Things to Consider Before Implementing AI in the Contact Centre

Rafael Cortes of Foehn offers some precautionary advice on the implementation of AI, particularly where human voice is concerned.

Over recent years, there’s been an extraordinary amount of opinion and speculation on the subject of AI in the contact centre. It often feels as though hype has taken over.

For many business leaders, eager to capitalize on the benefits, it’s a challenge to cut through the marketing-fuelled noise to build a case for investment. Even respected research consultants are issuing conflicting opinions on the technology and where it’s going.

Within that debate, perhaps the biggest dilemma surrounds the relationship between voice and AI in the customer experience.

The balance between the two is critical. Market surveys that confirm voice is still the preferred and most important channel of communication will also inform us that the biggest opportunity for increasing productivity lies in the deployment of AI, particularly in the areas of process automation and speech analytics.

The harmonization of these objectives has become one of the biggest challenges in the design and management of contact centres today.

The temptation to jump into AI is great, particularly for smaller businesses where automation and increased productivity are welcome antidotes to tight budgets and limited resources.

These businesses stand to gain the most from AI but, paradoxically, they also lack the time and opportunity to assess fully the two sides of the voice/AI equation.

For these businesses, and others taking the first steps to improve contact centre productivity, here are some issues worth considering before investing in AI.

Don’t Be Misled by Imitations

Over recent years, there’s been a growing trend amongst vendors to label applications (and their algorithms) as ‘AI-powered’ when, in fact, they’re not.

Algorithms that simply predict outcomes have been used in applications for decades, but some vendors are now taking advantage of the AI prefix to give these applications a cutting-edge appearance.

For example, the knowledge management systems that we have all used online at some time simply combine a predictive algorithm with a logic-based, decision-making model to deliver an answer.

Nonetheless, it’s common to see vendors re-badge this type of application as ‘AI-driven’.

True AI is based on machine learning where an algorithm ‘trains’ itself to improve its predictive capability by cycling data input and output, on a trial-and-error basis, at unimaginable speed.

When variances between output and actual become negligible, the predictive qualities of AI, running in real time, gives the impression of human thinking and judgement. But it’s not.

AI is limited by availability of data and the quality of that data. As usual, ‘garbage in, garbage out’ still out applies.

Look to Improve Existing Processes

With this reliance on existing data, machine learning often works best in the contact centre when it improves an existing process rather than introducing a new one.

For example, your CRM system can flag up a customer who is within one month of end of contract and therefore at risk of changing supplier.

Machine learning, on the other hand, can go much further, identifying the probability of losing a customer within six months of end of contract, given the number of complaints made, the value of the purchase, the model of the product and the age of the customer.

Scenario two could make the difference between a customer lost and a customer retained.

The Case for In-House Development

It’s important also to understand that AI isn’t plug and play. The machine learning algorithm must address the very specific requirements of any given scenario, and this takes time.

That said, the big vendors like Microsoft, Google and IBM have released cloud-hosted, machine learning toolsets that are simplifying AI development for the contact centre, particularly when it comes to chatbots and speech analytics.

These tools have made in-house development of AI more feasible for many businesses and this has its advantages.

For example, an in-house team will benefit from closer contact with processes and data and closer control over the monitoring and reporting required to tune the application and allow machine learning to evolve.

Quality of Data Is Paramount

Before embarking on an AI project, irrespective of in-house resources, it is imperative to consider the quality of data available. This is the biggest challenge faced by many developers.

Typical problems with data, such as spelling mistakes, missing fields and differing formats, all serve to confuse the machine learning process and necessitate the time-consuming process of data cleansing.

Data can be skewed by inconsistencies arising from variations in service levels or human bias imparted by agents reporting the data.

Ultimately, the data you need may not even exist digitally, like the qualitative information that is often locked inside the agent’s knowledge and experience.

An AI solution is not feasible without appropriate data. It’s a prerequisite for any AI project.

Consider RPA First

Typically, the contact centre is awash with simple but repetitive and time-consuming tasks where true machine learning is beyond requirements and over-elaborate for the purpose. This is where robotic process automation (RPA) is a better option with fewer challenges regarding data quality.

RPA is a software ‘robot’ that simply mimics human actions, whereas AI is the simulation of human intelligence by machines.

The technology works particularly well when giving the agent control over certain processes that would otherwise require a back-office intervention.

For example, an agent may need to validate customer details whilst communicating via a chatbot. For this relatively simple and repetitive task, a stand-alone RPA application would provide a rapid response within the chat session without the need for the agent to break away to a back-office CRM, where delays could put the interaction at risk.

This is typical of how RPA is associated with “doing” whereas AI and machine learning are concerned with “thinking” and “learning”, respectively.