Human underwriters still have the edge over their AI counterparts

Artificial intelligence is gaining ground, but humans still have their uses in private debt underwriting

It is a common and complacent misconception that highly skilled workers are less at risk of finding themselves usurped by artificial intelligence. Long after any doctor has been forced into early retirement by the synthetic brainpower of AI, the nursing assistant will still be lifting patients on to and out of beds – something robots still find it hard to do because of the sheer unpredictability of any physical action involving humans.

This is not a cheery thought for people in the cerebral area of private debt underwriting – and AI has already reached the smaller end of the market. ESF Capital, which lends between €500,000 and €5 million to European businesses, uses it to whittle down the number of potential borrowers to about twice the final total, says Alex Schmid, London-based founding partner and CEO of ESO Capital and chairman of ESF, an ESO Capital spin-out. The remainder are considered by humans. The firm also uses AI to scan the market for potentially suitable borrowers.

“If ESF didn’t use AI, in the long term it would not be able to compete well,” he says. “It would need a significantly larger team, which would mean a higher cost base.” However, when it comes to the final separation of the wheat from the chaff, “people look at the potential loans and say: ‘Does this make sense in my experience as a lender?’ At that point, human judgement becomes more critical in questions such as: ‘Is the owner trustworthy? Do they have a good or bad history? Can we work with this person?’”

A balance with the bots

However, it is reassuring for private debt underwriters that ESO Capital – which makes loans worth €10 million to €40 million – does not use AI in underwriting decisions. This is to do with the more manageable volume: it makes about five loans a year, which it decants from a funnel that starts at about 700 possible deals. “It’s relatively easy to reduce 700 to about 30 if it’s done by people with experience,” says Schmid. “Larger firms tend to have highly practised people who can make decisions more quickly and better than a machine through a few calls to character referees.”

Schmid’s assessment is echoed by Nicolas Nedelec, managing director at Paris-based private markets investor Idinvest Partners, which focuses on deals worth between €50 million and €70 million. “We don’t use [AI] on a regular basis, but sometimes in our due diligence process it can be useful,” he says.

Idinvest has used it to analyse potential borrowers, such as insurance brokers with millions of items of anonymised data on customer transactions that can only be processed by sophisticated and powerful AI.

However, Schmid believes AI could gain further ground in underwriting corporate debt deals both small and large as it becomes more powerful and as the datasets it crunches grow larger, more complete and more interlinked globally. It could even become expert in that last step of deciding whether a potential borrower is inherently trustworthy if it can combine “very detailed psychometric tests” with background checks and financially
based credit grading.

George Ralph, managing director at RFA, an IT services firm with a number of private debt clients, acknowledges that adoption of AI for underwriting is for the most part slow – for now. “What will happen, as happens with all technologies, is that one firm will adopt it and become incredibly competitive,” he says. “Then other firms will have to adopt the technology just to keep up.”

However, he adds that even when AI breaks through, humans will never be taken completely out of the underwriting process: “I think someone will always have to click ‘Accept’.”