Why your telecom expense management sucks – a statistical view
The software on your smartphone was designed in the temperate zone, and it shows.
In the tropics, a typical day in the monsoon season can be hot and sunny through until mid-afternoon, then an intense 30-minute deluge, then hot and sunny again. Thirty minutes of rain in an otherwise sunny day. And yet the little weather icon on your smartphone is likely to show for the whole day a black cloud icon with rain falling out of it. It is true that it will rain, but the forecast is as good as useless.
Weather forecasting is not so easy, but stupid generalisations like many smartphone weather reports make the problem worse. Rather than considering all available data, and then providing an analytics-based report, we get a simplified prediction based on only a few data points (will it rain today: yes or no).
Nate Silver calls this mistake an “hypothesis test” using arbitrary “significance levels” to “accept or reject” a single parameter value. It’s inaccurate, and in forecasting, it’s usually misleading.
There are many statistical techniques to improve forecasting: linear regression, linear discriminant analysis, decision trees, K-means clustering, Gaussian distribution, Bayes analysis and many others. Silver, the current guru of statistical forecasting, favours an approach that could be called “non-naïve Bayes”.
However, discussions of the best forecasting technique often miss the point that forecasting in general is a second-choice approach. Forecasting is what you do when you can’t “know”. We can’t know the future, so we forecast the future instead. But even the best forecasts risk confusing the signal with the noise.
In the world of Telecom Expense Management, the ‘second-choice’ approach is surprisingly common.
The ‘second-choice’ approach to validation is to sample the data, and to build inferences, estimates and guesses of the total situation based on statistical samples. The results may point to the real story, or they may be as sensible as that smartphone rain icon while you’re getting burned by the tropical sun. In the world of telecoms expense management, data sampling can get you burned too, and very expensively so.
Validating an organisation’s telecoms expenses is not a future-centric problem. All the data is available, in complete detail.
The ‘first-choice’ approach to telecoms expense management is to avoid sampling completely. Why sample, when you can easily analyse all the available data?
Telecommunications usage and charging data is complex. In most markets, it is by far the most complex type of usage data there is. Partly this is due to the fluid and flexible nature of communications, but mostly it is due to the habit of telecoms carriers of imposing wickedly complex (some would say tricky) pricing regimes.
Some vendors utilize powerful validation engines specifically optimized for doing line-by-line validation of an entire organisation’s telecoms charges and usage. Validation of every single event, every single interaction, every single charge. No sampling, no statistical guesses. Vendors apply analytics to this output to give a 100% accurate view of an organisation’s usage and spend and what it means for the business.
Critically, line-by-line validation also has an immediate monetary benefit. Errors or overcharges found on bills can be immediately raised for carrier refund. With sampling, errors can be difficult to substantiate, or else they are simply not found at all. (If your sampling approach ‘predicts’ a 7% overcharge on a bill, your carrier is unlikely to take action, but if your line-by-line analysis factually proves a 7% overcharge on a bill, the carrier can be compelled to credit the error.)
If your vendor is doing line-by-line validation and providing real value, they should be able to prove it. Ask them to provide a “value register”, itemizing the specific line-by-line savings they have made for you. If they can’t provide it, they might just be doing data sampling, with no line-by-line validation at all.
Validation of telecoms usage and expenses is a financially critical activity.
Some telecom expense validations give you hard data, and real financial benefit. Others are more like a bad weather forecast – trivial, attention-seeking, but basically useless.
Which is your organisation’s approach to telecom expense management?