Robo advice – can it deliver good quality debt advice?

“Robo advice” is being used increasingly to provide investment advice to people who can’t afford individual advice from an IFA.

Can a similar approach deliver good quality debt advice through an automated online process? Face-to-face advice is sometimes seen as both expensive and a process that not all clients may want or need.

In this article I examine what robo debt advice can do well, the problems it has at the moment, and to what extent these can be resolved.

StepChange’s website invites people to go through an online process before talking to an adviser, saying:

Although the examples in this article are taken from StepChange’s online process, I am just using this as a lens to examine robo debt advice in general.

What is robo debt advice?

How does it work?

I am using the term to refer to a computer program where a client is asked about their situation. The questions would typically relate to the client’s debts, income, assets and expenses. Others can cover the reasons the client is seeking advice and whether they have any specific vulnerabilities.

From these inputs, the program can create an Income & Expenditure (I&E), use that to determine what the client’s possible and best options are, and report back to the client.

Robo advice is sometimes called Artificial Intelligence (AI), but here that seems to be a misnomer. The algorithms used to generate the advice are fixed, they cannot adjust themselves through experience, there is no machine learning here.

Google can learn common misspellings and improve its guesses at what you may be searching for. But robo debt advice just asks the same questions and processes the replies in the same way, mostly treating them as correct. The StepChange online process queries only a few of the expenses if they are very unusually low or high.

This isn’t what a human would do. When presented with an unusual answer or one that doesn’t seem compatible with a previous answer, a debt advisor will ask more questions. Why does someone have rent arrears when their I&E seems to show a large surplus, say. Perhaps the client has a gambling habit, is making large payments to a relative overseas, or they have simply missed off a big debt or expense. If a client has disability benefits, that will normally lead to a discussion about the extra costs associated with those.

The potential advantages of robo debt advice

With millions of people needing help with debts in the UK and a limited number of debt advisers, any process that can help give good quality debt advice and reduce the human input may enable more people to get the advice they need.

If the client needs or wants to speak to an adviser after the robo advice, it may have helped by recording a lot of client information, making the subsequent “full advice” faster and possibly less stressful for the client.

Another advantage is that the robo advice can be accessed 24/7, at any time that is convenient to the client. The client can also save progress and so take several sessions to enter all the information.

And some clients may simply prefer not to have to speak to a person at all if they are unclear what may be involved, are worried about being judged, are unsure if their situation is serious enough to bother an adviser with, or have mental health problems.

Problems with the inputs to robo advice

Debts and the language of debt advice

There are two problems with the process of a client giving details of their debts.

The first is that it can be a lengthy process if someone has a lot of debts. And they may not remember them all – especially if there are old debts they aren’t paying.

The second is the language used to determine if a debt is causing problems and what stage those difficulties have got to.  A human adviser adapts the language they use automatically where it becomes clear a client has difficulty with some of the terms.

The StepChange online process uses some formal terms like Default Notice and tries to describe what these mean. In my experience, many people have no idea what a Default Notice is or if they have received one. They may say Yes because they assume that missing a payment is the same as “defaulting” or because the debt is marked as defaulted on their credit record. Even if they have kept the letters, they won’t check them because they know what the answer is… The problem with a Letter Before Claim is even worse, as the letters may not even have that as the heading.

It asks if you are up to date or in arrears, it doesn’t ask if you have been up to date so far but can’t make this month’s payment… The decision engine may not be interested in that option, but it would be reassuring to the client who will feel they have explained what is happening. It also treats all contact before a Default Notice as “no action taken” – which is not how many clients will see getting texts, emails and letters talking about making payments, debt collectors etc.

The StepChange process also fails to ask key questions that would reveal an immediate need for assistance. The client is not asked if they have received a Claim form for unsecured debts. or a Notice Seeking Possession for rent arrears. And if they say they have a Suspended Possession Order, they are asked how much they have to pay but not whether they are up to date with the payments.

At the end of the debt input, StepChange asks the client if they dispute any of the balances owed or if they think any of the debts are unenforceable.  It mentions statute barring and provides a link to more information. But there is no mention of other reasons such as affordabilty and CCA agreements. Very few clients will be aware of these.

Some of these problems could be remedied by a more thorough set of questions in the robo advice. But the underlying difficulty of whether the client has understood the language of debt advice correctly will remain.

Income & Expenses

The StepChange online process has a list of benefits. But it doesn’t ask if any have been applied for but are not yet being paid. It also uses some language the client may well be unfamiliar with eg a non-dependant contribution.

When an income varies, it asks for recent amounts and takes an average.

StepChange says:

It’s important for your budget to represent your actual situation so we can give you the right advice. Try to be honest and realistic with the amounts you enter.

Sensible… but there is a problem here which applies to any robo debt advice. What is “the actual situation“? Should the client say what they have been spending, how much they think they can reduce that to, or what they ought to be spending?

