I imagine this is a much harder problem than meets the eye. Even for my own travel, there are like a dozen things I'm trying to optimize for. Time, money, seat, travel to/from airport (traffic, train cost, taxi pickup/dropoff ease), lounge quality, layover length or complexity vs cost calculus, loyalty program, airport navigation, airport amenities, flight hospitality quality etc. etc.
When an AI can take into account all these subtleties and work reliably then yeh, I'd give it a go. But for now I'm kinda ~ok with the time cost of finding a non-rubbish flight.
I guess this is where 'travel agents' once used to shine.
Also: I want to be _absolutely_ certain the flight was booked -- that nothing was dropped, that no weird API failed to callback, that no letter was dropped from my name, that no silly email parser thought I meant London, Ontario, Canada.
> I guess this is where 'travel agents' once used to shine.
Score one for self-service software "improving lives" and "increasing productivity" by allowing everyone to do the job that would previously be done much more efficiently by dedicated specialists :).
Still, reading that:
> Even for my own travel, there are like a dozen things I'm trying to optimize for. Time, money, seat, travel to/from airport (traffic, train cost, taxi pickup/dropoff ease), lounge quality, layover length or complexity vs cost calculus, loyalty program, airport navigation, airport amenities, flight hospitality quality etc. etc.
I envy your dedication. Halfway through reading this list, I already feel like going by a train instead. I can't imagine doing it myself without some specialized software; I'd lose my mind trying to do it with the normal travel/accommodation search and booking websites.
> Also: I want to be _absolutely_ certain the flight was booked -- that nothing was dropped, that no weird API failed to callback, that no letter was dropped from my name, that no silly email parser thought I meant London, Ontario, Canada.
Very much this. It's my main concern here, too. I need live feedback on every input at every step throughout the process, because I don't trust software to not screw this up.
I was very to booking a flight to SJO instead of SJC. The interface just said "San Jose" so it looked fine. Might have been nice to go to Costa Rica instead.
> I imagine this is a much harder problem than meets the eye. Even for my own travel, there are like a dozen things I'm trying to optimize for.
But if you created "profiles" of how you might search with various weight settings, it seems like it could get reasonably close without having to re-enter all of the details each time.
Thanks. Definitely more complicate than originally imagined.
A lot of this can be done automatically with time (eg, already built an auto-seat algo), but there is a reason the beta is focused on frequent fliers solving logistics problems and making repeatable decisions :) That said, think you'll be surprised by how well it optimizes already.
During beta onboarding, you make an account with all travel info (no need to parse this). Travelers also confirm their booking before purchase ("checkout" links to a page with all flight info in a lot more detail instead of 'access' link). And BonBook sends a confirmation email post-booking (required by federal regs).
Not to mention cabin luggage allowances (as well as the likelihood they will be honored and you won't be checking your cabin luggage at the gate, or stowing your bag in row 45 when you're in row 9). Even finding the (cabin/checked) luggage allowances on an airline's website can be a task for AI.
'Yes. BonBook is live in private beta and can do a whole lot more than just find flights. Click here to request access.' - why keep it a secret? What else can it do? :)
I am not sure about the US market, but Skyscanner [1] works great in Europe, I wonder how this is different. Besides from being structured as an email, it offers quite similar functionality of finding optimal flights (in some sense of optimal). It may just be me, but I am definitely misunderstanding the inbox metaphor. Why is it an inbox? Why is there emphasis on _one email_? I don't book flights through email -- I go on an airline's website to book.
For now it seems to be always responding with "Hi there, I couldn't find any flights for your request. Please try again.", I could not test it out. Unless this is a demo version, and is working as expected.
I get results everytime. Maybe not the cheapest. But it's something, and it's fast...
Very fast. (Compared to going to a website.)
With a very simple sentence full of acronyms that we all use buy now, it is a fun experimental tool. (Then clicking through the Google link, you can actually continue and book.)
But, I agree, Skyscanner is awesome.
I told it to give me a direct flight and all of its flights have an Atlanta layover. It also ignored my seat pitch request, my checked bags request, and gave me only WestJet flights when Air Canada flights are cheaper.
Is this a frontend to google flights? Sabre integration might get you the info you need to do these queries.
