Profitability from predictions

Convert predictive analytics into money

In late 1997 I was helping some Egyptian DMC streamline operations, when terrorists stroked at Deir el-Bahri, the heartbreaking Luxor Massacre. About a year later, not one but two hurricanes (Mitch and George) broke havoc in the Caribbean, where I was operating my own DMC. In 2010 half the world’s travel flow was disrupted by Iceland’s Grimsvotn Volcano eruption, with European airline capacity cut by 75% for days. It’s impossible to predict such events, of course, but back then I would have given anything to get analytics help me solve the critical problem these tragedies have in common: passenger accommodation & extraction from affected areas.

And this is ONE of the many applications predictive analytics has in the travel trade. Although it might sound a bit Star Trek to many, PA is nothing more than finding hidden patterns in large quantities of data. Patterns that sometimes (not always) uncover unexpectedly profitable insights or equally costly inefficiencies, in the form or (generally) accurate predictions.

I’m not going to delve into the technicalities (leave the boring stuff for nerds like me). It should suffice to mention that PA is a toolbox that includes statistics, data science and some form of artificial intelligence. No need to know what’s under the hood, the market offers easy to use solutions that require no technical knowledge to bring results. Which can be spectacular.

predictive analytics cycle
PA is a cycle

What gives, man?

Predictions was something I longed for during my whole career, because I learnt early on that relatively easy statistical methods allow me to know how many vehicles I should contract next month, how many arrivals I might expect at a certain date from a defined market… Or, at the very least, what were my chances of getting into some girl’s Victoria’s Secret apparel. Results were always an approximation, though, which is way better than educated guesses. Today’s methods offer an accuracy never seen before, plus the combination of several sources of data can provide incredible insights. Even more incredible is the fact that there are platforms like REVVA, allowing non-technical staff, revenue managers and decision-makers get the information they need, packaged in a simple interface or dashboard. The beauty of it!
A word of advice against using numeric models for the Victoria’s Secret issue. Trust me, I tried: it’s a classification problem, not a regression problem. Ladies are unpredictable in their own predictable way, so advanced algorithms are required there.

Now, here’s a list of areas that can be improved by PA on any kind of travel business, no mater its size or sector:

  • Demand and sales forecasting
    As of today, Revenue Management systems and personnel still perform forecasting over historical data, with ages old “proven” methods. One of the very first things I learn in Stats 101 (in 1988!) was that time series (the past) is never enough evidence of what will happen in the future. Enter PA and machine learning techniques to bring us trustable predictions. A fascinating topic that I’ll address soon enough.
  • Automated alerts
    Several of our bed bank prospects observe that a client stopped sending bookings, or that a supplier all of a sudden is not giving availability, and it might be hours or even days before somebody notices. We worked on an automation that would alert on such cases, while the predictive engine will anticipate them if certain conditions appear to be materializing again. Any anomaly can be predicted and detected by a system boasting proper PA, no big deal there.
  • Marketing
    I guess it’s not difficult to see the huge impact PA has on conversion rates, ad campaigns, etc. Check your Facebook (for the 10th time today?) and see the advertising thrown at you. It’s not cookies, you know? A better example: try looking for a flight in and see how they’ll target you for a while with relevant averts on different social media (or even your inbox).
    This is something that deserves a whole article, I’m already working on it.
  • Recommendation systems
    Closely related to the above item, once an intelligent system knows who you are and what are your preferences, it will be able to suggest possible alternatives, depending on what you’re looking for. Although used for a long time in other industries (think Netflix), this has still to be developed, studied and proved profitable in the travel industry.
  • Segmentation
    Defining your customers by age or sex is a no-no. What we do today is to group them by buying behavior or wallet power, using clustering algorithms. That way, we can predict what each segment will buy, when and for how much. As you noticed, this goes hand in hand with the two items above, but as opposed as recommendation systems, this analysis can be done at a reasonable cost and with acceptable accuracy. Check my text on the matter >>
  • Dynamic pricing
    If you know in advance what a customer will buy and when, willing to spend this much, you can’t go wrong with an elastic pricing strategy. You’ll need to keep an eye on your competitor’s rates and demand, though. See my take on this subject >>
  • Other stuff
    There are more interesting applications of PA in the industry, like audience sentiment, image recognition, fraud detection and so on, but those are specific for certain sectors, or still far from profitability.Profitability from predictions

Make money right now

The not-so-distant future will be sawing all the above around an automated sales process. Here’s a possible scenario: a traveler issues a vocal requests to your booking engine for a flight+hotel trip in Paris next week. In a fraction of a second, the system “talks” with your CRM, checks past buys and searches from this very client, as well as his social media profiles (all of them!). At the same time, system verifies availabilities and rates from suppliers, comparing competitor’s prices from an RCS. As a result, the client’s search output includes a choice of his favorite hotel class, in his ideal area of the city, which will be reached via his preferred transport mean (not necessarily by plane), at an affordable price for him (the most profitable for you). Since we’re talking science fiction here, the system may also recommend and book ancillary services and activities, based on client’s personal tastes. What if client does not buy right away? System’s will remind him in a day or so, though phone or whatever we’ll be using in those days: “Hey, human! Do you still want to go to Paris? Check these new offers…. The Novotel Les Halles says that if you stay with them again, they’ll add free breakfast with your beloved Nutella crepes. There’s also a food festival in Montmartre during your searched dates, you know? Perhaps Rose and Alfred would love to go with you

Sounds appealing, right? And don’t get me started with payment options in this sci-fi film!

Ok, enough daydreaming. Can all this be translated into profits today, even if you run a small-sized company? Of course, you can, mon ami! Otherwise why would I be wasting my time with this infomercial informative piece?

In fact, we count among our clients small OTAs and DMCs which started right away benefitting from several of the above items. That’s my suggestion: start working at least on one item at a time, and I’d say go for segmentation first. No need to put in place complicated integrations, involve your IT staff or fork out large denomination bills. You’ll simply want to put your data to work. You have plenty of it!
Where, did you ask? If you own an electronic reservation system, be it online or offline, you already are collecting tons of data. If your booking system involves manually writing instead, you still have a chance to collect and process data, although I fear you might be reading this text on an IBM PC XT donated by the Red Cross.
If you don’t own any booking system, I assume you must be using some kind or ERP, and/or running AdWords and social media campaigns that bring leads to your website: those are excellent data sources to start working with, at least on segmentation predictions.
If your organization does not encompass ANY of the above, I can’t help but wonder what you understood so far… Are you a robot killer sent from the future to dispose of the asshole who started using “AI” as a marketing term?

Anyway, to wrap it up: getting money from predictive analytics requires overcoming two main challenges. First one, you guessed it: start collecting data! The rest is better left in the hands of solutions like REVVA, and that’s your second challenge: trusting a tiny portion of your hard-earnt cash to an unknown Spanish startup. If the second challenge is unacceptable to you, no problem. Go ahead and face seventeen more challenges on your own. Surely, you’ll save time and resources…

[ctt template=”3″ link=”dDcRw” via=”yes” ]Predictive analytics is not just another buzzword >> for posers, nor a secret weapon only accessible to vertically integrated German giants (not anymore).[/ctt]

It’s up to you now: either you try (for free) and reap the benefits with a ridiculously low investment, or stay in your Disney-like comfort zone, basking in the certainty that your gut predictions are -and always will be- a thousand times more accurate than whatever a stupid computer could guess.

Thanks for reading (and sharing, if you liked this)

Marcello Bresin