A very common situation is where a client has been cutting back on essentials for a long while to try to make debt repayments. Food, adult clothes, heating, buildings and contents insurance, TV license… As a result, what they have been spending may not reflect what they actually need to spend. The client may not even be conscious of having done this.

The StepChange online advice process will query an extremely low amount for food or nothing for clothes, but if the client replies that is correct they are given a set of reasons to select from, but not one that says “It is all I can afford“.

And should the client be encouraged to put in larger expenses if they are expected to increase? Energy bills will go up in April for 22 million households, a child going to a new school is very likely to need new uniform, etc. Inflation is set to be a significant problem in 2022.

What will change in the next year

At the end of the input process, the StepChange online advice asks the client whether anything is likely to change in the near future. The client is asked for the month the change is expected but not how large the change may be.  But asking for the month seems too specific and not asking for an amount or any indication of probability seems an omission.

Can these input problems be eased with more tech?

The language issue

In other fields, work is being done on the use of natural language to make human-chatbot interactions more effective and friendly to a customer. But the issue here is different – the technical language of debt advice is not just not understood by most clients.

I doubt there is a way round this that would not be extremely expensive to develop and could lead to different sorts of errors creeping in.

Credit reports

If information from the three Credit Reference Agencies can be obtained, it won’t give all the debts – see How can I get a list of my debts? – but for many clients it would be a pretty good start. It should make the entry process quicker and it may also result in fewer debts being left out.

I can’t see any downside to doing this apart from the cost of the CRA data.

Open banking

The client could be asked to authorise their bank to provide the last 6 months of bank statements. The robo advice could then analyse these incomes and expenses.

Expense categorization may not be completely accurate but banking apps are making great strides in improving it. Some of the problems are caused by the Standard Financial Statement categories – if a client buys all their toiletries at the supermarket, there will be a combined figure for groceries and toiletries. There is a good case that the SFS should be changed so that it more closely corresponds to what is easily achievable in Open Banking!

Inputs from Open Banking could be very helpful for robo debt advice:

• highlights variable income;
• identifies a lot of benefits in payment;
• unless the client uses cash a lot, it may give a more accurate assessment of what is spent on food, clothes, eating out etc than the client guessing at numbers;
• shows other payments that may need to be discussed – payments to a relative, or for gambling;
• shows what debts and bills are being paid.

But Open Banking’s problems are also large.

A couple’s finances can be complex.  If the account is a joint account, the transactions will be mixed up with those from the other partner. If it isn’t a joint account, just seeing the client’s transactions may not make it clear whether joint priority debts are being paid.

Often people take debt advice because of major life event – losing their job, having a baby etc. When someone is asked to enter information, they could make some guess as to what the new figures will be. But Open Banking simply says what happened in the past and is of little help in a changed future, especially with high inflation.

And the problem that someone may have been spending less than they need to in order to pay non priority debts will remain.

I think using Open Banking to gather information quickly about the past is useful for debt advice. The client could then be encouraged to update this information so it is more realistic going forward.

But automatically generating advice based on Open Banking is dangerous. It seems like driving a car while only looking through the rearview mirror.

That is always a problem to some extent if you ask a client what they have been spending, but the apparent completeness of an Open Banking analysis may give a false sense of accuracy – the precision bias fallacy.

Some problems with the “advice”

Negative budgets

A negative budget is the term for when a client cannot pay their priority bills and essential expenses, even if no payments are made to nonpriority debts. All debt advice agencies report that the number negative budget cases has been increasing massively, now perhaps between a third and a half of clients. This is likely to get worse with sharply rising energy bills.

Here even insolvency does not resolve the fact that the client has insufficient income. In face-to-face advice, this typically leads to help with budgeting, applications for grants, benefits or social tariffs, referrals to foodbanks, and other practical advice.

Little of this can easily be done by a robo advice process, no matter how good.

One or two recommendations could be suggested to a client. But a long list (Warm Home Discount, WaterSure, DHP,  TV license simple payments scheme, social tariffs for broadband, check Turn2Us for possible grants, etc) is likely to leave even a relatively competent client overwhelmed.

The answer must surely be to not provide written advice here but to explain the problem and offer a speedy appointment where the adviser can go through ways of improving the client’s situation, prioritise what can be done and discuss debt options.

This is what StepChange’s online advice does for tax debts – the client gets a message saying:

You don’t have enough money available to repay these debts… you must contact us as soon as you can so we can advise you on what to do next before dealing with your other debts.

But for arrears on other priority debts, StepChange says that it is urgent the client offers payments to priority debts. A client may be told to offer £1 a month for 720 months to council tax arrears and £1 a month for 1800 months to their landlord for rent arrears.

This is pointless and could be dangerous. Even as a temporary holding measure a client should never be told to do this without having had a full discussion with a debt adviser.

