Blog: AI? Cut the crap, you barely need BI
Almost every day I see or hear a ludicrous statement linking the travel industry with artificial intelligence and data science. Like “Big Data” was three years ago, and “Blockchain” sounds soooo 2017 now, these days AI is in everyone’s menu. Most of the time, it’s a marketing exploit: buzzwords are an infinite source of creativity for the sales bunch. No surprise, business owners in the travel arena were bombarded with revolutionary techno bullshit since the ’90s (“disruptive” and “innovative” seem to be the mainstream terms these days). Hey, stakeholders are not unaware of new technologies and trends. So I’d assume they won’t fall into crooked sellers’ traps (see here how to spot them >>).
The other field with amusing cases reported is recruitment. Last year, there was a sudden surge of data scientists job ads in travel, with absurd skill sets requirements (see examples here >>). This year, some newbie makes the same mistake now and then, but in general the HR departments realized that rainbow unicorns exist only in Silicon Valley.
Hence, this text is my tiny contribution to clear out some of the confusion about definitions, use cases and scenarios that involve AI and data science in B2B travel environments.
Get the basics at least
Top executives brag about the future of the industry, and how their supercharged alien technology subjugates consumers at light speed. Meanwhile, down their company ranks, almost nobody puts data to work. Only a few of the vertically integrated behemoths are actually cooking serious predictive analytics. A reduced bunch of big players are monitoring API performance (with Triometric >>), and even less of them are optimizing sales, inventory, etc.- Only last year a few bed banks and operators started to implement revenue management techniques (see article here >>). All of them, though, still keep data in silos.
As it is, there’s a tremendous lack of data-driven culture in travel companies. That is, not enough people taking action based on whatever internal and external data dictates (read an article on data-driven culture adoption here >>). Yes, everybody checks reports and spreadsheets. But that’s not even business intelligence! I’m not going to explain here what BI is (see here>>, if you want to know more about it), but I can point out what it is about: the past. Which is not a bad thing. It can explain what happened, and why the company or product behaved in a certain way, during an earlier period. Such information is great to improve processes or operations, for instance. Even more useful, though, is predicting future performance from historical data. Which can be done with simple statistic methods, because the data we manage in travel businesses is mainly structured (as in a spreadsheet) and anything but “big” (small amounts of it). Tell me, honestly: how many people do you see benefiting from data, in any kind of travel business? In early 2019, you´ll find much lip service to data analysis in our environment. But very few real practitioners.
From the advent of online booking engines and massive searches, datasets are getting bigger by the day. A neophyte might think “Ok, time to ditch old stats and try futuristic AI”… Truth be told, that’s not the case. Not necessarily, at best. Let me show you why.
Which is Which
Machine Learning (ML) is a branch of AI that uses huge (and I mean, HUGE) amounts of structured data to find patterns, verify trends, forecast possible outcomes and so on. At REVVA we used ML algorithms to analyze past bookings and future searches of an OTA. A great tool to predict destination demand or cancellations, forecast sales, and so on. Said OTA had a few dozen bookings and a few thousand searches per day, though. Not that big volume of data, you’d reckon! But more than enough to prove the method valid.
Now, if you take out the “ huge” from the above statement, you can do the same with old trusted statistic functions. No complex math, no sci-fi shenanigans: just simple formulas to calculate medians, standard deviations, regressions and so on. After all, the kind of questions we ask to data are, for instance: “How many passengers can I expect from market X during month Y, so I can allot Z seats?”. Classical statistics and ML provide a different framework and techniques, mostly based on the amount and quality of data you have to analyze. That said, anybody working in trading or casinos can prove that past performance is no great predictor of future performance. ML prevails because it can easily digest models with future data (searches) and delivers great accuracy, which improves over time. By definition, the more data you throw to the ML algorithm, the better it will perform next time. And that’s the “intelligent” part of this AI branch. But it requires a lot of human intervention (and no less than 6 months implementation) to achieve good results. In short, it could be said that ML is statistics on steroids. It’s much more than that, actually, but so far there are few direct applications in travel. Indirect applications are fraud detection and client retention techniques.
