What I learned about customer contact strategy from reading 10,000 service webchatsby
Pulling out stories from the likes of Sky, HSBC, British Gas and Tesco, Andrew Moorhouse examines the death of live chat and how companies are starting to rethink their entire customer contact strategy.
Choose your channels wisely…
In the summer of 2017, to much fanfare, a £16bn UK media giant introduced a new customer contact strategy; one that was prophesised to strip out 30% of OpEx costs from the call centre via headcount reductions. Using a cute piece of behavioural science, “Press 1 for a faster way to be served” the IVR deflected hundreds of thousands of inbound calls to an AI bot on WhatsApp or secure Chat-in-an-App.
This was the future. Conversational AI with seamless transfer to a live agent; and this enterprise was among the first to embrace the potential. Those leading the change were even known to utter, “We’ve gone Async.”
So, what happened to the live chat strategy? Well, it was deactivated four months into the pandemic and never turned back on.
So, what went wrong?
It was a modern-day version of the emperor’s new clothes. With all the buzz and hoopla to be first on the AI scene, nobody dared to speak up and challenge the new direction they were heading in. Async, it turns out, wasn’t the future, and the 1990s teen-boy-band-inspired thinking wasn’t working with the customer base.
You see, Sky is an exemplar of driving self-serve and has arguably the best self-help and online self-serve journeys. It’s impressive. Yet, when you try to cram the FAQ into a hybrid chatbot that quickly escalates to a human, you have a problem. Queries were not being resolved or even answered. There was huge repeat contact, corrosive levels of customer dissatisfaction, and a WFM operating model that didn't align with the projected inbound traffic. An omnishambles so severe, the only fix was to deactivate the system.
“Bye Bye Bye”
The death of live chat is an important story that's not being told right now. It’s not in the interest of any tech vendor to propagate this message. And those at the helm of live chat failures have been swiftly exited and their story also remains untold. Until now.
This unashamedly long-form article will go deep. A deep, evidence and case study-based approach. For the avoidance of doubt, this isn’t a thinly veiled whitepaper. It’s quite the opposite. The purpose is to challenge conventional, vendor-led thinking; tactfully provoke the current ways of working, and offer insights into how some companies are starting to rethink their entire customer contact strategy.
A health warning
Before we dive into the good stuff, firstly, some positioning. This piece pulls out stories from Sky, HSBC, British Gas, Tesco, and many more. There’s no malintent at all. No NDAs have been harmed during the making of this article. Indeed, I've used my consumer experiences to showcase what I see happening in the marketplace. So, with the health warning out of the way, let’s look at some of the reasons live chat might not be living up to the hype.
1. Don't discount your self-serve journeys in favour of AI
You are a consumer of a major UK High Street bank and you’d like to block your ATM card after misplacing it on a train. Seems like a simple request, right?
As we see below, it seems that some chatbots just don’t chat. In the rush to bring conversational AI to market, enterprise firms purchased the “out of the box” solution and built basic query bots. You’d hope that a transactional request such as “Block my card” would be understood. Yet sadly, as we see, the bot suggests a huge disambiguation list of 10 options and then goes to prematurely terminate the chat! Utterly ridiculous.
A failure here drives repeat contact and corrodes customer satisfaction. I persisted and via human live chat, I was able to block my card. The total asynchronous chat duration was 30-minutes. In contrast, take a look at the brilliant, simple, low effort option from Barclays for this same journey. It’s possible to block a card in a one-minute using a self-serve form. Even quicker in the App (website pictured as I couldn't take screenshots from the app)!
The leadership team at Barclays has resisted the allure of conversational AI so far, in favour of brilliant banking app design and some excellent humans in Liverpool. Though curiously, once logged into the Barclays website, it’s clear that conversational AI/hybrid chat is on the CX roadmap.
So it seems that one influencing factor in live chat failure is the use of perfunctory and rudimentary AI; and a desire to automate things that perhaps should be left alone. But what else is causing the death of live chat?
2. All intents are not equal… or resolvable
To have an informed opinion, I read 10,000 LivePerson transcripts to truly understand what was going on with resolution rates. Prior to this activity, I didn't have an especially strong viewpoint. But now, I can categorically state that (within a banking context), live chat is an absolutely terrible option for medium to complex customer queries. For certain customer intents, it’s horrible with no resolution possible.
There are two reasons for this which I'll explain, but first, take a look at the diagram below; this maps out the chat resolution rate against the NPS for each customer intent.
Why is it so bad?
We could easily spend an hour digging into the minutia on this diagram alone; but what’s very clear, at the bottom left-hand side, you have some highly complex but low volume intents with close to zero resolution and incredibly poor NPS. This is an agent mindset and capability issue, for example consolidating an ISA (a UK tax-free savings product) almost always ends with zero resolution as the agent pushes the onus of resolution back to the customer, "You'd need to check with them; it's not us." The mindset of Powerless to help is discussed further below.
