The six hidden requirements for chatbot success
For large B2C businesses, reducing costs through increasing the share of service interactions handled without human intervention is a top priority. Research suggests that customers are on board with this – Forrester finding the share of customers self-serving via web increased from 67% to 74% in 2014, with the percent using virtual agents increasing from 28% to 55% over the same period.
Deploying virtual agents or chatbots offers greatest scope for increasing the overall share of self-service from current levels. A survey by Personetics found that 14% of financial institutions believe chatbots are 'ready for prime time’ with another 62% stating they believe chatbots are an exciting opportunity which will become a reality in 1-2 years. Over 60% expect over 25% of current conversations to be handled by a chatbot in the relatively near term.
My own conversations with a number of financial service companies supports these findings. A key learning from these meetings was the scale of ambition. While it might only be 25% of conversations in the near term, many are aspiring to an 80:20 model over the long term with the majority of contacts currently handled by agents resolved by chatbots (augmenting self-service enabled via the website).
Only exceptions, which are typically complex in nature, and specified sales opportunities would handled by a human. And the aim is to meet these targets while protecting or enhancing NPS.
Resolving the potentially conflicting goals of experience improvement and cost reduction is a tough but not impossible challenge. It requires the chatbot experience – and the broader self-service experience - to be of very high quality. That will not happen until there is a fundamental change in understanding on how chatbots work and what they need to be effective.
At the moment the focus is primarily on the technology, that is important – speech recognition, biometric authentication, natural language processing, search, predictive modelling and next best action decisioning will all play a role. But other, less obvious factors will also determine success.
1. Chatbots rely on content, not just technology
Chatbots are like icebergs and attention to their hidden elements will determine whether businesses deploying them achieve their goals for customer experience quality and service staff reduction, or not. Without addressing the below the surface components, the experience will be poor, complaints will rise and the number of contacts requiring human intervention will increase – costs will rise rather than fall and NPS will fall rather than rise.
The visible component of chatbots is the technology. The hottest topic of 2017 in both customer experience and technology media is artificial intelligence (AI) with chatbots and conversational commerce cited as a primary use case. But the success of chatbot initiatives – at least in the 1-2 year period that most companies believe they will become mainstream - will be more to do with the breadth and granularity of knowledge that the solution can access than technology mastery.
Most of the effort in making chatbots a success will come in the less exciting area of content creation. Creating these knowledge assets is a significant investment that must be factored into cost equations. And in the technology-driven excitement about AI, that may have been missed.
Figure 1: Chatbot Iceberg
The critical importance of content to chatbot success was admitted (albeit in a backhanded way) in a recent interview with Chris Nicholson, CEO and co-founder of deep learning company Skymind. He lauded the value of AI in helping businesses understand from structured and unstructured data what is really happening; also predicting what will happen; and optimising decision making. But when it came to chatbots, he was highly sceptical.
“Chatbots are over-hyped, and the tech isn’t there. Maybe it will be in a couple years, maybe it won’t. The chatbot interfaces being implemented now are just taking Web pages and breaking them down sequentially as interactions. Not satisfying. Users don’t get much value there.”
Yes and No.
As the CEO of an AI business, Nicholson understands the technology aspects well. Currently natural language generation tools are limited to turning tabular data into commentary, example use cases being business report creation or weather forecasts. Amazon has created the Alexa prize to reward the creation of socialbots that can converse intelligently on news and culture. And this shows how far this technology is from becoming effective – certainly not within 1-2 years. Even then there will be the challenge of having sufficient depth and breadth of conversations available for the AI to learn from.
Nicholson is also right that the current experience – breaking web pages down and serving the content sequentially – delivers a poor experience. A chatbot raises expectations as to what will happen then frustrates by simply acting as a gateway to FAQs, which are frequently based on what the business thinks customers will ask rather than a data-driven assessment of what they really want to know.
But his overall conclusion - that the potential for chatbots is limited - is not one I agree with. Technologists focus on AI in isolation, CX professionals focus on getting the best out of the technology currently available in conjunction with other assets to meet the needs of their organisation and its customers. Nicholson’s view ignores how costs can be reduced and the customer experience enhanced by deploying current AI capabilities to serve richer content - whatever is required by the information needs of the conversation.
Any conversation will only be as good as the information that the AI engine can serve.
With any customer contact, the conversation is like a decision tree. Once the customer’s desired objective is clear, reaching it requires the exchange of information – the chatbot asking the customer for information (e.g. for identification, verification, clarification and qualification) and then presenting information in return, either content or data from the customer’s account (such as balance, recent transactions).
This will then shape the next step, which may be asking the customer to make a selection (e.g. from three different types of mortgage which they want information on, whether the information provided has answered their query or not) and so on.
This is giving rise to a new role of conversation designer requiring skills in AI (NLP, search), content generation, data mapping and decision flows. Most importantly any conversation will only be as good as the information that the AI engine can serve. This beneath the surface part of the chatbot iceberg is where significant effort is required.
2. Required content is already being generated, it just needs to be repurposed
Much of the content that chatbots need is already being generated by customer service teams. It may be provided verbally in telephone conversations, via web chat interactions or in emails. What is required is for this content to be repurposed - translated into a form that the chatbot can serve once its requirement is triggered.
It is easier to start off with translating what is already in digital form. For example, the content from every email sent to customers could be reviewed for inclusion in a knowledge base of answers that the chatbots can reference. The same applies to online conversations in digital channels.
The aim should be to create once and reuse thereafter. Achieving this requires close links between whoever is creating the content - service teams - and those responsible for digital touchpoints, typically marketing.
