Today’s advisor is industry veteran Scott Gillespie, managing director of tClara. Scott’s guidepost tackles what travel program managers should try to predict — and what to skip.


Predictive analytics is a shiny new toy that has caught the travel industry’s eye. It sounds useful — check. It sounds cool — check. It scores points in management-speak bingo contests — check.

Is there any real value here, or is it mostly hype?

Well, let’s think about what needs predicting in the travel industry. More specifically, what would be valuable for travel managers to predict fairly accurately?

Predicting Obvious Things

Personalized itineraries

Let’s start with the “Let’s be like Amazon and Netflix” itinerary examples:

• We’ll predict the airline, departure time and seat assignment that Joe wants on his next trip.
• We’ll predict the hotel brand and location Joe wants, his favorite floor, bed type and preferred pillow.
• We’ll recommend a cool restaurant and a gym for Joe to try.

Should travel managers pay for these types of personalized itinerary predictions? It depends on the price of those predictions and their usefulness, compared with the time and cost of asking Joe, or better yet, reading his profile preferences. It’s hard to see how anyone can predict Joe’s preferences if Joe travels once or twice a year. There might be a bit of Amazon-type value if Joe is a commuter — the same trip every week — but that’s a very small part of the traveling population.

Note that an effort to personalize Joe’s itinerary is different from presenting Joe with three curated flight options and three hotel choices. One is trying to read Joe’s mind; the other is combining the company travel policy with a bit of display logic.

Trip costs
Predicted trip costs might be good to know. Wait, the GDSs do those predictions in their sleep. They call it shopping, so no management bingo points there, and no real value to “predicting” a trip’s cost. Admit you’re either shopping or estimating and move on.

What about the chore of grouping travel expenses into a total trip cost? That’s been buzzy, but I don’t see the value in this calculation. What are you getting that can’t be derived from traditional expense report analysis?

Long-term price forecasts
Budgeters could use the long-term predictions, but how much precision do they really need?

Ever notice that no one remembers that so-and-so’s annual price forecast was off by 3 percentage points? It’s because it doesn’t matter that much to corporate travel budget owners. There’s too much flux in the travel ecosystem to make anything traceable back to a bad forecast. Said differently, there’s no cost to the forecasters for being wrong.

Trip-specific price changes
This one is worth paying for. Who wouldn’t want to know how a trip’s price is likely to change over the next few days or weeks? Think Hopper for corporate travelers. Combine with an Upside-type element that shows how to get even better prices and this is a predictable winner.

Flight disruptions
Another predictable winner. Lots of data here for predicting the chance and severity of a flight delay on most any airline in most any city. Needs to be integrated into every corporate booking tool.

Predicting Less Obvious, Way More Valuable Things

By far, the best thing we could predict is how much money should be spent on travel to achieve the desired business goals — and how that money should be spent. That goal is unreachable today. But we can break that goal into some useful pieces, and from there we can see quite a few things that can and should be predicted.

Predicting the impact on road warriors
We need to focus on the 10 percent of travelers who spend half the travel budget and create most of the value from traveling. Predicting how to improve their ability to bring back positive business outcomes is a worthy and practical goal. We must learn how to predict these traveler-focused impacts:

• Traveler safety: Based on destination, travel itinerary and travel suppliers
• Traveler health: Impact on chronic diseases, behavioral risks, mental and emotional states, etc.
• Traveler productivity: Based on itinerary, time zones, cabin, connections, hotel, etc.
• Traveler burnout and attrition: Based on traveler friction, trip frequency, non-travel breaks, etc.

Predicting trip effectiveness
The ultimate goal is predicting the ROI of the trip. That’s a remarkably difficult task, but again, we can break it down into a couple of simpler goals.

Start by having travelers rate their trips, in hindsight, on a simple scale, such as:
• Not effective
• Somewhat effective
• Quite effective

Then focus on predicting the Trip Scrap Rate (percent of trips rated not effective) and the Trip Success Rate (percent of trips rated quite effective). Carlson Wagonlit did some very interesting work in 2014; ARC, Amex GBT and tClara studied these issues in 2016. Our industry needs to do much more.

What’s The Goal?

The obvious goal of predictive analytics is to make useful predictions. The less obvious goal is to learn about the factors that drive the predictions, and then to manage those factors so that our road warriors deliver better business outcomes.

Plenty of value will be created with predictive analytics. Let’s focus on predicting what really matters.


