How to annoy customers with virtual assistants and web chat.
(Courtesy of #Sky on #LivePerson)
In a recent web chat with Sky (who I dislike a lot), even with my low expectations I was surprised at some pretty basic mistakes. For such a big brand I think they are worth looking at to understand what not to do. Scenario: I want to cancel; I really don't want to call because the voice recognition in their IVR is so bad and the wait time every time you call is pretty poor
1. Deliberately understaff
I am not saying they do, but a huge brand with possibly 4K+ FTE in customer service, can't make mistakes forecasting traffic and scheduling staff, every day, all day, it's just not statistically possible. And on voice always playing an IVR message saying something like "we're sorry, we are experiencing high demand...", every time you call is insulting our intelligence. My speculation ... they deliberately undercook their staff estimates to save money.
With regard to the web chat, we will cover that later but it was pretty clear they weren't staffed up.
2. Use a Virtual Assistant that doesn't help the customer
If a VA improves the customer experience, bring it on. In this exchange time in the Virtual Assistant was 5 minutes plus! What on earth can take that long that a virtual agent can do? (more on that later)
Here is the takeaway. If the VA is unintelligent (as in this case) and it cannot solve the problem, route to a human once the customer intent is understood. In reality, that is the only question that is necessary to route to the right team. Seconds!
3. Use a Virtual Assistant that is incapable of understanding the answer to the question it asked.
The dialog:
Even without AI, it is pretty easy to understand this with basic NLP but with Gen AI it is a fraction of a dollar to split this into intent and products.
The VA then asked four questions to clarify the original request. One of the UI basics was it asked which products with buttons for the answer, but you can't select multiples. Classic, basic usability mistake. This VA tech must be extremely poor.
4. Make the Virtual Assistant say things which are proved not to be true
VA: "You might have been asked these questions before, but it's just to speed up the process when you get through to the next advisor"
It just isn't true, in fact it probably cost Sky money in wasted agent time.
5. Get the Virtual Assistant to ask for known and irrelevant information
Because one authenticates on Sky.com to get into mysky (my account), my identity and information is known to the VA. It proved this by showing me some. We then went through a sequence of it asking the following information:
6. Personalise badly
Over 5 minutes into this conversation the VA says "Wow you've with us for an amazing 3 years, Thanks so much for your loyalty to Sky."
Key points
7. Make a customer spend many minutes with a Virtual Assistant
This should be:
It should be in and out in 30 seconds.
Recommended by LinkedIn
5 minutes!! Really?!! Did that Virtual ASSISTANT assist the customer in any way?
8. Make the customer wait 8 minutes to connect to a human
Putting this in perspective, it took 13 minutes to connect me to an agent, so it took 8 minutes to queue (please refer to point 1 again) once the Virtual Assistant had finished. By ANY SLA this is poor. But there is something possibly worse.
At the beginning of the interaction with the VA the chat was assigned to a named agent, but the VA takes 5 minutes to finish. I know (from experience in the past) LivePerson have a sophisticated predictive engine to ensure a chat is assigned to an agent as fast as possible. Have Sky just neutered this because of assigning the chat too early or was it the tech? If so, that is just silly because there is no good reason.
9. Once with a human, ignore the customer’s request
OK I am being a bit unfair. Customers are now savvy that the way to get a better deal is to threaten to cancel. So, to be fair, it is OK for the first interaction, it is retentions after all. BUT trying 3 times to tempt the customer, (not 1) when the exchange takes 4 minutes is just dumb. A customer trying to get a better deal will bite on the first try. Notably if this was on a phone call, such an exchange would be less than a minute AND the agent would have detected the customers firm desire to cancel and moved on quicker.
Why is this bad?
Well apart from the obvious, chat agents are measured by all kinds of metrics including conversion/saves.This chat agent was forced to stay in the chat when they could have just cancelled me (it took 2 minutes) because I was a no hoper. They could have spent time on a more qualified opportunity which would be good for them, good for the customer and good for Sky.
