What role for humans in the future of collections?

Author: Calum Fuller, Credit Strategy 

As artificial intelligence (AI) matures and becomes more commonplace, panellists at Digital Banking Club for Collections Live Debate considered how roles in collections will change as the technology matures.

While artificial intelligence (AI) can process and address a substantial portion of customer issues, the advent of AI does not necessarily come at the expense of human input, the Digital Banking Club for Collections Live Debate, powered by Intelligent Environments, concluded.

Expert panellists agreed that while AI will ultimately change the collections industry forever, it is by no means time to abandon traditional, human-based methods, they said. Instead, it is a time to work out which techniques achieve the best results at defined points in the process.

While they agreed that AI systems such as chatbots could introduce significant efficiencies, the knock-on effect for workers would be to move them up the value chain, argued the panel comprising Martha Bennett, principal analyst at Forrester Research, Stuart Sykes, group customer operations director at MyJar, Terry Cordeiro, head of product management at Lloyds Banking Group, and Simon Cadbury, director of strategy, marketing and innovation at Intelligent Environments.

In cases such as collecting customer debt, the experts noted chatbots are adept at managing the simplest, most common cases and can even guide the customer through an income and expenditure assessment to produce a proposed repayment plan, as was demonstrated by Intelligent Environments’ own Codee chatbot prior to the debate.

Indeed, AI is quite at home managing complexity too, the panellists noted early on in the discussion. Despite that, it is where AI encounters exceptional and unprecedented situations that processes can begin to encounter difficulties.

“We quite right rightly call out complexity, but one challenge for a lot of organisations is that they meld complexity and exceptions,” explained Bennett. “that needs to be pulled apart because computers are perfectly capable of handling complexity, but for anything that’s complex, there also needs to be enough precedent. It can be complex, but it needs to be repeated.” “In particular, AI is ideal for repeatable tasks where the need is repetitive but the outcomes vary,” said Cadbury. “AI is suited to areas that are more predictable and well-understood. It focusses on exploiting what is known.”

For exceptional cases, though, many providers of chatbots, including Codee, have built in a feature that allows users to break out of the journey and speak to customer services should they need to.

Notwithstanding that option, companies investing in AI need to be aware of its limitations, the panel said. AI does not understand context, subtlety and humour, nor can it exercise discretion.

“It really speaks to the vulnerability issue,” noted Lloyds’ Terry Cordeiro. “Identifying vulnerability involves comprehending and diagnosing vulnerable situations rather than vulnerable customers per se. Each one of those is going to be pretty individual, so a lot those are going to be exceptions.”

“Machines don’t understand context, people do,” he added. “As AI becomes more common, humans will be retrained and redistributed into different roles rather than replaced.”

Instead of acting as a panacea in the management of vulnerable customers, AI instead acts as a filter, catching and dealing with the most frequently-seen issues customers have and addressing them as far as possible, noted Intelligent Environments’ Cadbury.

If circumstances dictate that the AI programme cannot resolve a customer’s problem, they can be passed on to a member of staff who will have greater time, space and resources to give the customer the attention they deserve than they would have before the use of AI.

“[With AI], you’ve got data, good parameters and situations that fall within them – collections falls very neatly into that bracket,” said Cadbury. “Codee is effectively working through a number of dynamic workflows. When you dissect the most common customer interactions from a collections perspective, it will follow those workflows.”

One example Cadbury gave was that of paying off a loan, and whether they can pay it off all at once or over a period of time.

Quality of design

AI, though, is heavily dependent on the robustness of its design and the quality of the information it is provided with, Bennett told the DBC.

Systems, including AIs, that are not constructed on the basis of reliable information, without filters and defences against undesirable inputs, such as foul and abusive language or poor data, can produce problematic outcomes, the panel said. 

Heeding those points, the panel, and in particular, Cordeiro, were keen to highlight the role reputation plays and the need to eliminate that risk in the construction of AI programmes.

He identified the issue of inadvertently building bias into AI, a phenomenon which has its roots in social institutions, practices, and attitudes of the people building the AI or IT systems.

“Machines don’t cause reputational damage, people do,” Cordeiro cautioned. “Who is accountable? How do we know that the people putting information into the system are not putting bias into the system? Have you got visibility across people who are training the system? Have you got people from all walks of life, different backgrounds and the right mix of men and women? What are you putting into the system, and how do you know that it’s being done safely? All that feeds into what comes out as a result.”

