cutting-edge AI research

How investment in cutting-edge AI research is set to pay dividends for sales and customer service at Sappi

In September 2022, Sappi partnered with the Rochester Institute of Technology (RIT) in the US to fund a Digital Innovation Lab as well as a PhD research student. The Lab’s research focuses on how to leverage data to gain a better understanding of customers.

The collaboration has already yielded promising results. In the process of designing an interactive machine learning system (IML) for Sappi’s sales organisation, the team had to address a number of challenges. The lessons they learned were outlined in a paper entitled ‘Four Challenges for IML Designers’ and presented at the ACM CHI Conference on Human Factors in Computing Systems in Hamburg last Spring. This premier international conference about human-computer interaction has as an overarching goal to make the world a better place with interactive digital technologies. 


Using AI to enable better customer insight

Dr. Javed Khan, head of UX at Sappi Europe’s Digital Transformation Team, co-authored the paper alongside RIT’s Professor Konstantinos Papangelis and PhD student Muhammad Raees. According to Khan there was a lot of interest in the paper’s themes and ideas at the conference. “This is a really hot topic in the community,” he says. “How to design human-centred AI applications.”

The paper outlines how a machine-learning algorithm might segment customers based on three variables: recency, frequency and monetary value of purchases. This information could then be used by salespeople to deliver better customer service and optimise resource allocation.

Khan envisions a number of practical applications that could help salespeople in the field. While previously they might have looked at their customer base from a subjective, personal perspective, now they can augment their current practice with a data-driven approach to customer segmentation.

The AI can also help salespeople engage in meaningful discussions with senior management, armed with the data to back up their recommendations. Another benefit is the ability to make more informed decisions about where to invest resources – by quickly identifying strategic customers.

Feedback from key salespeople will also allow the algorithm to improve as it is used more and more.

Making the most of company data

If the real-life applications of the research are fascinating, it’s also noteworthy that Sappi, a global renewable resource company, will invest in digital technology if aligned to its Digital Roadmap and when the use case demonstrates clear business benefit. “I think this is very telling of how important data and AI is for all companies – how much data companies generate, and how it can produce valuable insights,” says Khan. “But those insights can only be valuable with the right user experience of such AI applications.”

The RIT Lab plans to continue its research into developing IML applications for sales, but in the future Khan would like to expand the scope. “The focus for now is around sales,” he says. "But it could extend to other areas, such as supporting Sappi’s sustainability team.”

Four pointers for creating effective AI for salespeople

During their research, the team noticed some significant difficulties with the IML system – it was great at crunching numbers but lost the ‘human factor’ when it came to user experience. These insights were broken down into four lessons that Khan says can help designers create more effective machine-learning systems for salespeople.

1. Although the algorithm is good at creating cluster segments about customers, which it reproduces as three-dimensional graphs, it’s not so great at providing meaningful labels for the user. If salespeople have to pick apart dots on a graph to decode the meaning they are, as Khan puts it, “going to lose interest very fast”.

2. The overwhelming amount of information that the algorithm produces can have a paralyzing effect on users, who may be confused about where to start and what to do next. The paper suggests that designers should create systems that have a limited set of actionable insights – such as, who are your strategic customers? who are the customers you risk losing? – to guide users in their decision-making process.

3. Another important element for designers to include is validation mechanisms. This is to take account of the fact that a good salesperson ultimately knows better than the algorithm – and should be able to override the system.

4. The research also highlighted the need to offer a wide range of parameters for clustering, so empowering users to customise segmentations based on specific criteria.

If there is one key learning, according to Khan, it is that, “Insights from machine-learning applications are not enough. You need to take a human-centred AI approach, designing applications with the end-user in mind.”

‘Four Challenges for IML Designers’ can be explored in detail here