In the past weeks two independently interesting things happened. Early September, as a member of the valantic team, I attended the German SAP User Group (DSAG) annual congress. As you can imagine, this event is about all things SAP users like and also those that they rather would like SAP to change.
But, no worries, this is not the topic this post will be about.
I was in charge of our showcase there, an integrated ecommerce scenario with sales and service portions that is triggered by IoT devices and controlled by a conversational AI, exposed as a chatbot. As the IoT sensors need to be triggered playfully by a remote-controlled drift car there was plenty of opportunities to talk to parties interested in not IoT but chatbots.
We have used this showcase a few times in the past year.
So, why am I mentioning it now?
Because the type and quality of the conversations changed. It changed from ‘what is a chatbot’ via ‘how can I use them’ to pretty concrete scenarios. Scenarios that mostly cover customer service and how one can use chatbots to improve call centre services and/or to increase agents’ productivity and satisfaction.
Co-incidentally, end of September 2019 Helpshift released a report that covered how a call centre approach that combines bots and humans gets more and more acceptance. In brief, this report confirms the findings of Salesforce’s third state of customer service report that was released earlier this year.
What are the call centre challenges businesses want to overcome?
The conversations that I had during the DSAG annual congress centred around how bots can be used to prequalify incoming requests so that already busy agents can hit the road running and are less burdened with mundane requests. This way, three objectives shall be achieved:
- Service centres can resolve more issues without significantly investing in additional staff. Instead, the existing staff can be trained better
- Agents have increased job satisfaction as they are working on more interesting tasks and are getting more valued
- The organization can show to its employees that it is not about bot vs. employee but more about a “bot-augmented” employee. While this is more of a side effect, its impact should not be underestimated. As I have written before, the implementation of AI tools in an organization is a major change that needs to be supported by measures that avoid or at least decrease the employees’ fear of being replaced by machines.
At the same time customers shall get faster problem resolution.
At the end of the day it is about enabling a better customer experience through faster responsiveness and a better employee experience.
This is nothing new. I have written about this in my column and other posts a couple of times, for example here – as have other experts, like fellow CustomerThink Advisor Jeremy Watkin. Still, technology needs to prove that it can be helpful and is more than just a way for vendors to increase their share of wallet. Implementation approaches needed to get tested and best practices needed to be developed.
Now I hear you asking…
How do I get there?
Glad that you did, after patiently reading through more than 500 words.
There are two dimensions to the answer.
First, start simple and move on to more complicated topics from there. This is akin to following the Think Big – Act Small approach that I regularly emphasize upon. The objective here is to make sure that both, bots and human agents, work at their best. Bots take over the ddd – dirty, dull, dangerous – tasks (with a focus on dull) and human agents the more challenging ones, that bots cannot yet solve without further learning.
Second, have your conversational AI available where your customers expect it to be – and where you want your customers to ask for help. This is basically about deciding which communications channels are important enough that you want to support them.
Two corollaries to these dimensions are that you want to have one single conversational platform that feeds all your chat- and voice-bots and that you want to have it help your agents, too.
Step by Step – Then Rinse and Repeat
Step one is to be clear about whether you want to consider customer service only or whether there needs to be broader support of business functions (hint: there should be …). Decide upon your use cases and their respective priorities.
Once this is clarified you need to find out whether there are one or more vendors that support your use near term cases, channels that you want to initially support, and offer a satisfactory road map. If you have a platform strategy that bases upon an ecosystem around one of the big vendors, chances are good that there is more than one. If there is none, then you might have placed your bet on the wrong horse (i.e. platform). In this case, you need to reconsider your overall platform strategy. You can read here what I consider a platform. Perform a vendor selection to find the best fit and go for that one.
Once you have identified the vendor, step two is to consolidate the knowledge that is needed to support your primary, i.e. service, use cases. Answer questions like: Which questions/issues would I like to cover with a bot first? Where does the knowledge reside? Which of it do I want to make available to external users? How do I improve the knowledge base? Answering this enables you to build the first set of questions that you want bots to be able to answer.
Once you have done this, step three is to design the prioritized conversations that the bot shall be able to support to improve life for service agents and customers alike. Be sure that these cover the simple requests that make up a lot of an agent’s workload. Select training data sets to train the bots and test data sets to make sure they perform. Train them and if they perform sufficiently, let them work.
Step four then is to implement a feedback loop that does two things: First it makes sure that there is continuous performance monitoring, secondly it helps in improving the quality of the bots’ replies. This is a reinforcement learning that can be based on the users rating the solutions given or, if escalated, the agent’s choice of solution.
Once the implemented scenario is stable, it is time to venture on. Recheck and re-prioritize your business cases and implement the next iteration. Rinse and repeat. And always recheck the portfolio against the current strategy to be sure that the most valuable cases (within the given constraints) are implemented first.