TL;DR My “Applied AI” series shares practical examples of how specialised AI systems can augment human workers. We can automate over 70 % of calls received by a call centre. This episode will show you how. You can jump down to watch a recording of a conversational AI resolving requests of a banking customer.
Current State: The Interactive Voice Response Systems (“IVRS”)
Remember the last time you called your bank? You were most likely greeted by a robotic sounding voice: “Please enter your account or credit card number”; “Please enter your tpin”; “If you like to get more information on our recent promotion XYZ press 1. If you like to … press 2 … if you… press 9”; “Thank you for waiting; your call is very important for us and will be answered in the order it was received.”; “[BAD MUSIC]”.
94 % of callers interviewed said that the IVRS of their bank is “not useful at all”. The general perception seems to be that they are only deployed to keep the caller busy till an agent becomes available.
At the same time, call centre agents we interviewed pointed out that: “We receive a lot of simple requests for information. I type the client ID into the system and read it out. I prefer to spend more time on providing advice and dealing with complex requests.”
Indeed a major bank’s head of contact centre told us that 70 % of callers asked for their account balance. Considering the same banks’ contact centre operation has close to 400 employees, that is a lot of people reading out account statements.
The obvious issue here is that today’s IVRSes are only triaging calls and collecting information. They do not answer the customer’s questions or resolve even the most simple requests.
Future State: The AI-augmented contact centre experience
How could a specialised AI system augment contact centre agents? To illustrate this, I recorded a call made to one of our conversational AI systems (demo, not a live solution).
The AI resolved 4 of my requests and up-sold me an instalment plan. When I signalled emotional distress, the bot passed me over to a human agent. The whole process took ~ 2 minutes, and my call was answered immediately.
Conversational AI features deployed
The first thing you might have noticed is that the system greeted me by name. It did not ask for my account number or any other verification details. This is made possible by a risk scoring engine that selectively allows actions based on authentication information available. In my case, the system recognised that I was calling from the mobile number registered with the bank. This gave me a personalised greeting and the option to be directly connected to the premier service. After that, the AI used biometric voice pattern recognition to confirm my identity. This allowed me to ask for my account balance and pay off my credit card bills. If I would have requested a riskier action, the AI would have asked for my passphrase and might even have challenged me to repeat a random voice OTP.
The second feature you will have noticed is “intelligent solicitation“. The AI only asked me for information I did not already provide. Hence it immediately executed my request “make full payment on my signature credit card using my current account” while asking for more information when I only said, “make credit card payment”.
It is imperative to “teach” your AI agent when it is time to hand over a call to its human superior. AI research made enormous strides in providing systems with contextual awareness. We are still several years away from a general AI that can create new & effective patterns on the spot when the situation demands it. This is why it is vital to have a clear handover strategy. It can be as simple as “if the bot does not understand the customer more than two times” or as complex as “if the customer seems angry or frustrated”.
It is further critical to keep human agents on staff and not even think about to replace them by AI fully. Live project deployments show over and over again that the combination of human & machine yields the best results.
The business case for an AI-augmented customer service
An unaugmented human can manage 96 calls a day (5 min/call no breaks for 8 hours). Advanced, cloud-based AI systems can handle several thousand calls per minute. Call centre managers report that 15 – 30 % of their total call volume could be taken by the AI from day 1. With investment in continuous dialogue and AI skill improvement, I find that this number can go up as high as 68-76 %. An AI-augmented call centre can consequently handle four times the number of calls of a traditional operation, can it?
It is crucial to keep in mind though that the 24-32 % of calls that are still routed to human operators are those of higher complexity. The call centre manager has to expect a higher average handle time (‘AHT’) per agent as a result. In my experience, expecting capacity to double is a good rule of thumb for an initial business case.
The next equation you have to look at is cost. Reality is that most conversational AI installations require a significant upfront investment to train the bots and integrate them with existing systems. On a small operation of 5-10 agents based in a low-cost jurisdiction that will unlikely generate a positive ROI – if cost-saving is the only driver.
I have seen SME customers build a case around customer delight & scalability, however. Imagine your clients could receive service 24/7, immediately and with consistent quality. Your call centre always answers even if the staff is on medical leave or busy with other clients.
AI can also lessen the burden of having to hire many agents in a short time frame when business growth explodes beyond expectations. A very “happy problem” I wish you to have soon!
I trust this example illustrated how conversational AI provides leverage to existing human capital in a contact centre. With AI as a force multiplier, even small teams can service more clients, with higher quality than their better-staffed competitors. AI changes the game from “who has the biggest team” to “who has superior strategic agility & tactics”.