RPA is the corporate gateway drug to AI.

October 25, 2018

The unsavoury truth regarding robotic process automation.

There is a perspective promoted by some senior managers and vendors that robotic process automation (RPA) will enable benevolent employers to free employees of their chains in a fit of philanthropy. That’s an economical portrayal of the truth; there are a pair of dirty secrets where RPA is concerned.

Current RPA is essentially relatively low complexity but with a near-immediate return on expenditure. Generally the thought and action processes already exist but have to be mapped, and there is an immediate and measurable resultant benefit post deployment. RPA is most often a tactical fix because reengineering all of the processes and the related systems in their entirety is just too hard. There is an argument as to whether or not it constitutes ‘real’ AI since there’s not a lot of learning going on yet, but it can be argued that current RPA is AI in the same way that Expert Systems are AI. If it appears to exhibit intelligence then it seems fair to label it accordingly.

We can classify automation tools into 4 categories:

  1. Simple automation: Pre-RPA forms of automation involved Excel macros for example. Beneficial to the individual but relatively inflexible.
  2. Specific input-driven automation: Early RPA tools were programmable but demanded specific inputs which are restrictive insofar as a change in the input format would result in an epic fail.
  3. Intelligent automation: Adding NLP, image recognition and/or other machine learning techniques meant that bots were able to work with unstructured data.
  4. Self-learning automation: Here the bots observe employees to understand processes then take over the processes. They may still ask for human input when they get stuck. Such tools are still in development.

Much has been said recently with regard to the capabilities of RPA and its ability to remove the mindless repetitive tasks from people’s daily tasks. Consider the requirements within an Operations department for the repeated transcription of data from one application to another. RPA can, we are told, free staff up to work on the more difficult parts of their roles. However, everyone seems keen to insist that RPA is categorically not a mechanism to drive reduced headcount. “It will empower staff to take on more complex work and save them time!” we hear. Sure it will. If it saves your staff time, you won’t need so many of them. That’s the honest truth and the first dirty little secret.

There was the case of a large corporation’s President addressing an audience of a hundred or so senior managers from the firm. He started his speech by saying that the firm would invest heavily in AI and robotic process automation then immediately went on to say that it would make 20 percent of the firm redundant. When the audience’s eyes collectively widened he hastily backtracked and said that they would actually be “retrained”. That’s was a rare moment of candour with respect to RPA’s dirty little secret (that isn’t particularly secret). It’s all about cost savings and that means fewer staff doing that form of work. It’s about time we heard the honest truth. Certainly there are second order cost savings because if your staff get a chunk of their day back then they can be doing something else productive instead of updating their Facebook status to “Freed from drudgery”. Estimates put RPA costs at 65% less than a FTE [Grand View Research] cost so although it’s not an order of magnitude saving it’s still very significant.  The same research group are pitching the RPA industry to grow to 3.11B USD by 2025 so business is well aware of the potential.

Let’s consider the potential scenarios that could play out post-RPA deployment.

  1. If business is increasing then maybe you won’t need to hire an additional person to handle the volume increase because the current staff now have capacity to suck it up. It’s a staff cost saving.
  2. If business is flat then staff might be deployed to do something different but what if there is no other suitable work that is within their capabilities? They won’t be left idle – there will be a staff cost saving.
  3. If someone leaves the team chances are they won’t be replaced. The other staff will suck up their work. That’s a staff cost saving.
  4. If nobody leaves of their own volition at some point there will be a review of staffing and it will be apparent that there is a team out there that is decidedly over-staffed. As soon as the net over-staffing level is observed to be in the region of 1 headcount or more there will be a headcount cut. Not necessarily within the first month or so, but it will happen. That’s a staff cost-saving.
  5. If business is contracting then RPA will simply amplify the inevitable cuts resulting in an even bigger staff cost-saving.

 

Let us be clear about something: This is A Good Thing.