Great idea. Would use (flew 140 segments the year before Covid).
One piece of feedback:
I asked for a "flight from LAX to..." and it recommended flights from SNA and ONT (two other airports in the Greater Los Angeles area). These airports are easily 1.5 - 2 hours away from LAX at commute times.
I suggest that, if the user requests a specific airport (as indicated by an IATA code), you limit responses to that airport code by default, rather than broadening the search to a region.
If I had asked for a "flight from Los Angeles to..." I would not have been surprised by the current behavior (and probably would have appreciated it.)
Would love to be able to have this buy my the flight at an optimal purchase time. Like I say “LAX to JFK for Jan 10th to Jan 15th, between $400 and $600” and it will wait until that is seen and then buy for me. That would be a real new feature.
why would you ever set a minimum price? If you're looking for a certain service level just have it look for that. Otherwise I'd be worried you'd miss out on a flight that fits all your criteria but costs $385 or something
I am one of the early customers and this works like a charm.
Was on a flight last week that was on the tarmac for two hours due to weather one evening last week; the pilot finally got on the PA system and announced we were headed back to the terminal. I booked myself a new flight for the next morning within 90 seconds. Pure magic.
Hi there, I couldn't find any flights for your request. Please try again.
Cross checking with skyscanner there are indeed no flight in the morning I asked, but there are in the afternoon, maybe propose alternatives with nearby airports and/or timeslots when there are no flight meeting requirements?
Wow.I book a lot of flights, often waste a huge amount of time find the best flight, and would use this for a lot of the easier bookings. I was amazed at how well it works (in the simulator).
Google Flights is really the last Google product I routinely use -- happy to have an alternative!
> often waste a huge amount of time find the best flight
Would you trust that this would book you the "best" flight? Every user has a different definition of "best" – cheapest, shortest, fewest stopovers, shortest stopovers, airline preference, lounge preference, seat preference, time of day preference...
Personally I find Google Flights gives me the right amount of detail and I book flights I'm happy with.
I would use it for flights where the best flight is pretty obvious, or in cases where satisficing is fine. I'd tbh mostly use it when OTHER PEOPLE ask me to find a flight for them.
I'd also use it to quickly book flights on my preferred airlines because I have free changes/cancellations on those, so the value of getting something booked immediate and then refined later (and possibly rebook) is fine.
Yeah, I'm kind of surprised by all the uniform praise in the comments - NOT saying it's not a cool tool, it's definitely very impressive and the author's done a nice job, but I could never really see myself using this. Mainly just because flights are super expensive, I can't see myself trusting a bot to make the best choice instead of spending 10 minutes of my own time to potentially save a lot of money. I guess it's a different world when you're a super frequent flier (or when your company is paying, of course).
It reminds me of the Carvana ads that called out home delivery for cars. Like, it's definitely neat, and I'd love to skip scummy dealerships, but convenience is wayy down my list of priorities when I'm dropping that much money on something.
Worth noting: if you don't specify a time of day in your search, you will see the cheapest flight on that day.
Bots can analyze several thousand flight options (which even the sim does) in a few seconds. So odds are it catches a deal someone might otherwise miss.
Also, because BonBook searches direct w/ airlines, it sees inventory not shared with Google (and regularly has better fares).
Interesting, the last one is pretty surprising. I kind of just assumed all airline booking systems (Expedia, Google Flights, Kayak, whatever) all used the same underlying service - maybe SABRE or something, but that's a total shot in the dark. Would love to hear what the difference is, assuming you're comfortable with sharing!
Can share a bit of background: (TLDR; airlines want to keep more profit so are witholding some fares.)
A system called GDS (Global Distribution System) was built in the 60s as a data transmission standard. Was used by airlines to publish fares (among other things). But it was complicated so aggregators (Sabre, Amadeus, Travelport) were born to ingest all this published data and make it easily searchable.
When travel went online in the late 90s, Online Travel Agents (OTAs) were using this data, and since then most new OTAs have too. (An example; Expedia's first patent minimized GDS searches per customer request, saving them $$).