What you also have is an incredibly large volume of intents in the bottom left, where there's an inability to resolve via live chat due to legacy processes and policies. For example, fraud reporting must be done over the phone; likewise for requesting a direct debit indemnity.
Looking a little deeper you can see that the lack of resolution is the single biggest driver of NPS detractors.
35.5% of NPS detractors are due to process, policy, and procedural issues.
Sorry, you've come through to the wrong channel. You'd need to go to the branch or phone; there's nothing we can do.
Sound familiar? The mantra of being Powerless to Help is the calling card of the disempowered and disengaged call centre worker. Hat tip to Ted Mckenna for coining this phrase. When 35% of your NPS detractors are due to your inability to resolve the customer query via the live chat channel, then it's time to take a serious look at those legacy policy issues.
I'd also take a serious look at the credibility of the data you have on your live chat containment rate. The CX director for the above operation was convinced that just 4% of all queries were routed to a separate channel; the real number is closer to 18%.
So an inability to resolve certain customer intents via live chat, combined with terrible AI is a contributing factor; but what else has caused so much trauma to contact centre operations. This next answer will result in me getting no Christmas mail from the AI tech community...
3. Don't believe the hype
When a £5bn UK retailer won a prestigious contact centre award for innovation and bringing the first AI-powered live chat offering to market; the transformation team were humble and admitted that when they looked at the numbers, their AI bot mostly cannibalised traffic from the FAQ page. Year-one inbound call volumes reduced by just 2%. It's since grown to an 8% reduction in inbound traffic, but take a look at any vendor sales material. They all promise a minimum of 30% call volume reduction. So go as far as 50%.
Lies, damn lies, and chat vendor statistics
Again, it's the emperor's new clothes syndrome. That 30% OpEx is so tantalising that nobody challenged this at Sky. I even had one of the Gartner 'Magic Quadrant' tech vendors claiming he could achieve an 80% call volume reduction. He was insistent it was possible. Let's say we disagreed to agree and then disagreed some more.
Let's run the numbers: take a large-scale enterprise operation. 1,200 staff and 10-million inbound calls per year. Repeat telephony contact is probably around 10 to 15% depending on how you measure it. Your staffing bill is going to be north of £12-million at about a 400-second AHT. Are you seriously suggesting that an AI chatbot will reduce inbound calls to 2-million? Or even 5-million calls. Or even strip out 3 million calls (30%)? Just not happening. Ever.
Over-egging the concurrency pudding
The concurrency myth is one of the biggest culprits in the spurious vendor ROI package. Here's one example of a live chat concurrency distribution curve. I recall chats were capped at 4 and the mean was 1.8. Have you ever tried to hit an average of 3 in your department? It's hard, because you'll always have a bulk of newly inducted (and poorly performing) people at way below 2; so you need those 4s and 5s to be coming through to get close to a centre-wide average of 3. Skilled banking teams hit 1.8 to 1.9 on average.
Almost every vendor will promise an average concurrency of 3, and then make wild extrapolated claims that this equates to a 66% reduction in AHT. I believe the concurrency fallacy is the final piece of the jigsaw that explains why live chat has failed at Sky, EasyJet, VirginAirlines, and many more.
On the topic of EasyJet, they opted (eventually) for a pure FAQ bot without the live-human escalation. It doesn't do much for customer satisfaction or channel shift, but it does help their WFM demands as they, along with many other airlines, were facing a deluge of live chat that they simply couldn't service.
4. Scrap your socials
There is a curious trend, however, that removing live chat has forced customers to seek other online chat avenues. The most damaging is perhaps hijacking innocuous marketing adverts on Facebook. Sky routinely sees upwards of 300 customer comments per advert; Easy Jet is the same, but the clear (unfortunate) winner has to be British Gas.
This well-intentioned advert garnered 7,300 comments and it's still live. Link at the end of this article. The comments section is not a pretty read and there are ~ 2,100 customers reporting that they have no heating via FB comment spamming...
The analysis speaks volumes about the challenges faced by any B2C enterprise firm. What's worse though, is it appears that many marketing-led UK "social teams" have no system access, no 'teeth' to actually own anything; whereas a direct message (DM) on Facebook Messenger reaches a different team, utterly disconnected from your prior online conversation. It's the worst practice on all fronts.
There's one final piece to cover again. It's not the fundamental case behind the death of live chat, but it's fun and worth prodding a little further.
5. Chatbots that don't chat
The banking example we saw upfront had a bot that failed to understand a very simple request to block a card. Here we have a leading retailer using a proprietary Facebook Messenger AI piece of software. The Natural Language Understanding (NLU) engine sits at the heart of all intelligence. Though in this case, there doesn't seem to be much intelligence at all.