Some clients would probably be alarmed enough at the suggestion to ignore it and take up the offer of a phone appointment. But many may disengage, especially if they have mental health problems – if that is all debt advice can offer, they may well wonder what the point is.

No income max

Income maximisation is vital in negative budget cases, but important for all debt clients. A client may need a DMP, but applying for Council Tax Reduction or WaterSure will speed up a DMP.

It may sound like a neat idea to embed a benefits calculator into the robo advice process to analyse what benefits the client should be receiving.

But that would involve a very large number of additional questions. Unless this was combined with getting information from CRAs and open banking, which would reduce the number of questions, the extra benefits questions could make the information gathering process overly prolonged, leading to people not completing it.

Sustainability and insolvency preferences

StepChange explains in their online advice that it only recommends an insolvency solution if it would take more than 10 years for someone to clear their debts in a debt management plan.

This is a policy choice by StepChange that I think is hard to justify. Why someone with a low amount of debt who would qualify for a DRO is recommended to pay a low amount each month into a DMP for say 8 years is baffling. It is highly unlikely to be a sustainable debt solution over that length of time, especially with inflation.

Robo advice on insolvency

There is a more general problem with any robo advice recommending an IVA or bankruptcy  – there is simply no way to avoid all uncertainty about the quality of the input data to that process.

StepChange assumes the disposable income is completely accurate. On this basis, someone with no assets and a DI of £85 a month is rejected for a DRO and recommended for an IVA. Bankruptcy is listed as an alternative solution, but there is no explanation as to why an IVA is preferred when there are no assets, nor of the risk that an IVA could fail.

If a DRO is suggested and this is incorrect or undesirable, it can be safely left to the DRO adviser to determine this.

But a client who has been told that bankruptcy is their best option by any robo advice can complete the application themselves without taking any further advice. This cannot be good practice – the robo advice has not asked in detail about assets and things that the client may have given away.

Similarly with IVAs. I don’t think there is a problem if the client talks to StepChange about an IVA. But the client may think for a while and then google for an IVA, ending up at a firm which is less than scrupulous about considering whether the IVA is suitable and sustainable.

I think it cannot be right to recommend insolvency, a decision that is very expensive or impossible to change later,  on potentially inadequate data. A “first, do no harm” approach should be adopted and instead of an insolvency recommendation, the client should be asked to make an appointment for advice.

The presentation of the options

There are three problems with the way StepChange presents the options.

The first is easily solved – remove all references to the options which are clearly not available… If someone is in England, there is no point in telling them about Scottish debt solutions.  Nor in mentioning an administration order for someone whose debts are more than £5,000. Including this makes things look more complicated for the client and gains nothing at all. MaPS and the FCA need to agree that debt advice doesn’t have to include unhelpful information.

The second is that an option which may be right in slightly different circumstances should not be ruled out. It is wrong to list a DRO as unavailable on a DI of £85 which may not be accurate.

The third is more fundamental. If you list a lot of details about each options, there is a lot for the client to consider. If you list only a few, then some important information may be omitted. StepChange tries hard to present the information in an accessible way but there is still a lot of it.

This problem exists in all forms of debt advice, see this 2017 report on Better Debt Advice. But to me it feels worse with robo advice where the client can’t check what you are saying, you can’t adapt your language and you can’t expand or draw comparisons with another option.

It is hard to see how good quality robo debt advice is possible

The problems with the input data mean that in many cases the I&E produced as the starting point for advice is inaccurate, not a budget the client could sustainably live on. Here even the best analysis may not produce good advice. The programmer’s term for this is Garbage In Garbage Out.

Even if the input data is good, debt cases are becoming more complex: negative budgets, the increasing number of clients with priority debts, the unreasonableness of much government debt collection, inflation…

Debt advice is now rarely about choosing between a DMP, an IVA or some other “debt solution”. Yet this is the entire focus of robo advice. The result doesn’t feel like quality advice at all, more an over-simplistic choice between a few options that are presented as the only solutions.

I think the idea of a robo process to get information for a human debt adviser is useful though. By allowing the client to enter it at a time that is convenient for them, and adding CRA and Open Banking data, a “proper advice session”, whether face to face, by phone or webchat, would be off to a good start. It would take less adviser time – and so be more efficient. And the client may find it less stressful than going through their budget line by line.

Update – StepChange’s robo advice in 2021

StepChange’s annual yearbook for 2021 was published in April 2022.

Of the 170,000 who completed a full advice session, 65,000 were via the phone and 106,000 via the online advice system.

34% of phone clients had a negative budget, much more than clients who went through online advice, where only 23% of clients had a negative budget.

StepChange says that online clients were “more likely to have higher income and surplus levels than clients who accessed the telephony channel“. Income is straightforward. However, in my opinion the Problems with the inputs to robo advice identified in this article could result in expenses being understated in the online advice channel, resulting in fewer clients with negative budgets being identified.