Calculate the following. You operate 1,800 flights from 85 bases every day. Each flight requires 100% presence of a pilot. Pilots may operate more than one flight per day (“segments”), up to a flying time limit of nine hours, but may be on duty for up to 13 hours. Pilots are permitted to take holidays at any time of year. How many pilots do you need?
As everybody now knows, Ryanair has been unable to answer that question, resulting in the cancellation of 40 to 50 flights each day for the next six weeks. The no-frills airline is likely to face compensation costs of at least €20 million. Given that the cause of the problem was a failure to adjust for a new holiday calendar, that is a very expensive mistake.
What not many people will know is that Ryanair is still run by spreadsheet. So are many companies and that is not necessarily a bad thing - it at least suggests that data and reporting play a role in the business - and it is what you might expect from a highly budget-conscious organisation. Why go to the expense of creating an analytics and business intelligence function when you can run everything in Excel? (And as the old joke about airlines has it, the only sum that matters is that the number of planes taking off equals the number of planes that land.)
But it does highlight the limitations of this kind of old-school approach. Human resources has a lot of mature ways to model out how many staff are needed to cover the work to be done. Call centres, for example, use a robust model called Erlang C which considers the predicted load (incoming calls), the desired answer and call handling times to determine how many customer service agents should be present in any given day part. Allowance can be made in the model for holidays and sickness, adjustable by acceptable elasticity in how long customers wait to have their call answered. Change calls to flights, build in 100% cover and you should be able to work out how many pilots need to be on duty.
The problem is that these kind of models, when built in spreadsheets, are effectively trapped within the function that created them, or even the laptop of the individual running them. Knowledge sharing is difficult, let alone the kind of predictive analytics which might have averted the current service failure. It is one of the first steps towards data and analytics maturity that such models get opened up and shared so they can be repurposed and reused, rather than rebuilt (or remaining unknown).
Compared to the compensation bill which Ryanair faces, the cost of introducing even a small-scale analytics team of, say, one to three individuals, is small. What requires more effort is changing the mindset from one that assumes each function owns data and models or that the actions of one function have no consequence for the other.
The analytics revolution which DataIQ continually advocates has still only impacted on a minority of organisations. As a result, mistakes get made that could be avoided. Failing to transfer a change in the holiday policy into the staffing schedule, as Ryanair just did, may have been simple human error. Failing to learn from that mistake and deciding to invest in a proper analytics function so that any repetition is avoided would be harder to forgive.