UK Bank IT Train Wreck Demonstrates Why Algos Can Be a Terrible Idea

Posted on by

Given the walk-on-water abilities attributed to technology nowadays, for anyone outside of IT in general – and financial services’ IT especially – the reports that a big bank has lost an estimated £180m ($200m+) on a failed IT project might be surprising. That is the main take-away from the profit warning issued on Tuesday by, in this example, UK sub-prime lender Provident Financial.

According to the company’s Trading Statement Provident will run a £120m loss as opposed to a previously-stated estimate of a £60m profit. While some of that reversal of fortune could be attributed to higher loan loss provision and defaults which plague highly pro-cyclical lenders like Provident – as well as tighter wholesale credit market conditions – most of the loss would appear to be because of an IT project catastrophe.

Even for battle-hardened IT industry stalwarts such as I, this one is particularly gruesome.

For those outside of the profession (which makes snake oil salesmen look respectable and reputable) perhaps the best way to give some context and illustrate the scale of what has befallen the bank, HBO’s Game of Thrones fans (who are reasonably up to date with the show) will perhaps recall the recent episode where Daenerys Targaryen’s dragon obliterates the Lannister army? In the aftermath, the bedraggled stragglers who survived are shown wandering shell-shocked amongst the devastation. Well, that’s how Provident’s IT department are feeling right now. Having been through more failed IT projects than I care to recall, that’s pretty much what it’s like.

It almost seems an act of cruelty to lay before readers the full extent of the IT disaster which has engulfed Provident. But given this is a subprime lender and a particularly unpleasant one at that – their “business model” involves calling door-to-door in some of the UK’s most deprived areas selling high interest loans to people in desperate situations – I’m more than happy to twist the knife here.

Provident were attempting to make two concurrent changes.

Firstly, they wanted to change their salesforce and sales processes from freelance agents who worked on commission to fixed-earnings “employees”. I say “employees” because the contract of employment between Provident and their new salesforce gave Provident pretty much carte blanche to treat their workers as disposable if it suited Provident to dispose of them. This was hugely risky.

With previous arrangement of self-employed “agents”, who would sell the loans, the agents had skin in the game.

If the loans went bad, then some (as in, a lot or even all) of their annual commission earnings could be clawed back. So the commission-based agents would use local market knowledge and personal persuasion – which might not be particularly pleasant to potential defaulters but Provident would play a game of hear-no-evil-see-no-evil and blame a few “bad apples” if abuses were too overt – to make loans to only good (or less-bad) payers. Removing the commission-clawback incentive removed the new salaried salesforce’s reason to be as vigilant as possible about who they agreed to make loans to.

But the real issue was in the IT side of the change which Provident made to support the new sales forces and sales process. The commission-based salesforce possessed a huge amount of know-how and sector knowledge – much of which was very locality-specific – which was never written down. It existed only in the heads of the salespeople. Even then, this wasn’t necessarily the sort of thing you’d be able to capture in a scientific-standard textbook. Let alone an algorithm or automated credit decisioning matrix.

To give an example (which is informed by the author’s personal experience from my time in mainstream retail lending), let’s say an agent was “door stepping” in a community with a large cohort of impoverished households. Some of whom might not be in employment and subsisting entirely on public assistance, some of whom might be in employment but also relying on social security, some of whom might be in employment and not in receipt of social security and so on. Superficially, the best risk might appear to be on a loan to someone in employment. But it all depended on the exact nature of the employment.

Some employers may only give out hours of work on an ad-hoc basis and have a practice of deliberately short-changing employees occasionally by suddenly cutting the number of hours they allocated. They would do this to determine who is the most desperate. Anyone who had any other option would quit and go to work for another employer. If you are in a community retail bank, you quickly learn which employers play this nasty game with their employees. Those unfortunate enough to have to rely on such “jobs” had frequent gaps in their earnings. For whatever reason, they had little alternative or ability to find another, hopefully less-bad employer. If they went to a pay-day lender for a loan, which because of their spotty income reliability they often would, it would be very difficult to repay the principle or even interest because their income could drop to zero (as their employer cut their hours) and they’d end up in a hole they couldn’t easily climb out of again.

Conversely, a household which got its income entirely from social security (such as unemployment insurance, disability benefits, housing benefit and the like) had, at least, an income which wouldn’t usually drop to zero. While they may never be able to repay the principle of the loan, they’d be able to make interest payments, often for quite a long period. Eventually they might default on the principle loan amount, but by that time the subprime lender would have made enough in exorbitant interest to make a profit. This is, in essence, the way sub prime and payday lending works. You might well never get your original loan amount back. But you, the lender, wouldn’t care because you’d gouged a vast amount out from the borrower in interest.