Deep Learning (DL), on the other hand, seems to have a better publicist than ML. Thanks to marketers’ efforts, the general public assumes that, being DL a much more complex practice than ML, it’s the magic bullet. Well, don’t waste your time: if you read or hear somebody affirming that they implemented deep learning into their travel-related solution, walk away. I’m not going to explain what DL and neural networks are, since I am no expert either. But I know enough to be able to qualify as a crook (or a moron) anybody that sells DL as a predictive tool, for instance. This is why:
- DL needs humongous amounts of unstructured data to find patterns and verify trends. Such data comes from audio, video, social media, and so on.
- DL is the aggregation of many distributed ML algorithms over a neural network, a device that somehow imitates a human brain’s process of learning. Sort of.
- Neural networks work over extremely expensive computing power, feed by even more expensive datasets. How many travel companies do you think can afford that? Certainly, no second-rank software house, even less a startup. Believe me, I tried.
Although ML is getting a solid foothold in travel products and services (for behemoths, mostly), DL is still way out of our league. To give perspective, current real-life DL applications are voice-recognition systems like Alexa, facial recognition in law-enforcement security systems, and autonomous vehicles. Impressive, to some extent. But it will take years to be as mainstream as certain vendors want us to believe. An interesting application I foresee for DL would be hotel/room mapping, a pain in the neck for so many buyers and suppliers in the accommodation industry. But as far as I know, nobody has yet developed such a powerful solution.
Hold your horses, Data Master!
We’ve seen that, for AI to be useful and productive, large or exceedingly large quantities of data are a must. Hence, even if you have ten thousand bookings per day, you barely qualify for a proper AI consumer. Next logical question would be, “ do I need data scientists to cope with this stuff? “. If you paid attention so far, you know the answer by now. Still, it looks like management has been relating data analysis and interpretation to programmers. That’s one of the main reasons why data-driven cultures do not proliferate enterprise-wide. It makes zero sense trying to step up from basic reporting to unsupervised learning algorithms, with no intermediate stop. Even more so with your teenie weenie, tabular, half a tera daily dataset. To put it graphically: even if you’re the sexier and richer hamster in the jungle, you won’t get to shag Mrs Lion.
Before spending ludicrous sums trying to impregnate Big Lady Cat, then, you’d better assess your situation, starting from your dataset volume. Suppose your daily production, adding up all your data sources, is between a few gigas and a terabyte. First thing you need to make sure is that your data gets lovingly accommodated and managed, never siloed. Which is a data engineer’s job (following a data architect’s one-time input). Now, do you need to hire said specialists? It’s not a bad idea if you have the budget. Please don’t try to recycle one of your IT guys for this (unless they’re ready to learn a lot of new stuff). A better idea would be to outsource the task to nice people like the Bluekiri >> bunch.
Next, to analyze the data, should you hire an academical hot-shot, business-wise autistic? Let me recap what’s compulsory to achieve useful advanced analytics: a gigantic dataset comprising multiple data sources, a certain degree of data-driven culture, and at least one person that would interpret and act upon data insights. As you don’t have any of those, it’s preposterous to throw five-figure cheques to a mathematical mastermind. Back to the zoological analogy, it would be as employing a monkey to help you seduce the lioness. Sure, the primate has opposing thumbs and knows a great deal of recreational reproductive activities, but you’ll still be far from seizing the pussy(cat).
If you’re a small or medium wholesaler, hotel chain, DMC or bed bank, your best bet is to update your multi-sheet reporting system to proper business intelligence, first of all. A consultancy and a couple of BI tools might be more than enough to get you on track. Before your dataset grows with your business, you should sort basic architectural needs (scalable database and storage, etc.), prepared for analytical usage. Finally, when you realized the absolute need of a company-wide data-driven culture, a specific product and some extra consultancies could be all that you need to stay ahead of competition, reduce operative costs, and so on.
Or, if you’re the kind of person who likes to swagger around trade shows, boasting about your team of data scientists fishing amazing insights through AI from a turquoise data lake, be my guest. We all know you only have four clients in two markets, anyway. ;)
Thanks for reading!
Originally published at https://www.linkedin.com.