Indeed, Cadbury noted that bots are not yet close to conducting themselves in a human-like manner and pretending they are will set expectations that it will be very difficult to meet. As such, he said, it is best to ensure humans are in place to quickly take over chats within the context of the conversation or previous customer engagements so that mistakes can be corrected quickly.

For smaller businesses considering utilising AI, it can be most effective to target “low-hanging fruit” when it comes to AI and machine learning, added MyJar’s Stuart Sykes.

“You need to have your building blocks ready,” he said. “You need to be looking at what data sources you need, what information you need to make good decisions. That’s where the difficulty comes in, certainly for debt collectors. They are only receiving the data from the lender, so if the lender is not prepared or does not have the data, they are limited in what they can do.”

He added, though, that lenders are beginning to become more receptive in their attitude to sharing data, meaning decisions can be made over the best method to contact customers, something Sykes said is “very important to the building blocks of machine learning”.

Bennett added that where the EU’s General Data Protection Regulation (GDPR) is concerned there is some debate over the level of disclosure required around how an algorithm works.

“One thing that’s clear is that the explanation in a court of law that ‘the machine did it and we had no idea why and how’ is not going to wash,” she said.

Triage

With those experiences in mind, the panel then argued that AI ought to be treated as a triage mechanism, rather than something that threatens to entirely remove human involvement.

In this way, AI can act as an efficient filtration system, moving human input higher up the value chain and into positions which involve addressing more novel and challenging issues and being more strategic.

“Triaging away all the repetitive elements – that technology is there today and you can solve problems today,” Cordeiro said. “I would counter that if you want to go deeper than that, you

 need to think more carefully about how your organisation is structured, the types of people who you will need operating the system, the fact the it continuously needs training.”

Should AI prove as successful as expected, the panel concluded, it will reduce the time and cost it takes to address a customer query and offer an additional channel through which customer services may be provided, 24 hours per day, at scale, throughout the year.

“People will get retrained, there are still going to be people [working in collections]. I believe that technology will offer more creative jobs for us and we will be doing very different things in five-to-10 years,” Cordeiro said shortly before the end of the discussion.

22 May 2018

Author: Calum Fuller, Credit Strategy 

As artificial intelligence (AI) matures and becomes more commonplace, panellists at Digital Banking Club for Collections Live Debate considered how roles in collections will change as the technology matures.

While artificial intelligence (AI) can process and address a substantial portion of customer issues, the advent of AI does not necessarily come at the expense of human input, the Digital Banking Club for Collections Live Debate, powered by Intelligent Environments, concluded.

Expert panellists agreed that while AI will ultimately change the collections industry forever, it is by no means time to abandon traditional, human-based methods, they said. Instead, it is a time to work out which techniques achieve the best results at defined points in the process.

While they agreed that AI systems such as chatbots could introduce significant efficiencies, the knock-on effect for workers would be to move them up the value chain, argued the panel comprising Martha Bennett, principal analyst at Forrester Research, Stuart Sykes, group customer operations director at MyJar, Terry Cordeiro, head of product management at Lloyds Banking Group, and Simon Cadbury, director of strategy, marketing and innovation at Intelligent Environments.

In cases such as collecting customer debt, the experts noted chatbots are adept at managing the simplest, most common cases and can even guide the customer through an income and expenditure assessment to produce a proposed repayment plan, as was demonstrated by Intelligent Environments’ own Codee chatbot prior to the debate.

Indeed, AI is quite at home managing complexity too, the panellists noted early on in the discussion. Despite that, it is where AI encounters exceptional and unprecedented situations that processes can begin to encounter difficulties.

“We quite right rightly call out complexity, but one challenge for a lot of organisations is that they meld complexity and exceptions,” explained Bennett. “that needs to be pulled apart because computers are perfectly capable of handling complexity, but for anything that’s complex, there also needs to be enough precedent. It can be complex, but it needs to be repeated.” “In particular, AI is ideal for repeatable tasks where the need is repetitive but the outcomes vary,” said Cadbury. “AI is suited to areas that are more predictable and well-understood. It focusses on exploiting what is known.”

For exceptional cases, though, many providers of chatbots, including Codee, have built in a feature that allows users to break out of the journey and speak to customer services should they need to.

