As presented at the Amazon Appstore Summit, London in November 2015. Skyscanner walks through its recent history, how buying travel and buying books are very different (but users don't care), how the principles of our products are applied to voice search and how we built a basic Alexa flight search.
4. Skyscanner launched
Expands across Europe with foreign site launches
First 100k visits in a day
Singapore business established
to support APAC growth
First app launched
Beijing office opened
Miami business established to
support Americas growth
Skyscanner moves from flights to travel with car hire
launch and hotel technology acquisition
Sequoia secondary investment values Skyscanner at $800million
2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Launched apps for
hotels and car hire
Acquired Youbibi and opened office in Shenzen
Acquired Distinction and opened office in Budapest
Sofia office opened
Joint venture with
Yahoo! JAPAN
launched
Skyscanner
powers MSN
travel worldwide
2016
Larger Budapest
office and
London office
open
5. Truly global flight
coverage
80% visitors from
outside the
UK market
45 countries
with over 50,000
visitors per month
Edinburgh, Glasgow & London
Miami
Singapore
BeijingBarcelona
Budapest
Sofia
Shenzhen
6.
7. Our ultimate goal:
to make travel booking as easy as it is
today to buy a book online
26. Ask Skyscanner for a flight
Sure, where are you flying from?
London
Great, where are you flying to?
Seattle
Thanks, when are you leaving?
Friday
And finally, when do you return?
November 13th
Great, finding prices…
27. Ask Skyscanner for a flight
Sure, where are you flying from?
Ask Skyscanner for a flight please
Sorry, I don’t understand…
29. SearchType
Origin
Destination
Start
End
Ask Skyscanner for a flight
Sure, where are you flying from?
Great, where are you flying to?
Thanks, when are you leaving?
And finally, when do you return?
Great, finding prices…
flights
london
seattle
2015-11-03
2015-11-17
SessionAttributes
32. Edinburgh
Quartermile One
15 Lauriston Place
Edinburgh
EH3 9EN
Glasgow
5th floor,
151-155 St Vincent
St, Glasgow
G2 5NW
Singapore
No. 08-01&04 & 09-
04
8th floor,
Robinson Point,
39 Robinson Rd,
Singapore
Beijing
Level 19, Tower E2,
Oriental Plaza,
No. 1 East Chang An
Avenue,
Dong Cheng District,
Beijing 100738
Miami
1395 Brickell Ave,
Suite 900,
Miami,
Florida 33131
Barcelona
Torre NN,
Calle Tarragona, 157,
4a Planta,
Barcelona, 08014
Thank you
Editor's Notes
Good afternoon - it’s a pleasure to be asked to present here and talk about some of the things we’re working on at Skyscanner – particularly around Echo and Alexa.
But first, I’ll give a quick introduction to Skyscanner and what we do. Skyscanner is a travel meta search, which means we search many providers on your behalf, to find all possible options and prices for flights, hotels and car hire. We started small in 2003 with our founder Gareth pretty much coding on his own.
Our initial goal was to have all budget flights in the UK….
To having 10 offices around the world and over 700 staff, and a lot of huge milestones along the way.
Now we can say we’re a truly global company, 80% outside our home market.
With our Skyscanner for Business platform, we aim to “power the travel internet”, with hundreds of partners using our services. From API access to white labels and simple widgets, we are extending our reach beyond our own brand.
Our original goal was to make travel booking as easy as it is to buy a book online. And we’ve made some great progress towards it.
But of course that goal is a moving target because nothing stays the same. Just as Amazon revolutionsed the sale of printed books, it went on to make electronic reading mainstream when the technology enabled it. In that same way we can’t be prescriptive about which platforms we build for, or complacent about what’s coming next.
Let's explore this for a minute. If you were to buy a book from Amazon, the outputs are fairly limited. It might be in paperback or hardback, it might be sold via Amazon, the marketplace or in Kindle form. But the input - the product you're requesting - is the same thing with the same quality.
Just taking air travel as one part of our industry, you can immediately see the differences. Firstly the inputs, the things you can request, vary greatly straight away.
There are almost unlimited date combinations, airlines who may or may not fly on those dates, the class of service you want and then various providers who will sell you a ticket, often with different pricing models.