But, because OTAs take a cut and airline margins are thin, IATA (group of airlines) came up with a new system called NDS (New Distrubution Standard) in mid 2010s. Since its an XML standard that anyone can consume direct, airlines have been trying to push everyone onto NDC by simply not publishing certain tickets via GDS, requiring OTAs to book X% of tickets via NDC and even removing inventory entirely.
Some examples: American trying to go full NDC recently but getting pushback and Turkish Airlines pulling out of Sabre.
Is there a reasonable way for me to consume this data directly? I'd love to be able to pull a big pile of flight prices and mess around with the data, but I'd prefer not to pay big money or have to register as a travel agent or anything.
So I've tried a few variations on ticket searches. Pain language great. Two fliers great, can specify times and stops.
Trans continental with connections, works - specifying destinations and round trips
Worked great too.
Local flight in Western Canada, came up with options, but very expensive, and in USD.
I don't use Google for flights - but your tool provides a very efficient way of communicating to google, to get what you want.
One thing I noticed so far, that Google prefers Lufthansa over KLM on search results.
Anyway, great work. Keep us informed!
Pertinent note: BonBook doesn't communicate with/use Google at all. Google link is just there for result comparison.
BonBook has its own internal search+recommendation engine. And private beta let's you actually book the flights in two clicks via the checkout link. That's why you can also change/cancel them with an email.
This is also probably why you are seeing LH > KL.. BonBook doesn't yet have access to KLM inventory.
Feature request: Cryptocurrency available as a payment option. The payment part is a major painpoint in flight reservations. Would be happy to help with integrating if of interest.
For me the demo pulled up flights, but in USD, I'd need it to be localised to be useful.
Airlines also tend to present different prices for different locations/currencies so I doubt this would be as simple as a UI change (though that would still go a long way!).
Yes, the demo doesn't localize results. And Beta doesn't have currency support just yet but it's on the list.
Beyond the price differences, airlines also grant scoped 'ticketing authority' in jurisdictions (eg. can only sell these tickets in EU or US+Canada etc).
I imagine this is a much harder problem than meets the eye. Even for my own travel, there are like a dozen things I'm trying to optimize for. Time, money, seat, travel to/from airport (traffic, train cost, taxi pickup/dropoff ease), lounge quality, layover length or complexity vs cost calculus, loyalty program, airport navigation, airport amenities, flight hospitality quality etc. etc.
When an AI can take into account all these subtleties and work reliably then yeh, I'd give it a go. But for now I'm kinda ~ok with the time cost of finding a non-rubbish flight.
I guess this is where 'travel agents' once used to shine.
Also: I want to be _absolutely_ certain the flight was booked -- that nothing was dropped, that no weird API failed to callback, that no letter was dropped from my name, that no silly email parser thought I meant London, Ontario, Canada.
Ps. Cool demo!
> I guess this is where 'travel agents' once used to shine.
Score one for self-service software "improving lives" and "increasing productivity" by allowing everyone to do the job that would previously be done much more efficiently by dedicated specialists :).
Still, reading that:
> Even for my own travel, there are like a dozen things I'm trying to optimize for. Time, money, seat, travel to/from airport (traffic, train cost, taxi pickup/dropoff ease), lounge quality, layover length or complexity vs cost calculus, loyalty program, airport navigation, airport amenities, flight hospitality quality etc. etc.
I envy your dedication. Halfway through reading this list, I already feel like going by a train instead. I can't imagine doing it myself without some specialized software; I'd lose my mind trying to do it with the normal travel/accommodation search and booking websites.
> Also: I want to be _absolutely_ certain the flight was booked -- that nothing was dropped, that no weird API failed to callback, that no letter was dropped from my name, that no silly email parser thought I meant London, Ontario, Canada.
Very much this. It's my main concern here, too. I need live feedback on every input at every step throughout the process, because I don't trust software to not screw this up.
I was very to booking a flight to SJO instead of SJC. The interface just said "San Jose" so it looked fine. Might have been nice to go to Costa Rica instead.
Checkout link in beta provides details. But you can request both :)
> I imagine this is a much harder problem than meets the eye. Even for my own travel, there are like a dozen things I'm trying to optimize for.
But if you created "profiles" of how you might search with various weight settings, it seems like it could get reasonably close without having to re-enter all of the details each time.
Thanks. Definitely more complicate than originally imagined.
A lot of this can be done automatically with time (eg, already built an auto-seat algo), but there is a reason the beta is focused on frequent fliers solving logistics problems and making repeatable decisions :) That said, think you'll be surprised by how well it optimizes already.
During beta onboarding, you make an account with all travel info (no need to parse this). Travelers also confirm their booking before purchase ("checkout" links to a page with all flight info in a lot more detail instead of 'access' link). And BonBook sends a confirmation email post-booking (required by federal regs).
Not to mention cabin luggage allowances (as well as the likelihood they will be honored and you won't be checking your cabin luggage at the gate, or stowing your bag in row 45 when you're in row 9). Even finding the (cabin/checked) luggage allowances on an airline's website can be a task for AI.
'Yes. BonBook is live in private beta and can do a whole lot more than just find flights. Click here to request access.' - why keep it a secret? What else can it do? :)
I am not sure about the US market, but Skyscanner [1] works great in Europe, I wonder how this is different. Besides from being structured as an email, it offers quite similar functionality of finding optimal flights (in some sense of optimal). It may just be me, but I am definitely misunderstanding the inbox metaphor. Why is it an inbox? Why is there emphasis on _one email_? I don't book flights through email -- I go on an airline's website to book.
For now it seems to be always responding with "Hi there, I couldn't find any flights for your request. Please try again.", I could not test it out. Unless this is a demo version, and is working as expected.
[1] https://www.skyscanner.net/
I get results everytime. Maybe not the cheapest. But it's something, and it's fast... Very fast. (Compared to going to a website.) With a very simple sentence full of acronyms that we all use buy now, it is a fun experimental tool. (Then clicking through the Google link, you can actually continue and book.) But, I agree, Skyscanner is awesome.
Beta turns booking into two clicks.
The beta is email-based booking, changes and more. The demo simulates the search experience via email.
I told it to give me a direct flight and all of its flights have an Atlanta layover. It also ignored my seat pitch request, my checked bags request, and gave me only WestJet flights when Air Canada flights are cheaper.
Is this a frontend to google flights? Sabre integration might get you the info you need to do these queries.
The sim has limited ability to make it lightweight, free from signup and responsive in ~15 seconds.
Beta is much more capable.
GFlights is not used, link is there for comparison.
It would be great if i could chat with chatgpt and it could book it for me, like a travel agent
Great idea. Would use (flew 140 segments the year before Covid).
One piece of feedback:
I asked for a "flight from LAX to..." and it recommended flights from SNA and ONT (two other airports in the Greater Los Angeles area). These airports are easily 1.5 - 2 hours away from LAX at commute times.
I suggest that, if the user requests a specific airport (as indicated by an IATA code), you limit responses to that airport code by default, rather than broadening the search to a region.
If I had asked for a "flight from Los Angeles to..." I would not have been surprised by the current behavior (and probably would have appreciated it.)
Solid feedback.
To keep the demo light, it doesn't have full functionality but being specific on airports vs cities is part of the live beta.
NYC is another example w/ EWR. Can say "NYC" or specify "JFK or LaGuardia". DC too for people wanting to avoid the drive w/ BWI.
Happy to share access. Feel free to shoot me an email (attila@).
Cool, thanks - I sent you an email
Would love to be able to have this buy my the flight at an optimal purchase time. Like I say “LAX to JFK for Jan 10th to Jan 15th, between $400 and $600” and it will wait until that is seen and then buy for me. That would be a real new feature.
why would you ever set a minimum price? If you're looking for a certain service level just have it look for that. Otherwise I'd be worried you'd miss out on a flight that fits all your criteria but costs $385 or something
You can set up alerts using Google flights that will email you when prices change
I save thousands using this feature each year
I like that feature! But now I’m wondering if Google is selling that data back to airlines.
Calling it 'scheduled booking'. It's on the roadmap :)
I am one of the early customers and this works like a charm.
Was on a flight last week that was on the tarmac for two hours due to weather one evening last week; the pilot finally got on the PA system and announced we were headed back to the terminal. I booked myself a new flight for the next morning within 90 seconds. Pure magic.
