Your choice regarding cookies on this site
Click on Accept to agree or Preferences to view and choose your cookie settings.
Working hand in hand with experts from Fujitsu’s AI Center of Excellence, French IT services company D.FI has co-created an AI-powered IT service desk capability that has enhanced productivity by a third.
Headlines abound about exotic applications of AI technology, from piloting supertankers to composing symphonies. But there are plenty of deployments of AI in much more practical and — for IT executives — familiar settings.
Groupe D.FI is a French IT services company that provides hosting, cloud and managed services, with revenues of around €80 million ($88.7m) and a workforce of around 250. But in recent years it has seen its ability to combine sales growth and higher productivity thwarted by a resource-hungry activity: the efficiency with which its teams can process and resolve customers’ technology service requests.
Thomas Meunier, COO at D.FI, explains the scope of the challenge. “D.FI may be a relatively small company but demand for our services has grown to the point where our customers are raising around 10,000 service tickets a month,” he says. That translates into a vast volume of data: over 2 petabytes (PB) for the 100,000 tickets it deals with each year. Moreover, that data is largely unstructured and uncategorized, typically comprising the customer’s free-form description of the problem they’ve encountered.
“As an ambitious company, you can’t accept that your costs grow as fast as your revenues. You need to be more and more productive. The business was growing fast — but so were the costs, as a result of having to hire people to serve the customer accounts and their associated tickets,” says Meunier.
In 2018, the company became convinced that AI might help, but its leadership teams needed to find a partner willing or capable of investigating how to apply machine-learning technology to the challenge.
Partnering for success
“We knew AI could bring something disruptive to this issue. But we found ourselves struggling to find a partner who had the resources, infrastructure and commitment to explore this with us — without a defined outcome for each of us,” he says.
That frustration was addressed when D.FI met with an AI team from Fujitsu.
Just as the D.FI’s thinking on AI was taking shape, the Japanese technology and services company announced it was investing €50 million in France to support digital transformation, in a program that included establishing an AI Center of Excellence in the R&D hub of Paris-Saclay, home to the renowned École Polytechnique. Fujitsu backed that by offering partners access to high-performance computing capabilities and the expertise of more than 40 AI specialists and data scientists. “Everyone comes into AI with buzzwords but almost nobody delivers,” says Meunier. “Fujitsu’s announcement showed its commitment was real.”
École Polytechnique’s Drahi X-Novation Center, home to the Fujitsu AI Centre of Excellence
Although the two companies were convinced of the opportunity to apply AI to manage IT service tickets, when they embarked on what Meunier describes as “a genuine co-creation project,” in November 2018, they did not know what the outcome might be.
The Fujitsu team set about exploring more than 30 algorithms to see which — if any — was a good fit for the issues of ticket processing. That uncertainty of outcome related to the nature of the underlying data. Only limited intelligence could be derived from the unstructured way customers report issues. Moreover, there were doubts over whether even the 2PB of clean data D.FI made available was enough to train the AI algorithms.
“I am a big fan of the co-creation approach that Fujitsu has built and offered to us because together we were suddenly a team. We were together facing the same journey without knowing what was going to happen,” says Meunier.
Over three months, the Fujitsu team devoted more than 500 teraflops of computing power to analyzing the data and working through the multiple algorithms. Only two algorithms, both based on natural language processing, emerged as promising, Meunier outlines. Those were loaded on a Fujitsu server designed to function as an appliance, pre-processing service requests as they came in from the customer ticket database and offering recommendations to service agents.
Meunier explains the process: “Before a ticket is workflowed to one of our service agents, it is now caught by the AI engine, which analyzes and interprets it using the natural language algorithm. The magic of the AI is that it can understand and describe what the customer has written, which might be something as tricky as: ‘The network is not working as fast as I think it should, plus I’ve pinged a switch with this IP address and nothing came back.’
“Once the description has been understood and rationalized, the AI engine triggers an action to check the suggested source of the issue — in the example, it would send a request to a specific switch, checking its speed. Once the ticket is understood and the situation identified, the AI then goes into the database for the first time to check what action has been taken when a [service agent] has encountered a similar situation in the past to resolve it,” explains Meunier.
When it has understood the situation, the AI presents a set of four recommendations to the service agent, with one preferred action. Based on the nature of the issue and the pool of available expertise, the system also makes a judgment on the best person to receive the ticket.
“Having identified the complexity of the problem, the AI engine knows if a level 1 agent can deal with it or if it needs to be routed directly to a level 2 or level 3. That process alone saves 15 minutes per ticket,” he says.
Productivity and accuracy
The results have been as rapid as they have been impressive. “As if by magic, we have killed a third of the time a service agent spends making sense and describing the content of a ticket and presenting actions that hopefully lead to a resolution,” says Meunier.
There has been a learning curve, though, both for the teams and the AI algorithm. By mid-2019, the AI was generating a list of accurate recommended actions to the service agent in about 70% of cases. But as more tickets have been added and the algorithm refined, levels have risen — and were expected to hit 90% by the end of 2019. “The proof point for unplugging human intervention [for about a third of ticket handling] is at about 90% accurate. Shooting for 100% is not the goal here because over 90% is asymptotic.
“This is a game-changer,” says Meunier, “but it is not something easily achieved without commitment.” Over the four-month co-creation effort, Fujitsu and D.FI teams spent seven weeks together. And because of the pioneering aspect of the project, confidence in the outcome was not always high. “It was something of a roller-coaster, with days when we questioned if anything practical would emerge from the technology and the data.”
Above all, the application of AI is solving the fundamental issue that D.FI was facing — the limitations on its growth set by the cost of its manual processing of tickets. “We have definitely broken the direct link that existed between the growth of customer contracts and our costs going up,” says Meunier.
Another consequence, of course, is that the company will require fewer people for the early-stage processing of tickets. But, says Meunier, the application of AI means D.FI can confidently take on new contacts and grow the business faster while also ensuring agents who would have previously dealt with low-level tasks can now be deployed on more valuable and stimulating activities.
“With this co-creation we can look to something like 30% additional jobs and 30% additional growth within our business with no cost increase. And that is something that is going to be considered as much competitive by our customers,” says Meunier.
The AI translates into demonstrable benefits for customers too. As a result the time saved by the AI engine, key performance indicates such as systems availability have risen. “Customers love what D.FI and Fujitsu have done with AI,” says Meunier. “It also gives us the legitimacy to talk to the customers about AI projects. Artificial intelligence, machine learning, and deep learning are all hot and current topics for our customers, who are turning to us for help in identifying how to deploy these emerging technologies and solutions to address business challenges.”
This is a practical example of AI happening quickly, says Meunier. “Between shaking hands with the Fujitsu team and production was less than six weeks. Why? Because Fujitsu had the expertise, infrastructure to make it happen.” And, judging from that enthusiasm, he clearly has other opportunities in mind where AI technology can be a highly positive business disruptive.
• Watch a video interview with D.FI’s Thomas Meunier.
Click on Accept to agree or Preferences to view and choose your cookie settings.