With the convergence of machine learning and growth in popularity of customer-centric approaches driven by the wider financial industry, chatbots have naturally popped up as a cost-effective and user-friendly bridge with customers, both old and young according to the ‘2018 state of chatbots report’. But not everyone is doing them properly.
“Financial services have some truly awful and cheaply built chatbots,” Alberto Chierici, CPO of the insurtech startup SPIXII, told bobsguide.
The real challenge appears to be the ‘chat’ part of chatbot – the human interaction. The fast quantitative intelligence is here already, but dumbing them down to be believable, contextual and empathetic is a far harder task.
But for now, chatbots can perform relatively simple yet, quick functions that reduce business costs and provide greater convenience for the customer. Here are four arrows in a chatbots quiver.
1. Behaviour analysis
This is perhaps the most compelling business case for chatbot technology in insurance and one that SPIXII’s Chierici was quick to point out: “If data informs design and experience-building, with the right KPIs and measurements, insurance companies can achieve great things,” he says.
“The SPIXII chatbot replaces the generic Q&A forms, which is nothing fancy, but we collected behavioural data through the user-bot interaction and that informed our machine learning algorithms to optimise the type of experience, such as how long to wait before asking a question or whether to use a formal greeting or a more casual one.
“By optimising the journey to receiving a quote we achieved a 100% uplift, without optimising. From quote to buy, we achieved an uplift of 20%”, says Chierici.
2. The omnichannel straddler
A buzzword associated with every digital banking stack of the last couple of years, an omnichannel experience means the backend remembers where a customer left off with a particular task. A chatbot can straddle all digital customer channels, collecting, retaining and retrieving data to expedite this process at the customer’s leisure accessible 24/7 often as easily as messaging a friend.
3. Customer onboarding
In a similar vein, chatbot technology combined with voice recognition can be used as a sophisticated form, using a retrieval method to verify credentials and information as answers are given, leading customers to an instant response to their query; no longer will we hear ‘press 1 for accounts’. In effect, cutting human resources and boosting operational efficiency.
4. Compliance and transparency
Of more pressing concern with GDPR fast approaching, is how chatbots might be used to generate easily accessible audit trails. As a source logging data and conversations, it makes a lot of sense to cheaply employ it to this end. It also provides the customer with greater transparency into how their requests are being dealt with, and generally improves customer satisfaction – a trait that has been lacking in insurance.
It is still early days for chatbots and, as with the above example, conversations of any length quickly show its interactive limitations.
Even the more recent rival-sister chatbots, Siri and Alexa, often have a hard time computing simple commands, replying with “Sorry, I didn’t get that. Searching ‘how do chatbots make use of natural language processing’”. The algorithms, and how they retrieve data from the natural language processing libraries, can become unstuck without clear signposting.
However, advances in machine learning and neural networks promise to rectify that. The generative model, by learning from thousands of unique sequences of ‘Context-to-reply’ can quickly find algorithms to work out which replies best fit and generate a response; over thousands of trials, runs and simulations it can hone in on accuracy. Pattern-based heuristic elements can also leverage AI’s pattern spotting capabilities to ‘remember’ how effective past ‘context-to-reply’ conversations went and choose the best one.
The future of insurtech’s chatbots: Front end charm, back end efficiency
For SPIXII’s Chierici the future chatbot must stick to the Value Meaning Engagement (VME) framework. “Regarding value,” said Chierici, “they must save time and money for both customers and insurance companies. They must be easy to use and well-suited to their context (meaning) and enjoyable to interact with (engagement)".
Ultimately, for Chierici, the future recipe for a mass adopted chatbot in insurance will be able to say yes to these questions: is it useful and effective? Is it easy to use? And lastly, will the customer enjoy it?