Notwithstanding that option, companies investing in AI need to be aware of its limitations, the panel said. AI does not understand context, subtlety and humour, nor can it exercise discretion.

“It really speaks to the vulnerability issue,” noted Lloyds’ Terry Cordeiro. “Identifying vulnerability involves comprehending and diagnosing vulnerable situations rather than vulnerable customers per se. Each one of those is going to be pretty individual, so a lot those are going to be exceptions.”

“Machines don’t understand context, people do,” he added. “As AI becomes more common, humans will be retrained and redistributed into different roles rather than replaced.”

Instead of acting as a panacea in the management of vulnerable customers, AI instead acts as a filter, catching and dealing with the most frequently-seen issues customers have and addressing them as far as possible, noted Intelligent Environments’ Cadbury.

If circumstances dictate that the AI programme cannot resolve a customer’s problem, they can be passed on to a member of staff who will have greater time, space and resources to give the customer the attention they deserve than they would have before the use of AI.

“[With AI], you’ve got data, good parameters and situations that fall within them – collections falls very neatly into that bracket,” said Cadbury. “Codee is effectively working through a number of dynamic workflows. When you dissect the most common customer interactions from a collections perspective, it will follow those workflows.”

One example Cadbury gave was that of paying off a loan, and whether they can pay it off all at once or over a period of time.

Quality of design

AI, though, is heavily dependent on the robustness of its design and the quality of the information it is provided with, Bennett told the DBC.

Systems, including AIs, that are not constructed on the basis of reliable information, without filters and defences against undesirable inputs, such as foul and abusive language or poor data, can produce problematic outcomes, the panel said. 

Heeding those points, the panel, and in particular, Cordeiro, were keen to highlight the role reputation plays and the need to eliminate that risk in the construction of AI programmes.

He identified the issue of inadvertently building bias into AI, a phenomenon which has its roots in social institutions, practices, and attitudes of the people building the AI or IT systems.

“Machines don’t cause reputational damage, people do,” Cordeiro cautioned. “Who is accountable? How do we know that the people putting information into the system are not putting bias into the system? Have you got visibility across people who are training the system? Have you got people from all walks of life, different backgrounds and the right mix of men and women? What are you putting into the system, and how do you know that it’s being done safely? All that feeds into what comes out as a result.”

Indeed, Cadbury noted that bots are not yet close to conducting themselves in a human-like manner and pretending they are will set expectations that it will be very difficult to meet. As such, he said, it is best to ensure humans are in place to quickly take over chats within the context of the conversation or previous customer engagements so that mistakes can be corrected quickly.

For smaller businesses considering utilising AI, it can be most effective to target “low-hanging fruit” when it comes to AI and machine learning, added MyJar’s Stuart Sykes.

“You need to have your building blocks ready,” he said. “You need to be looking at what data sources you need, what information you need to make good decisions. That’s where the difficulty comes in, certainly for debt collectors. They are only receiving the data from the lender, so if the lender is not prepared or does not have the data, they are limited in what they can do.”

He added, though, that lenders are beginning to become more receptive in their attitude to sharing data, meaning decisions can be made over the best method to contact customers, something Sykes said is “very important to the building blocks of machine learning”.

Bennett added that where the EU’s General Data Protection Regulation (GDPR) is concerned there is some debate over the level of disclosure required around how an algorithm works.

“One thing that’s clear is that the explanation in a court of law that ‘the machine did it and we had no idea why and how’ is not going to wash,” she said.

Triage

With those experiences in mind, the panel then argued that AI ought to be treated as a triage mechanism, rather than something that threatens to entirely remove human involvement.

In this way, AI can act as an efficient filtration system, moving human input higher up the value chain and into positions which involve addressing more novel and challenging issues and being more strategic.

“Triaging away all the repetitive elements – that technology is there today and you can solve problems today,” Cordeiro said. “I would counter that if you want to go deeper than that, you

 need to think more carefully about how your organisation is structured, the types of people who you will need operating the system, the fact the it continuously needs training.”

Should AI prove as successful as expected, the panel concluded, it will reduce the time and cost it takes to address a customer query and offer an additional channel through which customer services may be provided, 24 hours per day, at scale, throughout the year.

“People will get retrained, there are still going to be people [working in collections]. I believe that technology will offer more creative jobs for us and we will be doing very different things in five-to-10 years,” Cordeiro said shortly before the end of the discussion.