 

Almost all firms operate in a competitive marketplace (argue amongst yourselves regarding the various global political ideologies and their influences on market economics). Not all firms are encumbered by legacy processes and systems. New entrants have the benefit of being able to leverage the latest tool sets, technologies and processes and unless they have a radically different product they will compete on price. The biggest component of their cost base is likely to be warm-blooded, expensive, and prone to occasional lapses of concentration, particularly when undertaking relatively dull tasks.

I am told that there are around 70 firms going through the process of getting a banking licence post-PSD2 in the UK. Their distinguishing factor is not their innate ability to provide radically different banking services to gentlemen with beards riding fixed-wheel bicycles around Hoxton and Shoreditch. Their secret sauce is that they have precisely zero legacy infrastructure and can design their processes from scratch. They are your competition. They have no migration costs to absorb and little scaling challenge because they have built their infrastructure entirely on-cloud. At one point Monzo’s business was increasing at a rate of 5 percent per week; they now have over 1 million users and run around 730 microservices in Kubernetes on AWS (and are still, currently, losing money). That is your challenge.

The figure of 80 percent has been cropping up of late. The amount of time Data Scientists spend on data wrangling? 80 percent. The amount of time junior Accountants spend on data transcription? 80 percent. One wonders at the amount of time paralegals spend reading boilerplate. Clearly there is plenty of potential for automation. This may all sound like very bad news for the unskilled labour force, and so it may be. Each seismic change to our working environment has been accompanied by warnings of impending doom. However, economists put forward the Luddite Fallacy (new technology doesn’t lead to higher overall unemployment) and the Lump Of Labour Fallacy (there is no fixed amount of work within the economy) to challenge the notion that there is a causal link between technological change and employment levels.

It is worth noting that many of the processes humans still do are not trivial. Therefore, automating them may be challenging. This may account for some of the failures to achieve significant benefits in the market. For example, claims processing is in the insurance market is largely automated (95 percent apparently). Therefore, the remaining few percent are the hardest to automate so the benefit/cost is reduced.

In addition to the cost savings there are cases where RPA can do something better or augment the human capability to potentially deliver an enhanced service rather than just their capacity for throughput. They are interchangeably referred to as “co-bots” or “attended automation”. Robots make far fewer mistakes than humans when correctly programmed. Attended automation may return the biggest benefit since that way you address the repetitive aspects of each individual’s role rather than trying to automate an employee’s role in its entirety. The focus should be on looking at automating the same tasks that are done by many staff rather than automating the many tasks carried out by a few staff.  The ROI is way better where the former is concerned.

What started with Caxton’s press and Jacquard’s loom became business process automation (BPA) and has since derived the subtopic of RPA, its latest incarnation. The simple fact is that in order to survive you are best served by considering RPA sooner rather than later to maintain or enhance your competitiveness. It will either save you existing or increased staff costs, and in so doing may just save your firm from losing its edge. This brings us to the second dirty secret of RPA. It will drive staff redundancy but it won’t necessarily save you any net cash. This is because you will need to divert any savings into replacing your legacy systems and processes given that the majority of RPA is tactical. So a subsidiary contributor to Scenario 5 (above) is that if everyone else is automating and you don’t, or they have no legacy processes to work around, then your business will contract unless you can differentiate your product in some way. The savings you make from automation may be just enough to enable you to reengineer your processes and systems correctly so you are able to compete with the new entrants to the market. Alternatively, your headcount costs will move to zero - because that’s how natural selection works.

So my conclusion is that RPA is absolutely likely to reduce or limit headcount costs, although it won’t necessarily save you any money. But you have to do it anyway.

 

There are a number of firms trying to help with this issue. A couple of the biggest are:

  • BluePrism (blueprism.com) have been in the game for a long time, around 17 years or so have some history. They partner with large and small consultancies in order to build out and integrate their product.
  • Pega (pega.com) have been around even longer (35 years) and trumpet their ability to deal with the attended automation market. They have worked with some megabanks focussing on back office processing.

Additionally there are many others out there including:

 

Go forth and automate.

 

[Photo by Franck V on unsplash]