|Homepage / Publications & Opinion / Articles, Lectures, Preprints & Reprints (post 2002)
Problems With Partial Solutions
6 Mar 2008
We live at a time when technological change is accelerating and outpacing our corporate and human abilities to subsume and fully understand the consequences. Perhaps the most important aspects, invisible to many, are the creation of new business models, customer expectations and behaviours. Call and contact centres have never been in the vanguard of change, and do not enjoy a good reputation on the customer side of the line. Here we address a slice of this problem set and posit some partial solutions whilst also presenting a limited set of predictions spanning the next 20 - 30 years.
The customer view of the contact centre industry is as wide and varied as the number of people who have had to use them and the depth of their resulting frustration. That is not to say that contact centres don't meet a very definite market and customer need, and that they haven't improved markedly over the past decades. But, just like lawyers, call and contact centres tend to get a bad rap! On the user side there is definitely an impression that the people responsible for their design and implementation have the same mindset as those responsible for corporate IT and Security Departments. In short; they don't appear to eat their own dog food!
There is nothing quite as sobering as trying to use your own products and services at a time of some personal emergency or stressful business situation. Things never work as well as you thought they would, or designed them to. If only the designers of supermarket check-outs, bank-teller systems, EPOS terminals and more were to work on them for a few weeks they would do a far better job! I have often thought that if I could be God for a day I would condemn all architects to live in the last building they designed for a couple of years before they are allowed to design another one! The point being; in general designers of everything never hang around long enough to witness or experience the SNAFUs they have designed in.
Unfortunately, this is all compounded by a society and technology landscape that is speeding up! At one end we have the very old who can remember the arrival of TV and the telephone, and may therefore have difficulty using modern equipment, and at the other we have the youngsters who have always had everything - mobiles, computers, games machines - the lot! Their view of life is entirely different and leans toward the 'get a move on, get with the program' intolerance that questions the way we do everything. At the same time Sci-Fi and Hollywood present society with technology visions that we can't quite live up to yet. Engineering for such a spread is a real challenge, but the good news is that there is a lot we can do to increase customer satisfaction and corporate effectiveness generation-by-generation.
It is not unusual to find the division of customers and income following a 5:30, 25:30, 70:30 ratio set. That is; the top 5% of customers bring in 30% (or so) of all income, the next 25% of customers are responsible for the next 30% (or so), and the remaining income (~30%) is gained from the remaining (~70%) of the customers. But when the disposition of company resources, in terms of call centre focus, is analyzed it often tends to be more or less uniform across the entire customer base. The 5% customer base is very important and tends to have expensive problems, and moreover, no one wants to disappoint or lose them! They are influential and hard to get, but a really good investment in terms of RoI/head. In contrast; the bottom 70% tend to have simpler problems and their disappointment of loss is far less damaging.
Grading customers and service on the basis of call-in number range, email, and TXT address is the easiest and first step. The second is the appropriate allocation of people and machines as depicted in Fig 1. Then there is the provision of the 'shortest, and most appropriate, hop to a solution algorithms' that can be engineered. At one extreme this can be dial into your personal problem manager or agent for the 'very rich' thro to the talk to this computer for 15 minutes before you get any human attention for the 'less valued' customers. Irrespective of range though our focus has to be on the fastest route to satisfaction.
Two key statistics in terms of this process are:
Mean Time To Respond and Mean Time To Complete
MTTR and MTTC
Unfortunately people tend to focus on the mean (average) numbers measured. But as my Professor of mathematics used to say: Averages tell us very little! The real information is locked into the shape of the statistical distribution dictated by their Moments - the chief one of these being the Variance. So in an ideal world the averages would be as small as possible, and the clustering would be really tight - ie the variance would be very small.
