How machine learning could give KPIs new meaning
Machine learning can help companies to identify entirely new metrics for a fast-changing market, writes Tom Bacon
It is widely acknowledged today that machine learning is already helping companies achieve their performance objectives by optimising existing performance metrics. By exploiting the growing volumes of data on customer behaviour, pricing, competitive actions and operational statistics, it can help travel companies in a number of ways. Indeed, from optimising marketing or pricing to improved customer service and operational efficiency, machine learning is delivering benefits. However, a recent article in MIT Sloan Management Review indicates that companies are increasingly using machine learning to identify entirely new metrics, new key performance indictors (KPIs) that more directly correlate with overall performance.
So, how can machine learning change KPIs? Well, it doesn’t take ML to conclude that measuring ‘total revenue’ is better than simply ‘base fare’ for airlines in a world of significant ancillary revenue. Similarly, airlines track ‘load factor’ and ‘average yield’, recognising that they are to some extent trade-offs; success requires a balance between the two. However, ML can be used to find less obvious metrics that correlate to overall success.
ML can reveal new meaningful KPIs by:
· Meeting needs in a changing market: New metrics that may drive overall success better than old metrics given changes in the market or the competitive environment. For example, with the growing importance of ancillary revenue, new ancillary-related metrics are now more useful. There is not currently, however, consensus as to how best to look at ancillary.
· Focusing on greater personalisation and segmentation: New merchandising tools based on greater personalisation means different segmentation schemes are now relevant. ML can help identify which customer segments should be the focus of more targeted offers – and new segment-specific KPIs designed around better meeting individual customer needs may represent a better path to overall success
Let’s look more closely at ancillary fees. In last year’s annual report, Ideaaworks points out that those with ancillary spend approaching 50% of total revenue rely disproportionally on baggage fees. Potentially, for some airlines, baggage revenue per passenger is actually as important to overall revenue performance as the base fare. With some bag fees exceeding $30, the difference between a customer paying for a bag or not, will often be more than the fare premium for the next highest fare. In fact, not getting bag fees for a particular flight could easily mean 20% less revenue – potentially the difference between a highly profitable flight, and a financial disaster.
...not getting bag fees for a particular flight could easily mean 20% less revenue
Thus, baggage revenue per passenger may be a critical KPI for these carriers; monitoring both the fee and take-up rates to maximise this source of revenue is fundamental to the airline’s financial success. Similarly, ML could identify other critical revenue sources that ultimately drive flight or sector profitability and which should be focused on as a critical KPI for the organisation.
Going a step further, baggage revenue may be maximised with certain fare types or certain flights or customer segments. Isolating the factors – or customer profiles – that most impact bag revenue can help an organisation focus on the right levers for success. Is bag revenue higher for business travellers or young families? How does baggage revenue vary by channel? ML can help airlines identify new metrics, new KPI’s, that will further improve performance on a customer segment, channel, or market basis.
For legacy carriers, the percentage of revenue from full fare passengers once was a key KPI. A flight with less than 10% full fare passengers was associated with extremely high, potentially unachievable, breakeven load factors. Now with full fare less popular – but with a plethora of new fare choices – the legacy airlines need a new metric that correlates with higher revenue or increased upsell. In fact, ML applied to the old legacy metric – trying to maximise full fare – is not likely to produce the desired results: there are so few passengers who pay full fare on any flight. Now, flight profitability may be more directly linked with the percentage of customers not paying the new basic economy fares. And, as with the ULCC’s proposed dissection of baggage revenue, ML can offer the legacy carriers insight into the factors or customer segments that are most prone to the desired upsell.
Machine learning is a tremendous new opportunity for airlines as well as other travel suppliers. But rather than simply using ML to optimise performance based on traditional KPIs it can be applied to identifying new ones that can be more meaningful in the changing marketplace.