3. Use analysis of why customers are contacting you to focus content curation and creation
The logical starting point is to create content that will have the greatest effect – where content can mitigate the reason for calling and the volume of calls is high. Understanding why people are contacting your business requires root cause analysis that draws upon contact profiling (who is calling, about what, when, how many times), customer journey analysis (looking at sequential contacts across channels) and Voice of the Customer research (incorporating customer surveys, email content and both online and offline conversations).
Not all contacts are candidates for self-service. Where the customer has a lot of emotion invested in the outcome of the interaction – when there is a problem (a mistake has been made, potential fraud, etc.) or reassurance is required – are best handled better by a human. The same applies where the complexity is high – exceptions that fall outside business as usual and require a high level of explanation, clarification and information exchange.
Then there are those interactions which suggest the customer may have additional needs or may be at risk of leaving. For both cross-selling and retention, human interaction is likely to be more effective. These are the c. 20% exceptions that businesses would like humans to continue handling.
For both cross-selling and retention, human interaction is likely to be more effective.
But there will be many contacts where the involvement of a contact centre agent offers limited value to either the business or the customer. The most obvious example is when customers attempt to serve themselves online but cannot, so they call the contact centre. Understanding what they were trying to do and why they couldn’t achieve their desired outcome is critical to fixing the customer experience and reducing the additional costs incurred. If the call occurred because the information sought was not available on the web site or not accessible by a chatbot, the fix will be content creation or curation (as described above).
If it is because the customer feels more comfortable picking up the telephone than searching on a website or chatting to a bot, the solution is to create content – video tutorials for example - that reduce the effort required in learning the desired behaviour.
Not all contact reasons can be mitigated by content creation. There will be some that arise from process issues – customers initiating contact because they cannot complete an online application, access information or have been caused to seek reassurance. This is a more complicated resolution as it requires process redesign and potentially IT intervention. Increasing self-service requires those responsible for technology, process and content to work closely together.
4. Develop served digital channels to support content creation and migration
Some customers are unlikely to switch from telephone channels to self-service or chatbots without an intermediate step - digital channels where assistance can be provided via an online text-based conversation. These channels can be both synchronous or asynchronous.
An example of the former would be webchat where customers converse in real time with agents through the website. The latter would include messaging and social media. Synchronous is ideal if a customer is stuck trying to complete an online process. The advantages of asynchronous channels is that customers can leave a message and not have to waste time waiting for an answer, they know they will get one back within 20 minutes or 2 hours or whatever the service level is.
These channels also offer benefits to the company – webchat channels typically enable an agent to handle two conversations at one time, with this rising to six with messaging or social channels. The duration – elapsed time to resolution – is typically longer and these channels are not well suited for complex issue resolution with multiple back and forth between the agent and customer. But waiting times and abandonment – more important determinants of customer satisfaction – are lower. Also agent productivity is higher due to concurrency (handling multiple conversations at once) and the flexibility asynchronous platforms provide for handling peaks and troughs.
Some customers are unlikely to switch from telephone channels to self-service or chatbots without an intermediate step - digital channels where assistance can be provided via an online text-based conversation.
Digital service can also be delivered more easily from offshore locations. The communication is text-based, so there are no accent problems where customers cannot understand what the agent was saying or make themselves understood – an issue that caused many businesses to repatriate customer service functions.
Finally, these conversations provide a readily source of text data for determining why digital natives - the most willing self-servers – are getting in contact. This provides insights to support the migration of customers from voice-based contact to digital assistance to full self-service through capturing feedback as to how to the digital self-serve experience can be improved or better marketed.
5. Migration requires creating a channel pyramid and then flipping it
One way to think about this is to think of your contact profile to be a pyramid, then flip it, as per Figure 2. [Click to enlarge]
At the top would be chatbots. Initially these would resolve only a small element of total queries, acting primarily as an online IVR and routing customers to the digital assistance team best qualified to handle the interaction. These digital assistance teams – web chat and messaging (whether via social media or an owned platform) are the middle layer, supporting the migration to full digital interaction at a lower cost than voice. At the bottom of the pyramid is voice contact - the highest cost and highest volume channel.
With root cause analysis and the creation of relevant online content, chatbots will resolve more and more queries - the aim being that eventually they handle the majority. Digital channels continue to play an important role in supporting traditional offline customers become digital. With the voice channel shrinking to just handling the 20% exception-type contacts outlined above – complex interactions or ones where empathy and reassurance are critical to the customer having a good experience.
6. Reduce risks by piloting with close supervision of experienced agents
Chatbots are an immature technology, so the risk of guinea pigs having a poor experience is high. Some forms of AI can be piloted internally, for example next best action engines can be trialled with agents rather than customers directly, this is not really possible with chatbots, so a different strategy is required.
The best place for locating such a trial is an innovation hub within the contact centre. This will typically have the right mix of skills and ethos to make it work.
As with any technology, there needs to be significant user testing prior to any deployment. Next the use of a chatbot should be transparent and, ideally, driven by customer self-selection. (Usually there are enough early adopters willing to act as guinea pigs and they will be more tolerant of quirks.) The scope of operations should be limited in first instance – online routing is a good place to start – and grow over time.
Finally, close supervision will be required. Much in the same way that a new agent is supported on a one-to-one basis when taking their first calls, experienced agents will need to watch bot conversations closely to ensure they are proceeding as intended and step in should the conversation go off track.
* * *
The potential for chatbots to enable customers to serve themselves without human intervention is high. Achieving the levels aspired to will take time as chatbots are an immature solution so a gradual approach will be required, both in terms of scope and migration of customers across channels.
For those seeking to deploy within the next 1-2 years, content will be critical to success. Analytics will be necessary to identify what content is most relevant then it will need to be curated or created. And knowledge management is an ongoing cost that needs to be factored into business cases.
Jack Springman works as an interim director, helping businesses to deploy digital technologies to deliver strategic objectives and desired customer outcomes.