There are 700 airlines in the world and 14,000 airports - in theory you can create any route you want between two places. Even if I know I'm going from Edinburgh to our office in Singapore, with one transfer in between, I'll be able to choose over 100 routes, multiplied by dozens of sales providers. And I'm just making one search of the millions we get each day.
Of course we can't complain, we love our users - and we want to answer these queries as quickly as the book search, despite the added complexity. We travel as much as anyone and demand the same fast answers as our users.
But the complexity I just described was a problem 12 years ago when we started. Take everything I've just said and multiply it by the many platforms people use, with different devices, screen sizes, input and output methods, and the different contexts in which they are used. Our work is never done.
At Skyscanner, we have asserted to being a “mobile first” company for some time now. Not as a strategy as such, but as a way to answer the challenges posed by these expanding needs and platforms.
And to try and explain why we work this way, and the benefits it gives us, I’ll talk through some of the thought process that got us there in the first place. It’s good to start thinking about the question, “what is mobile?”, and we’ll use these guys on the screen as personas.
Is it someone on their phone, the simplest way to put it?
Is it someone on the train on a laptop? They’re mobile by definition and by the challenges they’d in using it.
Is it the guy with no obvious device, but earphones? Maybe he’s got a kindle hidden inside the Financial Times.
At Skyscanner, we have asserted to being a “mobile first” company for some time now. Not as a strategy as such, but as a way to answer the challenges posed by these expanding needs and platforms.
And to try and explain why we work this way, and the benefits it gives us, I’ll talk through some of the thought process that got us there in the first place. It’s good to start thinking about the question, “what is mobile?”, and we’ll use these guys on the screen as personas.
Is it someone on their phone, the simplest way to put it?
Is it someone on the train on a laptop? They’re mobile by definition and by the challenges they’d in using it.
Is it the guy with no obvious device, but earphones? Maybe he’s got a kindle hidden inside the Financial Times.
At Skyscanner, we have asserted to being a “mobile first” company for some time now. Not as a strategy as such, but as a way to answer the challenges posed by these expanding needs and platforms.
And to try and explain why we work this way, and the benefits it gives us, I’ll talk through some of the thought process that got us there in the first place. It’s good to start thinking about the question, “what is mobile?”, and we’ll use these guys on the screen as personas.
Is it someone on their phone, the simplest way to put it?
Is it someone on the train on a laptop? They’re mobile by definition and by the challenges they’d in using it.
Is it the guy with no obvious device, but earphones? Maybe he’s got a kindle hidden inside the Financial Times.
In truth it’s all of them - but that’s just the start of the explanation.
In the first two examples, we’re not really talking about a platform but a context. And when you drill into more of those, and the additional platforms that mobile has created, you’re talking about infinite possibilities.
And that has has been documented as the ‘mobile mind shift’. It’s not about phones, it’s about hitting people with the information they want, at the time they need it, without them having to do very much.
Travel is a needs-based industry and we can’t predict when those needs are going happen.
In truth it’s all of them - but that’s just the start of the explanation.
In the first two examples, we’re not really talking about a platform but a context. And when you drill into more of those, and the additional platforms that mobile has created, you’re talking about infinite possibilities.
And that has has been documented as the ‘mobile mind shift’. It’s not about phones, it’s about hitting people with the information they want, at the time they need it, without them having to do very much.
Take an example closer to home - maybe someone doesn’t want a flight price, maybe they just want to know where they can go, or which airline takes them there.
Small queries that should be delivered by microservices to fragmented clients. Those are ideal building blocks for modern low-friction products - apps and web, and clients with no visuals like IoT or of course, voice.
In our report, the Future of Travel, skyscanner2024.com - we talked about the notion of “travel buddies” - assistants that know your recent history, your lifestyle preferences and even that you need to slow down your life a little bit.
It might know that you’re a regular business traveller, you never check bags onto a plane, and you always want hotel rooms within a mile of your meetings. You always fly first class and always want a four star hotel.
Or you might have a preference for hotels near the beach in sunny places with rooftop gyms, and you like to be told when those kind of deals come up, new places to try with your unique needs and likely travel dates.
Travel will be more about the traveller coming first, giving intuitive ways for them to get what they want, maybe before they even know it.
To be honest, thanks to Moore’s Law, that future is almost here already.
Travel buddies, that know your behaviour patterns and suggest appropriate travel or even make the bookings for you, are most of the way there.
