dedicated teams per vertical (airlines, OTAs and agencies, hospitality, corporate travel) and owns much of the reliability and availability work. The stakes are mundane and absolute: an agent who says 'my system is down, call back in fifteen minutes' loses the customer to the next agency, and a hotel with rooms ready but systems down leaves a tired traveller in the lobby. Post-migration, the work has shifted from moving to optimising cost, speed and security.

▶ 28:07▶ 29:20▶ 29:06
08

Hire by watching people play, not by listening to them

Sabre runs a gaming round in campus and even senior hiring: at the final stage, candidates play a board game rather than answer another question about teamwork. Anyone can claim to be competitive and a team player; cooperative games, where partners are formed dynamically and you may not know who is on your side until the end, expose the truth. He treats it as the recruiting analogue of sport: the competitive spirit you cannot fake, plus a read on how someone thinks under a rule set.

▶ 56:14▶ 56:30▶ 56:47
The numbers, drawn

What the episode measures

Every figure below was said on air, with timestamps included and caveats kept.

A ticket, or an order

what changes when EDIFACT gives way to NDC
EDIFACT: one request, shattered Booking Inventory Schedule Fares The airline may not know your name, age or number, because an OTA sold the seat. NDC: one order, richer content One order, and a direct relationship Extra baggage for a family of four, a vegetarian meal, a rebooking when Dubai slips four hours. 70 to 80 of roughly 450 to 500 airlines have moved
Midpoints of the two ranges he gives are used for the bar. He puts the slow adoption down to carriers still working through post-Covid operational recovery rather than any argument about the standard.▶ 8:57

The seat dies at the gate

what you can still do with stock that has not sold
A t-shirt Discount it, bundle it with joggers, buy one get one free 2 kg cornflakes Break the pack into 20 g sachets and keep selling A seat, a room Once the plane pushes back it is worth nothing, forever So operations research sits at the centre.
His answer to why travel distribution looks over-engineered next to e-commerce. The kirana can keep breaking the pack down; the aircraft cannot. Demand prediction and the willingness to reprice a slow Tuesday both follow from that one asymmetry.▶ 14:29

Nineteen data centres, moved while running

the migration, in the numbers he gives
19 data centres 40,000 servers moved 400,000 CPUs 50 PB of data 4 years start to finish 90-95% now public cloud Customer disruption: none A tyre going flat and being repaired while the car keeps moving.
All of it onto Google Cloud, with a mainframe remnant still carrying airline accounting, billing and reconciliation integrations. The zero-disruption claim is the company's own. He credits the skilling programme as much as the architecture, and says Sabre is the second-largest consumer of Google Cloud training after Google.▶ 20:17

Everything around the seat that nobody sells

the SKUs an order-based world would unlock
What you are shown today 14A, 14B, 15C, and a price Not legroom, seat shape or entertainment. What is sitting unsold Legroom In-flight entertainment The spa and the bar The bathrobe A two-hour room The Wi-Fi 400 parking slots on an IPL night
The hotel beside the Chinnaswamy is his sharpest example: today those slots can only be sold to room guests. Sell the parking from 2pm to 11pm, he argues, and the restaurant and bar revenue follows it in.▶ 50:15
Worth keeping

Lines that stay

Everybody wants to work with the unicorns. I would say we are the Phoenix, ready to take on the unicorns.

▶ 1:14

We were the worldwide web before the worldwide web existed — close to 2,000 agencies across the US connected to a single mainframe, before the internet.

▶ 3:02

Once the plane takes off from the gate, that's it — I cannot sell that ticket any more. Or that night, nobody's staying. I lose the inventory.

▶ 15:29

It's like your car tyre goes flat this morning, and you repair it while the car is still moving. That's how complex this is.

▶ 21:46

Everything is not about the shiny pennies. Sometimes we lose the forest thinking of a tree.

▶ 41:24
Clips that travel

Short on time? Start here

Engineers who have never seen travel's plumbing

What a GDS actually is

The clearest definition in the episode: who holds the inventory, why the airline's own site is also powered by a GDS, and what moves in the background before you fly.

3:49 → 7:23 · 4 min ▶ Watch clip
Travel and hospitality operators weighing up NDC

EDIFACT, NDC and the airline that knows you

The protocol shift explained through a delayed Dubai connection: why today the carrier doesn't have your number, and what changes when a ticket becomes an order.

7:23 → 11:25 · 4 min ▶ Watch clip
Product and pricing leads in perishable-inventory businesses

Why the seat dies at the gate

The t-shirt, the kirana's cornflakes sachet and the aircraft that takes off. The cleanest articulation of why travel pricing is its own discipline.

14:13 → 16:43 · 2 min ▶ Watch clip
Anyone running a mainframe-to-cloud migration in production

Nineteen data centres, moved while running

The migration numbers and the flat-tyre-while-driving standard for customer disruption, plus what stayed on mainframe and why.

19:30 → 22:20 · 3 min ▶ Watch clip
Hiring managers tired of teamwork answers

The board-game interview

Cooperative games as a hiring round, used up to senior levels, and the argument that play reveals what a self-description cannot.

54:53 → 57:40 · 3 min ▶ Watch clip
Glossary

The jargon, unpacked

GDS (global distribution system)
The system that holds and distributes airline and hotel inventory to every selling channel: consumer apps, OTAs, high-street agents and the airline's own website. Sabre was among the first.
EDIFACT
The decades-old messaging protocol still used by many airlines and agencies, which breaks a booking request into smaller commands sent to separate inventory, schedule and fare systems before the answers are reassembled.
NDC (new distribution capability)
The replacement standard that carries far richer content and lets the airline hold the customer relationship, turning a ticket into an order; 70 to 80 airlines had adopted it at the time of recording.
PNR
The passenger name record, the single booking reference that today ties a trip together, and which the order-based NDC world is designed to supersede.
OTA (online travel agency)
The online booking platforms that resell inventory from many suppliers so a traveller can compare routes and fares; India has roughly twenty to twenty-five large ones, per the guest.
Ancillaries
Everything sold around the seat or the room: legroom, meals, Wi-Fi, in-flight entertainment, the spa, the bathrobe, even parking, all of which the guest wants bundled into one personalised offer.
Property management system
The hotel's operational backbone covering check-in, housekeeping, utilities, accounting and billing, distinct from the reservation engine that distributes rooms, and something Sabre supplies as well.
Connections

If this resonated, go here next

EP 135 · Enterprise

The mainframe never gets switched off

Chander Damodaran (Brillio)
Data & digitisationAI & machine learning
Deep dive
EP 139 · Enterprise

The weakest link has a login

Ben Goodman (Okta)
SaaS & enterpriseData & digitisation
Deep dive
EP 124 · Enterprise

Cost is the part you can control

Srividya Kannan (Avaali)
Data & digitisationSaaS & enterprise
Deep dive
Full transcript

The whole conversation, searchable

229 segments

Auto-generated captions, lightly cleaned. Click a timestamp to open that moment on YouTube.