As highlighted in a previous post, organisations are seeking to deploy chatbots to reduce costs to serve while improving the customer experience through increasing service availability (24/7, 365 days a year), reducing call waiting times, shrinking resolution times and standardising quality at a high level.
The long-term service objective most large B2C businesses aspire to is the majority of customer interactions – information requests, account updates, simple transactions – being resolved automatically through self-service or chatbot channels, with only complex exceptions handled by humans.
Online services and mobile apps mean many of these interactions are already self-served, so the more important question is what percent of interactions currently handled by humans could be switched to bots?
At this early stage of chatbot maturity, that is difficult to answer, but the likely impacts can be bounded by reference to similar technologies.
What could the maximum impact be?
The closest analog to chatbots is robotic process automation (RPA). RPA automates the repetitive tasks handled by humans in back office roles – the activities created by multiple systems and lack of integration, copying and pasting from one system to another being the simplest example.
RPA is quite mature as a technology and is gradually becoming more sophisticated with the incorporation of predictive analytics and decisioning technologies. Together these two technologies enable more machine-based judgement, reducing the need for human decision-making in processing activities.
Based on observed results and the increasing scope of what can be automated, industry experts expect intelligent automation to remove the need for 40-50% of full-time employees (FTEs) executing back-office processes.
Industry experts expect intelligent automation to remove the need for 40-50% of FTEs executing back-office processes.
This can serve as an absolute maximum for the FTE savings that chatbots could deliver. It is very unlikely that bots will have a greater impact on front office processing than on back-office processing. Front-office processes are less susceptible to standardisation – customer behaviour can be influenced but not controlled.
Historic experiences of poor online service - starting a journey digitally but having to complete it via the contact centre – means trust is low and scepticism around new channels is high. The natural desire to avoid stress, not to mention the cognitive effort required to learn a new behaviour, means customers will be less compliant than employees in changing behaviour and front-office staff will be less susceptible to replacement by bots than their back-office colleagues.
What could the minimum impact be?
In the first instance, chatbots may be a substitute for other digital channels rather than replace human contact. Rather than login to a mobile banking app to check their balance, customers who are already in Facebook may ask a Messenger bot to provide the same information.
Also, it is not beyond the realms of possibility that chatbots could actually increase the need for contact centre agents. It has certainly been the experience with some companies that customers with mobile apps call the contact centre more than non-mobile customers.
How much of this increased contact stems from negative factors such as greater visibility of errors versus positive ones such as higher levels of engagement is open to question. In the days when a mailed monthly statement was customers only way of checking their transactions, banks could often correct errors prior to the customer noticing – much harder when customers access their accounts several times a week.)
It is not beyond the realms of possibility that chatbots could actually increase the need for contact centre agents.
Still it is possible that the more easily basic needs are met, the more customers will want to deepen the relationship and that reduction in some contacts could be offset by increases in others. Those with long experience of the contact centre industry are keen to point out that its demise has been predicted since the Web 1.0 and it is yet to happen. Why will it be any different this time around?
Best guess - split the difference (plus or minus a few percent)
Most longstanding businesses have a customer-facing operating model – people, process and technology – that relies on the first to make up for deficiencies in the latter two. For these businesses, digital transformation remains a dream. The costs and risks of replacing legacy systems are too great relative to the benefit of doing so, until such systems are judged not fit for purpose by regulators and stiff fines become the norm, or customers do the same and defect in droves.
These legacy systems embed processes that were never designed with customers’ desired outcomes in mind so humans are needed to act as integrator and translator. By comparison with digital natives such Amazon or Facebook, the need for human contact has largely been engineered away.
But the role humans are largely being asked to play – the manual integrator of system silos, translator of the internal to an external view – is in large part a robotic one. Activities are often repetitive and low value, require empathy but little opportunity for creativity and are remunerated accordingly.
The latest technological developments mean that data access, prediction calculation and action execution can be automated to a level where only exceptions need to be handled manually.
Availability, productivity and effectiveness are also closely tracked. The nature of the role and the degree of performance monitoring being contributory factors to the high levels of turnover typically experienced in contact centres.
At a recent symposium, the futurist Brian David Johnson argued that if a machine takes your job, then it probably sucked. If we take this logic in reverse, then jobs where there is a high level of staff turnover may be better handled by bots.
What will make the difference this time is that advances in the predictive capabilities of artificial intelligence (AI) combined with the maturing of automation tools have together created a tipping point. As a recent article in MIT’s Sloan Review highlighted, completing an activity requires four things – data, prediction, judgement and action.
The latest technological developments mean that data access, prediction calculation and action execution can be automated to a level where only exceptions need to be handled manually. Judgement is also being increasingly automated using either business or probabilistic rules applied to accessible data available and predictions delivered by AI solutions.
Key to this is natural language processing (which enables accurate prediction of customer intent) which will enable customers (eventually) to achieve what they want through Alexa-like devices, so requiring even less effort than taking out a mobile, finding the number of the call centre, navigating the IVR and waiting for an agent to be available. (Though in the short-term, executing such requests via web chat or social media interface rather than voice will be the norm.)
But the role of AI and automation technologies in identification and authentication, matching intent to pre-prepared conversation snippets, accessing data from multiple systems and executing customer requests should not be ignored.
So the reasons for thinking it will be different this time have solid foundations which means that the best guess we can make for what will transpire is probably somewhere in the middle of the 0-50% range with a margin of error, perhaps 20-30%. This fits with a recent Personetics survey which found that over 60% of financial institutions expected over 25% of current conversations will be handled by chatbots in the near term.
For any organisation looking to invest in chatbot technology, targeting a 20% reduction in contact centre FTE would be both meaningful and a reasonable starting point for the business case.
Jack Springman works as an interim director, helping businesses to deploy digital technologies to deliver strategic objectives and desired customer outcomes.