Understanding the NLU
I don't profess to be an expert in this field at all but it's a space where every tech vendor claims to have the 'best' NLU. And I'm very skeptical of anyone that claims to have out of the box functionality. You should be too.
Vodafone in Germany spent an entire year building the best Telco NLU engine. They had 400 agents responding to live chats, and for an entire year, there was a team of 100 analysts that tagged each intent and utterance to improve the effectiveness of Vodafone's NLU.
"Hey Erica, how much did I spend at Walmart last month?"
Bank of America allegedly has spent $100m to $110m on similar activities on developing their "Erica" voice bot, so there really are no shortcuts to achieving this level of understanding. Though coming out of Amazon, Google and Facebook are some NLU models trained on billions of interactions, with crazy names like "Hugging Face", Bert (Google), and RoBERTa (Facebook).
In my view, the only true way to test any vendor claims on intent classification is with a "bake off." (an English idiom to describe a contest between companies to win a contract). But let's wrap this section up and applaud EE's human approach for really getting live chat right. In a world of canned empathy, this is an examplar of a great social media response:
6. The resurrection of live chat
Despite the lack of publicity, the challenges with live chat aren't a rare or unique phenomenon. Virgin Airlines, EasyJet, and many more have since deactivated their conversational AI and live chat platforms and hunkered down back to using telephony as the primary contact option.
The issue is not artificial intelligence. It's human intelligence. This isn’t about bad tech. It’s about badly informed customer strategy decisions.
Here are some thoughts on how to resurrect live chat and ensure you don't suffer the same WFM challenges as Sky and others:
- Aggressively employ the paradox of choice principle. Customers do not get the best resolution rates by following a 'Choose your own adventure' approach. Log a complaint? Do it via a contact form. Want to consolidate an ISA, then let's do this via a co-browse session. Have three separate questions; here's a skilled advisor (typically 6 to 8% of inbound contact will have more than one intent).
- Triage, Triage, and Triage again. Triage is the most under-discussed element of conversational AI; because nobody fully understands the minutia of NPS and resolution per intent. We're not talking about a generic report showing 80% containment. To take decisive actions and definitive automation decisions you need to go deep.
- Define your conversational AI strategy. Leaders are talking about AI first and conversational banking, but what does this mean? These three objectives are similar but very different: Do you want: 80% of all contact to be digital; 80% of contact to be automated; or 80% via conversational AI? Define your strategy first. Don't let live chat be a bolt-on. Starling Bank had the benefit of a clean sheet to start with. They have one contact option, via the app, executed brilliantly. Their page is worth a look for FAQ/deflection inspiration.
- Unify your customer contact taxonomy. This is not sexy. Or easy. But will probably give you the best foundation for the future. Most organisations have 200+ IVR contact reason labels; 150 call drivers; 100+ AI customer intents; 40 to 60 agent drop downs. Unify all of these into the simplest list possible and use them as the #1 list for all customer contact. Until you do this, you cannot start designing optimal customer journeys.
- Don't believe the hype. I think this one was drilled home, but don't get sucked into the tangibility bias of techno-determinism. Omnichannel platform tech vendors love to sell omnichannel consolidation platforms. It's their raison d'etre. But customer contact strategy is best defined by you. You have an architect for designing a home. Have an architect for your contact strategy, one that isn't selling you a tech solution! AI isn't the silver bullet here. Keep that for your simplest transactional queries and triage the hell out of complex, messy, and infrequent intents. And in this mix, don't discount self-serve journeys. There are probably many, many well-designed self-serve options your customers don't even know about. Focus on promoting these as opposed to ruining them with an AI chatbot.
- Customer journey design isn't a post-it note exercise. In a data-driven enterprise, it's time to trust the data. Customer experience by committee isn't necessarily the best approach. What's the optimal, single best channel for delivering the best resolution, for us and the consumer. Again, one for your chief customer contact architect; not the entirety of your marketing department...
Conversational AI is here to stay - growing at 30% CAGR and the demand for more AI will hit your operation, from all sides. When you have those tricky journey design and prioritisation/automation conversations, hopefully, there's sufficient evidence in here to push back and dispel those tangibility biases that the c-suite love so much.
I truly hope you've enjoyed reading this.
- British Gas Advert: https://www.facebook.com/britishgas/posts/10159711794180649
- LivePerson Transcript Analytics supported by Dr Danica Damljanovic
This article adapted from a piece that was originally posted on LinkedIn.
Conversation Scientist | Management Consultant | MBA Lecturer | Founder of an advanced analytics firm using AI to eke out human performance gains.
My deep area of expertise is the codification of human behaviour. I deliver high-impact people analytics; and am driven to determine what top performers do differently.