In order to realize the best possible systems today we really have to jointly understand the system design and performance abilities, the customer problem set, and the current and future technological landscape/capability. These we can broadly categorize as follows:
Some Critical Problems
The HAL 9000 Problem
Even 15 years ago speech recognition programs were recognizing single human utterances with accuracies 1 - 2% better than humans. Today that number is routinely 3 - 5%. So why can't we have a HAL900 style conversation with a computer? Unfortunately recognition does not impart understanding! Until a computer has both cognition and context awareness it cannot decode the simplest of phrases:
"Let us pray"
"Let us spray"
These all sound the same, and to date, it is only humans that can sort on the basis of the knowledge that we are in church, in the garden, or in the kitchen! And there are millions of such linguistic and comprehension ambiguities that have to be addressed. Similarly, on the speech synthesis side; despite being able to artificially generate acceptable human speech, we cannot get yet get a computer to utter:
"I love you"
or any other phrase with irony, sarcasm, or true affection for the very same set of reasons! Context and comprehension are a very necessary component that cannot be neglected or by-passed by continuing to polish recognizers and synthesizers.
To date there is only one major 'computer cognition' program on the planet and that has been under way for over 20 years, and has only recently amassed sufficient data to provide some useful output. This problem is way beyond exhaustive solutions - it has to involve some defined basis (see www.cyc.com) augmented by real-time situational learning.
So when might we see something we can use with real customers beyond basic command and control? For acoustically quiet environments a good deal can be done today! But for even the most modest amounts of background noise system performance tends to degrade rapidly toward the unusable, or customer annoying! My personal estimate is that really useful contact and call centre technology capable of dealing with a mobile phone in a noisy office, car, street, or home is still at least a decade away. And the full HAL9000 experience? I wouldn't hold your breath - it could be another 30 years!
A Peak to Mean Solution
Company call and contact centres suffer the same peak-to-mean demand problem experienced by power, telephone, mobile, and computer networks. Demand is never even, always driven by correlating phenomenon (strange attractors) from outside the industry, and is a cause of customer dissatisfaction and operational inefficiency.
In the electronics and networking world this problem is well understood and addressed by clever routing and prioritizing algorithms that hide latency from customers. Nevertheless, for real and non-real-time services alike it still costs $$. For power grids it means lots of spinning metal on standby waiting to service peak demand. A high peak to mean ration always means expense!
For the contact sector the problem rapidly migrates beyond all efforts to automate and translates into more bodies. Is there anything that can be done with the human workforce, which, as ever, constitutes the major cost? So far the most-subtle solution resembles that adopted by the power companies for electricity supply with big power stations supporting the base load and smaller stations that switch in to address the peaks.
So a base staffing of FTE people can be employed on a shift of rota-basis to address the base-calling rate, whilst other call centres and casual/part-time home workers can be brought in as demand dictates. Such a solution affords a significant improvement in quality control - and migrates systems into a new and wholly dynamic mode. Significant improvements in customer satisfaction as well as other key metrics have been realized to date.
Whilst it is currently impossible to talk to a computer in the strict sense, a solution is possible in the guise of an 'Amazing Turk' (aka the chess playing automaton of von Kempelen circa 1769). By placing a human operator between a computer and the customer a series of scripted questions and responses can be 'clicked through'. This gives the customer the illusion of:
The reality is that the human agents click through the same process or fault tree menu with the same human recorded, or computer generated, responses.
If a point is reached where the tree is exhausted and the problem is unresolved, the computer-generated response can be of the form:
At this point the operative who has been engaged in the click-through comes on line, and as if by magic, is fully briefed and up to speed. And the customer is now far more accepting of a foreign or even difficult accent or dialect.
The agent now works through the problem and at the same time generates an extension, and/or new branch or set of branches to the tree that are also shared by all other agents.
This hybrid solution isn't perfect, but in a way it might represent what will happen as we gradually edge toward the HAL9000 ideal.
Like the businesses they support and serve, the evolution of the call and contact centre is far from over. Indeed, we might wish to consider that there is more to come than we have seen hitherto. But for sure we will see some very new diversions and creations as business models are driven to change by technological, demographic, general population acceptance, globalization and competition.
As an engineer my take is that we can design out a good deal of the applications considered essential and 'top-dog' today, and will have to become more fleet of foot as the pace of change accelerates. To this end let me close with some forecasts: