Big data: profitable projects over big projects
Paradox: all the communication singing the praises of big data projects has done nothing to clear up the doubts that operations staff harbor in their daily activities. They are still trying to determine how such projects will benefit the business. The good news is that the hype is starting to fizzle out, leaving behind projects that may be less spectacular, but which are tangible and profitable.
“Our team of 10 data scientists have been working on data for two years. Millions have been ploughed into a Hadoop cluster and heavy artillery. But everyone in the group wonders what they’ve been doing. They haven’t produced anything.” This story (and similar stories have been heard in CAC 40 companies) is in stark contrast to what has been publically boasted in the media, from announcements of game-changing big data projects to certain organizations preaching how they led the way and just how modern, visionary, non-uberizable and converted to data they are.
But it is a completely different story in practice: our contacts in those companies have told us just how difficult it is for them to reconcile such claims and the world-enlightening promises of data science with their own experiences inside the company and on the front lines.
Questions marks are rightfully being raised about the trend towards big data, data science and analytics among business functions and IT professionals: “OK, but bottom line, what will it bring to the company?” What value will it actually create for the business functions and the company? What benefits come from exploiting data?
McKinsey came to the same conclusion in an article published in July 2017 on the results of analytics and data projects in insurance companies. Projects delivering high impact were only reported by… one in six respondents! Yet companies were investing as much as $80 million a year, and over half of the CEOs concerned considered data and analytics a top-five priority. 80% waste: that is currently the gap between communication and the real world.
Data do not automatically generate value
Does the use of data really bring something to the company? Meaningful information or a new insight? A recommendation that nobody had thought of before? Support for the decision-making process?
Companies often fail to ask these questions. The reasoning behind many projects runs in the opposite direction: “let’s build a huge system, and we’ll be damned if it doesn’t deliver something!”
The value of this “something” does not seem proportional to the hard work expended, as confirmed by this customer from the retail industry: “of course data scientists detect patterns. But if all they’ve got to say is that we need to stock up on stationery before the new school term begins or more six packs when there’s a football game on TV in the evening… then luckily we haven’t been waiting on their insights.”
Data can be exploited to produce relevant conclusions… without bringing anything to the company.
Generating value is all very well, but profitability is even better
IT services firms and software vendors have been surfing the wave by explaining that big data are synonymous with costly projects and prohibitively expensive solutions. With high buy-in costs and hit-and-miss results (one in six cases, according to the McKinsey survey): is it all worth it?
We have become too used to announcements and claims. When AlphaGo beat the world Go champion, its raw performance was acclaimed and there was talk about the future of unsupervised learning, and so on. But the investment involved seemed to have been forgotten, such as the dozens of employees working on the project for seven years and unlimited computer resources. Not everyone is Google. Not everyone can afford to spend such vast sums on publicity.
Back in the real world: “Our reject rate for this part is 40%,” said a representative from an automotive OEM. “I’m convinced we can halve that number by tapping into our machine tool data. But the part costs about $4 and any late deliveries could cost the company dearly in terms of penalties. Is it really worth investing millions into machine learning at the risk of disrupting our chain?” Good question…
Data analytics may improve performance… but without generating a return on investment, especially if the investments are considerable.
The time has come to say goodbye to the hype and instead focus on real data analytics projects
Gartner’s famous 2017 hype cycle put machine learning just after the first peak (the “Peak of Inflated Expectations”): the moment when the first results are disappointing in light of the hopes raised by the technology. The hype dies away and other subjects grab the media spotlight.
This is exactly the current atmosphere, and it is good news! Now is the time when true projects are launched and real use cases emerge. Our customers have been telling us: “I’ve got a real problem. I’ve got a case where data analytics could be profitable. It’s not as spectacular as revolutionizing the world, but it’s useful for our company.”
Generate value from iterative processes in close liaison with the business functions
The use of data can be profitable. I’m convinced. Our customers are also convinced. The proof is in the projects that we have been leading. But the projects creating the biggest buzz or the major “three years – three million” initiatives – that will pave the way to profitability.
Because nobody can predict the best way of exploiting data in a given sector. Only iterative-based work practices can find the path. The path is lined with a succession of specific benefits driven by feedback from internal and external customers and the actual benefits observed, not the result of a think tank’s brainstorming session.
When the circulation of data has been automated and become increasingly reliable, and when its daily benefits for guiding and supporting decisions have been substantiated, and when the data culture has been disseminated throughout the company, then… the team of 10 data scientists in the CAC 40 group will be a major asset for achieving even greater progress.