Related
• From Data Management Firms, Clients Demand The Future
• Lufthansa Group: Closers Get Champagne (By 2020)
• Chrome River Closing CRM Loop

10 Comments

  1. Great article, Scott. Well thought out as always. Another point to make in all this is that the nature of corporate travel is there is no “nature.” Every single program is different in their needs, goals, what they see as important, and what they believe they should measure. What’s important to company A is completely the opposite of company B and so on. One of the biggest and most important steps in any process, for implementing any technology (or program for that matter), is to truly step back and get crystal clear on what you want out of the program, why you want it, and the direct results getting it will provide. It’s the 80/20 rule or more likely the 90/10 rule. A very small amount of what you do in running things will create the most impact. You have to know what that is before going down the road of implementing any type of solution. Once you have clarity, then moving forward is possible and will generate real results.

  2. Dave, you make good points here – no surprise!

    Jenna, Michelle, what do you think are your program’s priorities, in the sense of wanting or needing to make valuable predictions for your programs?

    Amy, you have visibility of a lot of accounts. Any sense as to a common denominator among them for what they’d really like to see predicted?

  3. In Scott’s article, he asks “What about the chore of grouping travel expenses into a total trip cost? … What are you getting that can’t be derived from traditional expense analysis?”

    The answer is plenty.

    While Scott may be right about the incremental value of “predictive analysis” related to Total Trip Cost, the metric itself has value for advanced travel programs. Up to 40 percent of costs incurred by a traveler in a city are often not hotel-related.

    Total Cost of Ownership has been a highly valued metric by procurement for years and Total Trip Cost is a component of that metric (historically it’s just been difficult to accurately calculate). Looking at expense categories like meals or black car hire or taxi on their own as a category have merit. Taking a total cost view adds opportunities to drive incremental program value whether from a supplier management or policy compliance perspective. With the right technology calculating this metric is no longer a “chore.”

    Some of the Total Trip Cost insights we have seen:
    • What is the total average cost per day of a trip to New York or Boston or Kalamazoo — air plus incremental in-city costs. Taxi and meals are sometimes more expensive than the hotel. Are there travel alternatives?
    • Does one business unit or even individual traveler spend more overall when they go to a city than others? If yes, why?
    • Ability to create employee-level scorecards and then filter on trip parameters (i.e.. domestic/international), or employee role and level.
    • Total Trip Cost calculations can also be applied to meetings to allow all related spend by attendees to be aggregated to the total meeting cost.
    • Predict the total cost increases in a city impacted by major events (Super Bowl, NBA Finals, World Cup, Olympics, Apple annual meeting, Comicon) based on the effect in other cities.
    • Booked vs. paid vs. expensed analysis — unless you group transactions into a trip, this comparison is not possible on an aggregate level.

    We identify many trips that exceed the 3 percent metric Scott throws out, particularly for trips that incur change fees. Business trips change all the time and when you add up those change fees and ad/collect fees on nonrefundable tickets, the booked vs. expensed metrics often have a much larger than 3 percent variance.

    Scott draws out many good points in his article and most that I agree with. In fact, very recently he and I engaged in constructive dialogue around the value of a trip to the business, or capturing the ROI of that spend by analyzing trip effectiveness data.

    Hopefully this provides some perspective around his question, “What you are you getting that can’t be derived from traditional expense analysis?”

  4. Scott Gillespie’s assessment of the potential of applying predictive analytics to travel programs was, in my view, appropriately subdued. Predictive analytics, even with tools that do the calculations for you, is pretty heavy stuff. I don’t recall being in a meeting about business travel (or any meeting for that matter) where an executive challenges somebody’s claim, for example, that left-handed travelers prefer aisle seats to window seats by asking, “How many standard deviations was that from the mean?” For many questions that might be usefully answered by predictive analytics, the travel program’s data set is too thin. Above all, the answer is not specific to the particular travel program and is available from various third-party sources. The impact of delays at JFK on other airports comes to mind. If a travel program wants to get into predictive analytics it should, as Tom Tulloch advises in another context, look at the total cost ownership. In other words, factor in that you’ll be bidding against Goldman Sachs for MIT (Caltech, for those of you on the West Coast) graduates.

  5. As usual, Scott’s analysis is thorough, reliable and totally useful! We are very fortunate to have this gentleman’s insight and analysis of key business travel issues.
    Rolfe Shellenberger

Leave a Reply