Heads of customer support have to constantly act in a field of conflicting business targets: cost reduction vs. improved NPS. Competing targets often reflect themselves in different performance indicators. Example: longer call handling time translates into increased cost, reduces support accessibility as well as customer satisfaction. Shorter call handling targets though lead to increased call transfer to back office or even field service, hence dramatically increasing overall call resolution cost as appropriate measures cannot be initiated as early as possible in the chain of support. In this area of conflict, the introduction of learning systems can substantially contribute to target fulfillment. But which are the key challenges project leaders have to keep in mind when starting their first ML projects?

Machine Learning formula of success

Prior to being able to define operating targets, business management needs to have a clear understanding of the technology and its implications. Machine Learning optimizes a calculable objective, this is what the success of machine learning is all about. Decision-makers have to undergo a mind change though: operating targets need to be transformed into a calculable target function. And what is even more important: the target function needs to be internalized and set as a best practice example all across an organization by business leaders and stakeholders. This is the basis for successful implementation of machine learning.

ML: Bare-knuckle efficiency requires meticulous preparation

ML algorithms hold the potential for highly-efficient optimization. That is why they are so powerful but it also requires a much more stringent definition of success and aspired business objectives when compared to non-ML projects. When listing of influencing factors like shorter average handling time AHT, increased first time right (FTR) or net promoter score (NPS) used to be sufficient in the past, it is merely a starting point in ML projects today. The algorithms’ success depends on many additional factors. In order to be able to formulate business objectives for initial projects, responsible management should pay attention to the following aspects:

1. As a basic prerequisite full transparency regarding the objectives and capabilities of machine learning need to be established
2. Existing data needs to be re-evaluated. Which data can be optimized by ML? Only high-quality data can help to achieve optimal results. Find more about it here.
3. In a next step, the influencing factors like AHT, FTR or number of recallers need to be identified and quantified.
4. The correlation of additional factors and influencing factors need be established. Example: in case AHT is considered as single factor for optimization, escalation to the next support level will be singled out by the ML algorithm, in order to achieve the shortest handling time possible. Only when the factors cost is considered by the learning system as well, the ML algorithm will not automatically decide in favor of escalation to the next level. Reason: escalation means higher cost as a more expensive expert needs to be involved in call resolution.
5. As a consequence. a calculable target function needs to be established.

You will find additional tips for how to start projects with learning systems successfully here.

Success means not be afraid of delegating single decisions

Management in charge should be aware of the fact they have a very powerful tool at their disposal. It needs to be controlled differently than more conventional technologies though. With regard to business objectives management has to accept the fact that single decisions need to delegated to a learning system. When it comes to detailed decision-making like “delegate after third repair attempt” or “always try repair xyz” ML algorithms are simply faster and more efficient. Management in charge can experience the delegation of detailed decisions as a loss of control, a fear which needs to be addressed in open manner. It is the management’s responsibility to formulate an all-encompassing target function. With these pre-requisites being fulfilled, the influencing factors can have direct impact on the learning system, hence management can then focus on relevant decisions and provide strategic direction.

ML success needs to be measurable

ML is a valuable ally to management in charge when it comes to achieve business targets. But all success achieved needs to be clearly monitored and evaluated, as both influencing factors like AHT and FTR can contribute to business objective optimization with every ML cycle. By regularly applying simulations to historic data management in charge can attain a higher level of confidence with regard to the successful operation of a learning system. Furthermore it may be advisable to also establish a real-time monitoring system with regard the target function during the early phases of an ML project. As a consequence not only a higher level of security will be generated but also the success of the learning system can be measured, providing a higher confidence level with regard to the results generated by the learning system, hence facilitating the delegation of single decisions with a clear conscience.

Well-prepared for ML projects

Substantial improvements can be achieved through ML especially in telco customer incident resolution.  This holds especially true since manifold data as well as clearly measurable success factors exist. By a clear definition of specific and quantifiable operative targets, learning systems will grow to be a powerful tool in the hands of businesses, helping them to agilely optimize their objectives.