We already have big data - history, location awareness, connections to other people - which produce data sets the human brain would never comprehend. The challenge for companies like ours is to comprehend that data for the traveller - using whatever tools and technologies make sense at the time.
At Skyscanner, we try hard to embed ourselves into all new technologies as they emerge. We send price alerts, but we don’t mind who delivers them, or when. Platforms often know their users better than we do, and we leverage that.
We have a bot which responds in the Telegram app (show animated demo) because their API and SDK made that very straightforward. We’ve experimented with TV apps showing QR codes, early voice search in Windows and many other things.
Basically, we don’t really care where the users are, we’ll try and meet them there, in that particular context. We try hard to stay ahead of the curve.
Which brings us to Alexa…
We knew of Echo and how cool the hardware was. Then we had a pleasant surprise in summer, when one of Amazon’s architects showed us how simple it can be to create an Alexa app.
It was an obvious thing for us to jump on, another innovative service we can connect to our own platform, and a really exciting way to give people quick answers to travel queries. And it’s another important piece of the “travel buddy” jigsaw which gives answers in another context.
Before starting if I could give a developer any advice on how to approach Alexa - start thinking about your conversation before you think about writing any code.
Fire out early on what an Intent and a Slot mean. For a developer, the tech is straightforward - an Intent basically maps to a function or a method, Slots are parameters supplied with it.
Get it very clear in your head what you want the app to do, and the flow of conversations that lead to your answers. I’d even recommend reading things like “The Jack Principles”, which make the point that you never quite know what to expect - “you don’t know Jack” - and neither side of the conversation is really in control at any time.
Design the structure of your conversation, the likely dead ends, the core phrases you think people will ask. The variations you need to factor in.
Test the phrases against another human. Ask them for a flight to Amsterdam and get them to deliver an awkward response. Get that on the flowchart and test something else.
Don’t assume people will be rude, they might say “yes please” when you only really expected “yes”. They might be asked “when do you want to fly” and reply “My mum’s birthday”. There are infinite possibilities you might have to deal with.
To begin with, I had high ambitions to cover all sorts of complex travel queries, but there are two main problems with that.
First of all you end up with complicated phrases which are hard for Alexa to parse, and more likely to be misunderstood. When you get beyond six or seven words, it gets that bit more difficult to process succesfully.
Secondly, as I found out, you end up with the potential for mixed responses because Alexa doesn’t know which Intent answers which phrase. For a Skill like ours which needs up to six pieces of information to fire a travel search, that’s critical.
To get round this, we decided to create a reasonably linear conversation structure, but flexible enough to fill in the blanks we need.
{transition in slides}
We always need the search type - flights, cars, hotels
We always need at least a destination (plus departure for flight and cars)
We always need at least a start date.
But we also need to recognise people are unpredictable - they might be asked “do you know when you want to go” and say “yes” rather than the date.
But we also need to recognise people are unpredictable - they might add extra words like ‘please’ as I said, and you have to be able to handle that.
So we did it this way. We nominated that “places” would be the only free text that Alexa had to parse. Everything else would be controlled.
So we have an Intent for Dates with one single slot – the only part of our app that will see a date.
We have an intent for Search Types which is any variation on “flight”, “hotel” and “car rental”, and things like “a trip”, or “a holiday”.
We then have a controlled list of other phrases - “yes”, “no”, “correct”, “thank you” and so on – but the clever bit is knowing what the user is saying ‘yes’ to.
To make that last bit work, we carefully maintain and store any information already given to us, so we don’t ask for it again - which lets us continually ask the right questions until we have enough for a search.
If someone says “yes” or another similar phrase, the session knows the last question asked, so we know what “yes” applies to.
If you’ve done any kind of session management it should be straightforward - my last piece of advice would be, connect it up to a back-end server, so you don’t start the user from scratch every time. Ask if they want to start from their last search again, because Alexa is a bit forgetful on her own!
We have over 40 different nationalities at Skyscanner and to try and make sure we trained Alexa to respond each time, we threw lots of different voices and accents at her. This is a small sample of the kind of reactions we had.
We are proud to support Alexa and Echo, and plan to launch the Skyscanner Skill for Alexa before the end of the year - and keep adopting new features of Alexa as they come up.
We’ll go into a bit more technical detail on how we did this, and release some of the code to the community, on our Skyscanner for Business site, later this week.
Thanks for your time, enjoy your lunch!