Happy BonBook could help. Thanks for the glowing recommendation
Hi there, I couldn't find any flights for your request. Please try again.
Cross checking with skyscanner there are indeed no flight in the morning I asked, but there are in the afternoon, maybe propose alternatives with nearby airports and/or timeslots when there are no flight meeting requirements?
Getting the same results, here is my query: "Find me a flight for the 10th, 11th or the 12th from BKK to London, ideally heathrow."
Can't handle multi-date (yet).
Yes, this is a bit more complicated but on the roadmap. Great suggestion.
Hi HN: Looks like LLM API calls started being rejected. Working on a solution...
UPDATE: We're back!
Seems like a neat idea but I just tried 4 times and got the same response every time:
> Hi there, I couldn't find any flights for your request. Please try again.
> -BonBook
I was trying input like "Find me a flight from BWI to PWM December 27 to January 1".
We had a bug with one of our external provides earier this morning, but its now resolved.
Wow.I book a lot of flights, often waste a huge amount of time find the best flight, and would use this for a lot of the easier bookings. I was amazed at how well it works (in the simulator).
Google Flights is really the last Google product I routinely use -- happy to have an alternative!
> often waste a huge amount of time find the best flight
Would you trust that this would book you the "best" flight? Every user has a different definition of "best" – cheapest, shortest, fewest stopovers, shortest stopovers, airline preference, lounge preference, seat preference, time of day preference...
Personally I find Google Flights gives me the right amount of detail and I book flights I'm happy with.
I would use it for flights where the best flight is pretty obvious, or in cases where satisficing is fine. I'd tbh mostly use it when OTHER PEOPLE ask me to find a flight for them.
I'd also use it to quickly book flights on my preferred airlines because I have free changes/cancellations on those, so the value of getting something booked immediate and then refined later (and possibly rebook) is fine.
Yeah, I'm kind of surprised by all the uniform praise in the comments - NOT saying it's not a cool tool, it's definitely very impressive and the author's done a nice job, but I could never really see myself using this. Mainly just because flights are super expensive, I can't see myself trusting a bot to make the best choice instead of spending 10 minutes of my own time to potentially save a lot of money. I guess it's a different world when you're a super frequent flier (or when your company is paying, of course).
It reminds me of the Carvana ads that called out home delivery for cars. Like, it's definitely neat, and I'd love to skip scummy dealerships, but convenience is wayy down my list of priorities when I'm dropping that much money on something.
Worth noting: if you don't specify a time of day in your search, you will see the cheapest flight on that day.
Bots can analyze several thousand flight options (which even the sim does) in a few seconds. So odds are it catches a deal someone might otherwise miss.
Also, because BonBook searches direct w/ airlines, it sees inventory not shared with Google (and regularly has better fares).
Interesting, the last one is pretty surprising. I kind of just assumed all airline booking systems (Expedia, Google Flights, Kayak, whatever) all used the same underlying service - maybe SABRE or something, but that's a total shot in the dark. Would love to hear what the difference is, assuming you're comfortable with sharing!
Can share a bit of background: (TLDR; airlines want to keep more profit so are witholding some fares.)
A system called GDS (Global Distribution System) was built in the 60s as a data transmission standard. Was used by airlines to publish fares (among other things). But it was complicated so aggregators (Sabre, Amadeus, Travelport) were born to ingest all this published data and make it easily searchable.
When travel went online in the late 90s, Online Travel Agents (OTAs) were using this data, and since then most new OTAs have too. (An example; Expedia's first patent minimized GDS searches per customer request, saving them $$).
But, because OTAs take a cut and airline margins are thin, IATA (group of airlines) came up with a new system called NDS (New Distrubution Standard) in mid 2010s. Since its an XML standard that anyone can consume direct, airlines have been trying to push everyone onto NDC by simply not publishing certain tickets via GDS, requiring OTAs to book X% of tickets via NDC and even removing inventory entirely.
Some examples: American trying to go full NDC recently but getting pushback and Turkish Airlines pulling out of Sabre.
Is there a reasonable way for me to consume this data directly? I'd love to be able to pull a big pile of flight prices and mess around with the data, but I'd prefer not to pay big money or have to register as a travel agent or anything.
