Anthony Flack, Head of Digital Delivery, Santander
Introduction – Santander moved from 1,000 chats to over 20,000 chats/day, delivering over 4.5m conversations in 2020. Progressing rapidly to deliver straight-through automated processes via the app and Online Banking.
In early 2020 the lockdown brought into sharp focus the challenges and opportunities to change the way customers have conversations with the bank. How to deliver digital self-service, whilst retaining where appropriate a human interaction, became a critical requirement as branches were reducing hours and call centres were at risk of becoming overwhelmed.
Chatbots, live agents, and automation via robots, all existed prior to the lockdown and were gaining traction in many sectors including banking. In late 2019, Santander was running a smallscale pilot for Chat and Live Agent within the secure authenticated Online Banking service, exploring if it could work to replace an existing secure messaging service. This provided the advantage of the customer being authenticated. The bank had already built a prototype bot that made great use of Google’s Dialogflow and the decision was made to integrate this with LivePerson’s Live Agent platform, using IBM as an expert integration partner.
The real opportunity was significant, as a well-executed Chat delivery can provide both a better customer experience and reduce the cost of service for the bank – a true win-win. However, it was initially difficult to gain traction and to establish sufficient scale to justify the risk of moving resources off the telephone to work on chat. Leading up to the lockdown, Santander were dealing with approximately1,000 customer conversations per day through the chatbot and live agent team.
The business case is heavily based on containment, providing answers from the bot, and concurrency, the ability for live agents to handle more than one conversation at the same time, which is of course not possible on the telephone!
Chatbots, live agents, and automation via robots, all existed prior to the lockdown and were gaining traction in many sectors including banking
To launch successfully there needs to be enough content of the right type on the bot, and critically enough capabilities to offer true straight-through automations to allow the customer to self-serve, without requiring a Live Agent to assist. Delaying launch until all these automations are in place is one approach but does delay the delivery of overall value. Santander elected to press forward, hoping to learn, evolve and improve quickly whilst adding the automation concurrently. Inevitably this will impact initial customer experience but did prove to be the right way forward.
Scaling 1,000 to 20,000 conversations in a week – lockdown
The pilot was working well with good CSAT scores and approximately 50 percent containment in the bot, with a team of 3 experts continually reviewing quality reports and making content changes. The bot had been integrated to the mobile app opening another important channel. Then lockdown arrived, and the challenge was to immediately provide a public chatbot to act primarily in the Q&A space. Public means not authenticated, which significantly restricts the range of things a customer can do, and inevitably leads to customers being directed to the phone – not a good outcome!
The public site bot was launched, and to support speed to market the LivePerson bot was utilised alongside a huge commitment to train over 3,000 branch colleagues to act as Live Agents for the public site. The bot was given an identity – Sandi and a tone of voice was developed aligned to the 9 brand behaviours familiar to all colleagues
It very quickly became apparent that without authentication there were too many customer journeys that resulted in dead ends, so Santander quickly adopted a form previously utilised for PCI compliance. This allowed certain transactions to be completed, through partial authentication from the customer to the Live Agent, either sat at home or in a branch. This proved immensely successful as, whilst not fully automated end to end, the customer’s requirements were fulfilled in one interaction. Soon over 20,000 conversations per day were through these channels, and volumes have remained at circa 16,000 per day ever since.
True automation opportunities
The elixir is of course to meet the customer’s need with full automation, whilst always retaining the option to engage with a person if appropriate. Prioritisation and decision making are data led, by a two-pronged approach. Firstly, identifying the transactions that generate the greatest number of calls, and then validating the current demand for these queries in chat. To achieve this, Santander embarked on the delivery of what became known as secure dialogues. A secure dialogue uses the power of chat AI to identify intent, combined with questions aiming to disambiguate the customers intent, to reach the final identified need – for example ‘card damaged, replacement needed’, is then integrated to APIs to complete the customer request, without the need for human intervention or manual back office processes. To date7 such dialogues have been implemented and many more are planned, including solving a complex technical challenge to introduce two factor authentication into the chat.
The interaction of the chatbot and the mobile app is an interesting one, and there are experiments running to test the suitability of delivery through secure dialogues as opposed to developing more costly complex solutions through the app user interface. In one example, the ability to report a card lost or stolen was piloted as a secure dialogue then introduced as an app-based experience - this led to cannibalisation but is proving very successful, with an expected usage in excess of 400,000 per year.
There are four major areas of focus as we move forward. The first is to exceed 80 percent containment on all bots, the secure bot is already at 90 percent. Secondly there is more work to do to match customer demand with the availability of colleagues to provide the human touch, potentially 24/7. We are looking at the inclusion of new channels, both internally for example Financial Assistance and also technologies such as Apple Business Chat and WhatsApp. And finally, the plans to add more truly automated end to end secure dialogues.