To get us started, Fabrice, can you give us a few key figures about Digital Insure?
FABRICE POMIERS: Digital Insure is a company founded in 2013, launched commercially in 2015. It employs 120 people based in Paris and Caen, France. Digital Insure has more than 300,000 policyholders in its portfolio, manages over 50 billion of insured capital and is growing rapidly. A member of the French Tech 120 class of 2020, Digital Insure is also ranked 1st Insurtech in the Frenchweb 500 ranking, for the 2nd year running.
Fabrice, what was your need when you approached reciTAL?
FP: We’re always looking for ways to optimize our relationship with our customers and distribution partners. As a broker, we work with a dozen insurers and are constantly expanding the range we manage. Initially, we focused on creditor insurance, but today we also operate in the provident and health sectors.
Faced with the rapid expansion of our product range and the growing complexity of our business, we need to help our customer service department in the broadest sense of the term to gain better control of information, and to be able to access it quickly and reliably.
Gilles, how does the reciTAL platform meet this need?
GILLES MOYSE: The reciTAL platform is an automated document processing platform. It has been specifically designed to process information quickly and efficiently.
It’s a platform that facilitates the work of managers in processing flows, i.e. the classification and extraction of information, but also in processing the stock of documents that is in fact their knowledge base: general terms and conditions, information leaflets, prospectuses, and all the documents that make it up.
At Digital Insure, we’ve introduced the Search product, which sends the right answer directly back to advisors, managers and staff when they ask a question to their knowledge base. This is precisely the subject Fabrice is addressing.
Can you tell us how AI makes this possible?
GM: The aim with AI is to make it invisible, even though it works on several levels.
The first place where AI comes into play is in the analysis of the document’s layout, in order to break it up and present the relevant part directly to the user. When a document is uploaded to the platform, the artificial intelligence – what we call layout models – is activated. – i.e. layout analysis models – will analyze the document in terms of the text it contains, and also in terms of its layout: is it a table, is it a graphic, what is the header, what is the footer, what is bold, what is centered, etc.? The advantage of these models is that they enable documents to be broken down into much smaller units, which in turn enable the user to directly find the right paragraph when typing a query. Instead of retrieving a document that may be fifty, sixty, two hundred pages long, it directly retrieves the relevant information.
The second point where AI comes into play is in the interpretation of a user’s question. We don’t just deal with the keyword, we bring a much finer understanding of the language to try and detect the user’s intention, whether he’s asking for a rate, an amount, a duration, a telephone number, an email, to go beyond the right paragraph and give him the answer directly.
This ability to return a precise answer to the question posed and link to the associated paragraph in the document is due to the Snippet functionality we’ll be rolling out very soon at Digital Insure. The idea is to access relevant information ever faster.
Fabrice, how did the implementation of the reciTAL solution go?
FP: It went really well! We have organized ourselves internally, with our experts in charge of training and quality control working daily with our customer advisors.
We have carried out essential preliminary work on the documentation itself. We’re still a fairly young company, with a strong focus on new products, so we both gathered existing documentation and turned to our consultants to ask them what they were missing. While we were working with reciTAL, we also completed the documentation with all the new elements we were able to collect, both in terms of content and form.
Once the final document had been produced, we carried out trials with advisors. There was a whole iteration between the fund where the documentation was enriched and the use of the tool. What kind of question is this? How do you place them? How relevant is it? etc. This work was carried out with the help of the reciTAL teams.
How well did the teams acclimatize once the solution was in production?
FP: They use it every day! We are very vigilant about its use, and also about the fact that we are welcoming a lot of new advisors. We need to train these new advisors in our products, and we use Search to help them integrate. There are advisors who have been with us for several years, and who see their skills broaden each time their scope expands, but there is also a legacy that comes directly to the new entrant, so this is where it seems most relevant to us. In particular, we are in a phase of development of our Customer Service department in Caen, where we have an intense need for training.
The second point is that we are developing a provident business in addition to our loan insurance business. As each of these products has its own specific characteristics, the reciTAL solution helps us to organize and distribute our knowledge more effectively.
Do you see an impact on your performance?