Depends on the detail. There are a few simple ones (eg flightlab).
Super cool stuff. Thanks for the info.
Fair, beta isn't for everyone. It's geared towards frequent fliers and people who want to save time. But lets break down the points:
-> cheapest?? Frequent fliers are primarily concerned with time efficiency.
-> shortest/fewest-stopovers/shortest-stops?? These are all synonyms for efficiency.
-> airline preference?? Beta curates based on airline loyalty.
-> lounge preference?? Not handled when booking flights..
-> seat preference?? Furthest forward aisle/window, avoid no-recline, some want extra legroom (beta has an algo that handles this).
-> time of day?? Easy, just add it to your request (works in sim too).
Happy to give you one. Biased but the live beta is better.
> No cheap carriers by design Why no love for the frugal ?
Different mindset.
BonBook is targeted at travelers solving logistics problems who make lots of repeatable decisions and fly often (eg. need to be in X at Y).
No saying what can be done with time.
While that's a good point, I think it would be wise for them to incorporate budget airlines if they are looking at the European market.
There are so many routes that are served exclusively by budget airlines.
Noted re the EU. Though there's another side to it as well.
BonBook does more than just booking (eg. one-email changes) and many budget airlines make this difficult with their restrictive fare rules.
They’re probably not distributed by whatever GDS API the app is using
Beta (and sim) intentionally does not search them.
So I've tried a few variations on ticket searches. Pain language great. Two fliers great, can specify times and stops. Trans continental with connections, works - specifying destinations and round trips Worked great too. Local flight in Western Canada, came up with options, but very expensive, and in USD. I don't use Google for flights - but your tool provides a very efficient way of communicating to google, to get what you want. One thing I noticed so far, that Google prefers Lufthansa over KLM on search results. Anyway, great work. Keep us informed!
Pertinent note: BonBook doesn't communicate with/use Google at all. Google link is just there for result comparison.
BonBook has its own internal search+recommendation engine. And private beta let's you actually book the flights in two clicks via the checkout link. That's why you can also change/cancel them with an email.
This is also probably why you are seeing LH > KL.. BonBook doesn't yet have access to KLM inventory.
And haven't added non-USD currency support yet.
Sounds super useful!
Feature request: Cryptocurrency available as a payment option. The payment part is a major painpoint in flight reservations. Would be happy to help with integrating if of interest.
Yes, its on the roadmap. Zooming out, crypto gets super interesting in terms of industry dynamics.
If you shoot a note to the contact email, will take a look.
Is this US only? Nothing is working for me in Australia
The email sim doesn't allow booking but should still work to demo functionality. And it's not geo-limited. What are you seeing?
For me the demo pulled up flights, but in USD, I'd need it to be localised to be useful.
Airlines also tend to present different prices for different locations/currencies so I doubt this would be as simple as a UI change (though that would still go a long way!).
Yes, the demo doesn't localize results. And Beta doesn't have currency support just yet but it's on the list.
Beyond the price differences, airlines also grant scoped 'ticketing authority' in jurisdictions (eg. can only sell these tickets in EU or US+Canada etc).
How is this enforced? Purely IP-based? Sounds like we now need a VPN when buying tickets.
>Sounds like we now need a VPN when buying tickets.
Frame 6: You guys are buying tickets without a vpn?
I personally see "Hi there, it looks like the internet dropped the ball somewhere. Please try your request again.".
Yes, the internet did in fact drop the ball. Fix is live.
Gmail sent me back a failed mail response:
''' Address not found Your message wasn't delivered to test@bonbook.co because the address couldn't be found, or is unable to receive mail. '''
Yes, test@bonbook.co is not valid for booking. And private beta requires onboarding.
The inbox simulator provides high-level functionality of the beta without access.
Really cool and useful, how long have you worked on this?
Thanks. Built this sim in a few days. Iterating the beta since summer.
This is nice. How do we choose currency?
Email sim doesn't have this (can't actually buy tickets).
And haven't built out currency support for the beta yet but its on the list.
Great, thanks
insanely easy, make it on SMS?
it's on the roadmap!
Wow, this is a brilliant idea! It sounds perfect for fewuent travellers.
Thanks for the kind words