However, it isn’t that straightforward. The personal circumstances of the borrower made this latter group of “better bets” a complex web of traits which would either help or hinder their ability to repay. Substance abuse or chemically dependent borrowers were almost always bad news for the agents selling payday loans. But then if the borrower could earn money from prostitution, then their loan repayment ability prospects went up. Compulsive gamblers were a good target for subprime loans, until they got involved with even worse loan sharks who would use violence to enforce repayment, in which eventuality the (comparatively speaking) “mainstream” subprime lender — such as Provident — would not be paid.

If eagle-eyed readers have cause to recall Yves’ preface to Paul Walsh’s article on how we exist increasingly in an environment where we are all, to a degree, buying or selling each other or ourselves, that’s because this is exactly what is happening here.

Against this backdrop, which is a microcosm of human misery and suffering, to those who preyed on it in the form of Provident’s commissioned-based salesforce and Provident’s clueless management – who may have known what really went on in the field but couldn’t publicly acknowledge it or might just as well have been completely ignorant in the way only today’s C-level executives can be – along came (presumably) IT consultants selling something possibly as least as dubious as Provident themselves sold: an IT project which would “transform” Provident’s “business”.

As Provident apparently decided to do, the IT project would capture the knowledge-base of the commission-based sales force and – I am not kidding here – turn it into an “app”. Readers might, at this point, be wondering if they’d read that right. Would, seriously, a bank actually turn the grim and grizzly reality of doorstep lending into an algorithm?

Would someone – for real – capture metrics and use-cases such as “crack addicted sex worker – good for 3 to 6 months interest only but then increasing probability that their pimp would be able to confiscate their entire takings” or “compulsive gambler – fine so long as they stick to the fixed-odds slots but don’t lend to them if they are getting into the horses at the track” into a database?

Even Provident probably balked at doing that. But as a result, their new technology platform was hopeless at credit decisioning and even worse at delinquent loan account recoveries for the social segment they had as their customer base.

I’ll quote in full from Provident’s own statement:

The new home credit operating model, which involves employing full-time Customer Experience Managers (CEMs) to serve customers rather than using self-employed agents, was deployed on 6 July 2017. This followed a period of higher operational disruption than planned between the announcement of the proposed structural changes on 31 January 2017 and deployment of the new operating model. The impact of higher than expected agent attrition and reduced agent effectiveness on collections performance and sales resulted in the announcement on 20 June 2017 that forecast pre-exceptional profits from CCD would be reduced to around £60m.

The primary objectives set for the third quarter of 2017 were to embed the new operating model and to progressively restore customer service and collections performance to acceptable levels in preparation for the seasonal peak in lending during the fourth quarter. The rate of progress being made is too weak and the business is now falling a long way short of achieving these objectives. Collections performance and sales are both showing substantial underperformance against the comparable period in 2016. The routing and scheduling software deployed to direct the daily activities of CEMs has presented some early issues, primarily relating to the integrity of data, and the prescriptive nature of the new operating model has not allowed sufficient local autonomy to prioritise resource allocation during this period of recovery.

Carnage is the only word to use. “…reduced agent effectiveness on collections performance”? I’ll wager that as a result of the IT project which delivered the “routing and scheduling software deployed to direct the daily activities of [the salesforce]” Provident scarcely, if at all, knows who its borrowers are and where they live.

After reading the full Statement (which includes some shocking metrics on recoveries rates), I concluded the only thing I’d rather be less than a Provident Financial customer is a worker in Provident’s IT department given the task of trying to sort this mess out. The main reason is, there simply isn’t a solution. The IT system Provident needs can never be built.

An anathema though it is to the army of tech boosters, paid Silicon Valley shills and unicorn-valued “disruptive” startup CEOs, some things – like the contents of Provident’s commission-based salesforce’s heads – cannot (and, in this case, probably should not) ever be captured in an algorithm, ported to an app or written up in a set of IT system documentation.

The fact that someone – not disclosed in Provident’s Statement but a reasonable assumption is that it is one of the larger IT consulting firms — could ever sell Provident on the notion that this was possible speaks volumes for management’s gullibility. But in that, they are merely reflecting and echoing a broader gullibility which seems to have gripped our entire collective culture. The sooner reality sets in, like it has for Provident Financial and their hapless investors, the better.

Readers may wonder why I bother, but in – as scant and as reluctant as I can make it – fairness to Provident, the programme they embarked on to automate via an IT solution their sales process may have been an attempt to clean up their act. By procedural-ising and automating the fieldforce’s customer relationship management and the collections and recoveries of arrears, there was the potential for Provident to better monitor and supervise its barely house-trained commission-based salespeople. But for businesses like Provident, it could well be that, if you try to clean up the business, you don’t have a business. How many other enterprises will suffer the same fate?

Print Friendly, PDF & Email