FP: We’ve just put the solution in place. What we’re measuring at the moment is the use of the tool and the quality of responses in general. We carry out standard checks on the responses recorded on our secure messaging system, and have noticed an improvement in the availability and reliability of the responses provided by the tool. The second step will be to look beyond the quality function and better measure efficiency.
How do the Digital Insure and reciTAL teams work together?
FP: We’ve found a partner who listens. We met at the Prix Finance Innovation 2019 and discovered a company growing like us, with the same issues: how to support and manage this growth and remain successful with quality services even in the current period.
As we have strong service commitments to our partners, in particular insurers and distributors, which are passed on to our suppliers, we appreciated the attentive ear of Gilles and his teams, enabling us to fulfill our own commitments.
Of course we like the agile fintech aspect, but that’s not the only aspect. We’re interested in building something reliable, durable and solid with reciTAL.
GM: Absolutely, that’s how we felt too. It’s great to work with Digital Insure because we feel we’re on the same wavelength.
We move forward together, we move fast, we’re agile and it’s great to talk to someone who’s like us. I also found that the tests carried out were very rigorous. We feel that our solution has a real impact, a real interest, with added value, which is very much appreciated.
In fact, our joint efforts to improve performance and optimize our processes have been fruitful. It’s a real pleasure to be working with Digital Insure today.
Fabrice, looking to the future, what do you see as the ideal augmented advisor?
FP: We need to simplify access to our products and support our customers. Our complex products must be easy to underwrite, with seamless processes. The augmented advisor is one who has a real ability to advise. We’re 100% digital, but that doesn’t mean 100% machine. The added value we bring today isn’t just about the digital journey, it’s essentially human.
When we’re selling individual protection products and Madelin-type personal protection products, where there’s a real duty to advise, the augmented advisor really comes into his own: he’s the one who knows more, as much as or more than any distributor, and who can really help the customer and find a solution.
Our aim is to distribute products via different types of more or less specialized networks, so the real success of augmented consulting lies in supporting non-specialist networks with more complex products that they don’t sell today. That’s our ultimate goal, within 2 or 3 years.
GM: That’s also what we’re seeing with our other customers today: nobody wants a completely automatic company. Firstly, because it doesn’t make you dream, and secondly, because it doesn’t work.
In concrete terms, when customers ask questions that are too complicated, the machine switches them over to the human? Is this the augmented advisor?
FP: In fact, it’s already human. Today, humans rely on their knowledge, the training they have received and the information we have made available to them. We are contacted either by secure messaging and respond in writing, or by telephone and respond directly. What the reciTAL solution improves is the immediacy of the information. Be able to go straight to reliable information and reduce waiting times. We reduce the number of “Thanks for calling, I’ll write to you or call you back”, because we provide quicker answers.
When the broker is face-to-face with the customer, that’s really what happens. For example, customers with real estate acquisition projects that are conditioned by deadlines feel real anxiety because these are life projects. Giving them an immediate response means helping them make their project a reality. And may it become a reality with us.
Do we need to anticipate all the questions that are going to be asked?
GM: That’s a good question! What’s interesting about our solutions is that there’s no need to anticipate questions, but it’s important to raise this point because it can be confused with chatbots.
Chatbots are machines, they’re basically FAQs with a set of questions and a set of answers. It’s static.
FP: But it’s a real choice, Gilles. If we chose to go for your solution, it’s because we had also studied other solutions, notably chatbot solutions. I’m not saying we’ll never use chatbots, but for other types of products that are more suitable.
GM: Yes, precisely because chatbots are actually editorial projects. An FAQ needs to be editorialized with a set of questions and a set of answers. The advantage is that you can plan a very precise rendering of the answers to these questions. But there’s a double drawback. The number of responses is inevitably limited, and these responses must be constantly updated.
To overcome this first drawback, one of our customers has integrated our Search solution directly into their chatbot. And finally, they figure they might as well do without the chatbot and solve the problem of updates that way. The ultimate goal is to have something that works directly on documents.
FP: Knowing that we already have to keep our documentation as exhaustive and up to date as possible means we have to maintain a double repository. We prefer to focus on a single repository.
Interviewed on 8/5/2021.