With businesses losing $1.6trn annually due to poor customer service (according to Accenture research), it’s a pressing issue for many brands.
Can artificial intelligence militate against some of this loss? There are a number of different technologies in the market. Here’s an overview…
Multiple choice chatbots
Chatbots are becoming increasingly common. Though they are continually lumped in with AI technology, their chief aim is simply to reduce the complexity of service navigation by using the format of a conversation.
They offer pre-defined multiple choice options and little or no free text input. In essence, they are a form of information architecture, which also can provide customer information (e.g. order details).
Examples of this kind of chatbot can be found in many different industries such as retail (Very) and travel (Expedia).
From a UX point of view, long forms with multiple fields, which can be a daunting prospect for users, may be made more accessible and less taxing when chunked in a chatbot.
This is perhaps the logical progression from current online form best practice, where one field at a time is presented to the customer, before moving on to the next.
Chatbots may also lead to websites and apps that shun a deep, nested menu experience, and choose to surface information through ‘interacting’ with the user.
Such technology offers time-saving potential in customer service. It could be integrated into live chat functionality (see bot-assisted agents below).
Further reading on chatbots:
- What are chatbots and why should marketers care?
- Why chatbots are an important opportunity for retailers
Most consumers will be familiar with virtual agents, which have been on the market for a number of years.
These are often no more than a way of searching an FAQs section. Though a stock image of a call centre agent implies some personalized service, the language processing involved is less to infer meaning in a request and more to match keywords and phrases with FAQ content.
As savvy consumers have become more familiar with these solutions, the realization that one is dealing with a glorified search functionality can be underwhelming.
Robert LoCascio, CEO of LivePerson notes that “Customer satisfactions on [these] traditional front-end bots is below 70%, which is really low for customer care and sales.”
This well-established style of virtual agent gives an interesting lesson in the dynamics of automated / intelligent service. The problem to be solved is how to accurately set customer expectations – offering supposedly intelligent service that ultimately disappoints must be superceded either by sufficiently accurate and error-free AI, or human service supported by AI (in the background).
Much of artificial intelligence relies on a ‘human in the loop’ – some form of human feedback or assistance that helps an algorithm learn or moderates its output.
Bot-assisted agents are the next step from virtual agents. Human agents quality control a bot’s suggested answer. LivePerson calls this a cyborg model and say it delivers 30-35% gains in efficiency (presumably time savings).
This style of assistance garnered much publicity when KLM implemented it, using a system developed by Digital Genius. Agents are supported with suggested answers based on historical use data, including that collected on the customer, to give personalised answers quickly and correctly.
Social media manager Karlijn Vogel-Meijer says this approach gives “the best of both worlds – a timely answer, a correct answer, and a personal answer. The best of humans and the best of tech.”
The screenshot below, via Nvidia, shows how this works within the Digital Genius interface. Confidence levels are provided and agents can choose to personalise or approve automated messages.
The automation threshold is set at 90% in the example below, meaning when the algorithm is confident enough, it will answer automatically.
In late 2016, KLM was taking an average of five questions a minute through Facebook Messenger alone (13 per minute in peak times). The airline already has c.250 social media agents, and so any solution to deal with rising message rates without continuing to add agents is obviously of great value.
A similar solution is provided in the market by DeviceBits, which can be used to optimize content surfacing in a knowledge base, or to provide predicted answers to an agent.
Afiniti also provides a similar product, which routes customer calls to the appropriate agent, based on analysis of that customer’s data.
Real-time emotional analysis
Cogito is one of the most interesting companies providing AI customer service. The company’s website claims to increase customer and agent satisfaction, reduce customer effort and attrition, with resulting gains in revenue and savings.
The software does this by flagging customer emotions, intentions, and social signals during calls in real-time. If you’re wondering what this looks like, see the screenshot below.
Cogito claims that working with healthcare company Humana, it saw a 28% improvement in customer satisfaction (likelihood to recommend).
However, though the sample was impressively large (100,000 conversations) and used test and control groups, I couldn’t find further detail on how strict the experiment was.
Were the same agents used? How did the customer enquiries and demographics differ over each test and control group? Were they conducted at the same time?
Though a 6.3% improvement in customer resolution is not to be sniffed at (including a 30% reduction in dead air), some suggest that such technology is not yet sophisticated enough to interpret individual or even regional differences in conversation cadence.
To this point, MIT Review quotes Rosalind Picard, a professor at the MIT Media Lab who has pioneered emotion tracking and says, “Many New Yorkers practice a ‘high interruption’ style. Interrupting can thus be likeable and build rapport with them. But the same behavior with some other callers could be seen as rude.”
The endpoint for artificial intelligence and customer service could be considered to be a fully automated solution. Due to the current limitations of natural language processing, it’s unclear when this might be possible and indeed if an automated experience lacking empathy could ever provide the customer satisfaction a human agent can.
Though home and personal assistants such as Alexa are currently very sophisticated, error rates would be way too high if such systems were implemented on the front line of customer service, particularly over the phone.
However, it’s worthwhile remembering that home intelligent assistants may affect a company’s product or service offering, whether they like it or not.
Companies have to decide how their own information and services will be presented to a customer through a home/personal assistant. This may mean some sort of HTML markup or further integration with functionality such as ecommerce.
The customer’s experience via a personal intelligent assistant should be treated like any other touch-point, offering an interaction consistent with the brand.
Content curation (e.g. merchandising)
AI is being used on the softer side of service, such as merchandising and personal shopping online.
North Face has employed IBM’s Watson technology to test an interactive tool that asks jacket shoppers fairly basic questions (but with some free text input) and then narrows down the product range.
The question is whether this experience is particularly helpful. North Face doesn’t have the biggest product range, and many people arriving at a website have already done some research more effectively on the wider web as to what sort of jacket they need.
Apparently the 60-day trial saw 50,000 consumers use the service, spending two minutes longer on the site than without the tool. 75% said they would use the tool again, though it’s debatable how many visited the website out of curiosity, and how many made a purchase confident that the algorithm had not missed a cheaper or more suitable product.
This was a quick walk through several AI customer service tools. As ever, in this tech-rich world of marketing, those that offer pragmatic functionality – where the AI lives in the background (i.e. bot-assisted agents) – seem the most likely to succeed in the short term.
Longer term, who knows the fate of the human agent. There is certainly plenty of impetus for automation, but for now it can seem like a cool gimmick (see the Henn na Hotel).
If used on any scale, AI has to be proven to definitively ease the customer journey and ultimately grow revenue.
Discover the world of AI-powered marketing at Econsultancy’s Supercharged event in London on July 4th, with speakers from ASOS, BT and Just Eat.