Instead, I had to read the desperate last pitch to keep my Sky Q box for £5/month to get what I get on my TV without it. Made Sky look desperate in my eyes.
10. Answer each customer response slowly
Average response time in this chat was about 1 to 2 minutes. In a chat 30 seconds seems like an age. With things like canned responses (if intelligently used) response can be speeded up.
In this case I will bet each of the agent responses where canned, so there is no excuse. I will take a pretty good bet that the agent was juggling 3+ concurrent chats. At that level of concurrency, it is impossible to be responsive. Measures I have done in the past show that customers respond slower so concurrency can be balanced with the customers' speed of response. The trouble is that Brands have been sold the pup that chat is cheaper than voice because of concurrent conversations. The reality is it is unbelievably hard to achieve that. I remember the best chat team I have come across achieved an aggregate concurrency of 2.33, but they were specialists. If you scale it up generically then you are going to be in the 1.5+ range. At that rate when you include cost of software vs. telephony etc. it is the same or sometimes worse. I often saw a chat AHT of 40 minutes for a contact type that would take 8 minutes on voice. So is chat a money saver or a convenience for customers or suitable for specific use cases?
11. To stitch yourself up, offer an NPS survey after all the palaver.
The chat ended after 24 minutes! What is my feeling about the chat? Not great. If I had got through to an agent immediately, I would just been a bit "meh".
My speculation is the agent will be blamed for my poor score (because I know NPS, there is no point in giving anything other than 0, 9 or 10) so I gave a 0. When it had nothing to do with the agent, it was entirely due to:
Sadly, I would guess this agent was in some outsourcer somewhere (as you can often detect a non Sky employee by their robotic style), so It could be that all these metrics reflect negatively in their next quarterly review (but in reality one chat isn't going to stand out).
Yes, I totally agree that NPS surveys should be randomly presented or offered on everything. I am not saying don't do surveys, just design the experience properly first.
12. Get the wrong people to design the user journey
For me this is the key point. Somewhere in some (possibly) design by committee world, some people thought all this was great. There are two key points for me:
The key thing here is if one gets through to a real Sky advisor the experience is usually great, therefore the issue is how you get there and who you speak to.
From a technology perspective my subjective view is this is poor, it's not the tech as they usually buy the best. Their voice recognition in IVR is a case in point, you spend time for it to misunderstand you then you get asked all that again when you get through to an agent.
With the chat, I am not sure LivePerson will be proud of the unintelligent use of the Virtual Assistant (which I am guessing is theirs), they always used to coach clients on best practises, even down to conversational design in chats. What has happened ?
Therefore, I will put this down to "just because you can, doesn't mean you should". Having witnessed a Sky person talking over one of the cleverest mobile CX insight analysts I have ever met, before they entered the mobile market, it really doesn't surprise me. Is this a case of the wrong people making the wrong decisions
These opinions are my personal opinions and not those of my employer.
Commercial Operations Manager at Webio Ltd
1yIf the 1st point is correct, that will lead to staff attrition which will then validate the initial action and become a self propagating cycle
Automating actuarial modelling processes in Life Ins' at WTW
1yHaving spent an annoying 5 minutes with the so called virtual assistant, being asked my marketing preferences (especially as being told about new shows is a bit silly if I have just cancelled), I just received this. Looks like the VA can’t store what it asks 😂
Co-Founder and Chief Strategy Officer (CSO) at Webio Ltd
1yCc Andrew Charles Moorhouse this will sound familiar to you …
Helping Independent BPO's grow and exit at a higher value.
1yHi Simon hope your well as always a succinct description,my view for what it's worth another example of shit for brains when it comes to deploying tech good or bad without any real thought strategy for the process or for the customer.if retention and good service are real drivers then perhaps reducing these wait times and having relevent interaction may just stop people leaving in the first place. Also messages that say on a minute by minute basis due to high volumes of calls blah blah doesn't fill me with confidence that the brand is on a good trajectory for customer service.