Is AI Falling into the IVR Traps?

Two decades ago, the “Service Guru,” Tom Peters, did an interview with me entitled, “Push, 1, Push 2, Push 3, Push your customer over the edge!”

IVR systems, designed to more quickly bring customers to the right place, instead, have been channeling them to “service hell.” They are also used to divert customers away from humans to reduce this most expensive component of service by providing them with easy answers to simple questions.

Does this sound at all familiar to recent conversations about AI? Aren’t AI systems in danger of being snared by the service traps that have maimed the intent, use, and reputation of IVR systems?

AI has the same issues as IVR, but the consumer is now hypersensitive due to the service hassles of the COVID-19 pandemic era.

The lessons and implications of the IVR experience translate perfectly to AI. So, let’s quickly review the pros and cons of IVR use.

The Good About IVR...

• You can authenticate customers based on the numbers they are calling from.
• Provides easy access to first-level answers to the top six or seven issues that customers call about (consumers will tolerate up to seven picks).
• When stable menus are used, the customer quickly learns which option to pick.
• When “barge-in” is allowed, customers can get to desired selections quickly without listening to all the options.
• If you publish the top level of the menu wherever you provide the 800#, customers can easily select the right option in advance, rather than forgetting what option they wanted.

The Bad About IVR...

• Long front-end messages, including “Your call is being recorded...” (we all know that).
• Long list of complex selections where you think you may want #4 but then try to listen to #5 and get confused.
• No ability to go back without listening to a second long message.
• No ability to barge-in when you know you want selection #4 you have to listen to #1 - #3.
• Open-ended invitation – “how can I help you today” – which is misunderstood 75% of the time and repeated multiple times, causing rage.
• No/murky opportunity to get to a representative.

AI has the same issues as IVR, but the consumer is now hypersensitive due to the service hassles of the COVID-19 pandemic era. Our National Rage Study found that 43% of consumers have yelled at a service person (or IVR) and 9% have even sought revenge against a company for poor service.

Tech vendors all espouse how AI will enhance effortless response and personalization, but we find that the most prevalent causes of rage are still:

• Escalating to a human and...
• Listening to long messages.

The Cost of Bad IVRs and Bad AI

The cost of bad IVRs is exactly the same as the cost of bad AI: customer struggle and market damage.

If you compare the revenue and word-of-mouth (WOM) cost of abandonment or struggle to escalate to the cost of easy escalation, you see much greater profit and delight (with WOM) in easy escalation. But Operations and Finance only look at the short-term costs.

Ease, and, for relationship surveys, “being easy to do business with,” are the KPIs most highly correlated with “definitely will continue purchasing,” trust, and “willingness to recommend.”

On the other hand, difficulty or struggle in escalation erases all positive memories and results in negative ones: and results (see FIGURE 1). The National Rage Study showed reaching a human and finding a useful phone were the key causes of rage: rage being a memorable emotion, more so than the original issue.

AI Experience Exactly Parallels IVR: Causes of Rage

Here are the top causes of rage with IVR that we are beginning to see with AI.

Access

AI tools use the same welcome as IVR, albeit sometimes text rather than voice: “Welcome, what can I do for you?” Then the customer must communicate their intent: what they want to do or a question they want answered. There are four causes of rage during access.

1. Long Messages and Wait Times

Messages that frustrate include, “your call is important to us,” “for faster service go to the website” (and saying “www.” to make your company look even more stupid), “your call may be recorded” (everyone knows that), and “we want to send you a chatbot link.”

If I’ve heard these messages and/or rejected them a dozen times, AI should not and should instead stop their use. An even worse message is a marketing message about great experiences on the lost baggage line, which one airline used to its detriment.

Remember, waiting seems twice as long as it actually is, especially if there is silence or bad music. Apple now at least lets you select the type of music you listen to. If the wait is more than five minutes use a virtual queue: see later in this article.

The record wait time I’ve seen is the State Department’s passport line, with a reported wait of 4 hours and 20 minutes. They did provide ongoing reporting to give you a feeling of progress in queue.

2. Not Knowing Me

That AI spits out multiple messages points out a critical and understandable source of rage, namely that companies don’t know the customers, no matter how often they do business or spend with them.

For instance, why do frequent flyers need to be told every time that the call is being recorded for training purposes or asked if they want a text link: which they have rejected the last 20 times they have called?

AI could easily determine the last time customers were told about the recording and if they have rejected the link and eliminate both of those messages: saving time and reducing my frustration. (Legal would admit that if you have recently been informed of the recording, you don’t need to be told every time.)

3. Cannot Understand My Intent

AI should be able to know if you’ve just had a transaction failure and ask if that is why you’re contacting. Further, it should help you communicate your intent by suggesting words that are actionable based on whatever input you’ve made.

4. Finding a Human

People like and, in most cases, prefer to interact with other people, particularly with urgent, important, emotionally-charged matters. While 90% want to self-serve, at the first glitch, they want a human.

Few things annoy anxious customers more than denying that access to them. And it is happening with AI just as it has been happening with IVR.

Giving customers the ability to “zero out” quickly and easily over the channels if need be is key to ensuring customer loyalty and repeat and referral business.

A general rule of thumb is to have all channels in view, though you can encourage one over others. For instance, AARP’s Help Page has a type-ahead driven “How can we help you?” entry point. At the same time, lower on the page, it provides the phone number, chat, email, and even the mailing address. The chat has a fast enough response that it has almost totally supplanted email as a channel.

The cost of bad IVRs is exactly the same as the cost of bad AI: customer struggle and market damage.

It is ok to hide a channel as long as you do it knowingly. One Boston beer company with a small service unit intentionally dropped the 800 number in favor of social and email. Their policy of same-day response assures consumers rapid, tailored replies.

The beer company’s manager knows that a few older consumers may be put off by not having an 800 number. But he has told us they are not the target market, and the company is willing to live with the recognized cost of not hearing from the older market segment via the telephone.

The manager has fully educated his quality and marketing departments on the science of the multiplier (12-20 problems for each complaint received), so he’s still getting good impact with his data.

On the flip side, operators at Chick-fil-A provide many guests with their cell number on a sign. The CEO of Sonos had a website contact us page that provided his direct email address at the bottom, as did Jim Robinson, the former chairman of American Express in the AmEx Monthly Newsletter.

You learn a lot via unfiltered input: and such an action sends a message to the rest of the organization that they should be open to complaints.

The solutions for access rage are remarkably simple and common sense.

• Publish Help and Contact Us at the top of the home page, not hidden at the bottom.
• Stop using stock messages. If you recognize the customer via phone number or email, use AI to determine if the last time they got the message and delete repeats.
• Publish response times near the channel, e.g., chat response standard 20 minutes. For long waits on the phone provide virtual queues; as long as you keep to the promise, little dissatisfaction is caused. Do like Apple and give the customer a choice of what they listen to: Eastern Onion used to have jokes and Aflac lets you listen to a duck quack.
• Link input channels to your user history so you do know the customer. For all my criticism, United Airlines at least says, “Welcome back John,” every time I call, based on my cell phone number. United Healthcare uses voice prints for authentication.
• For intent, have the AI (and IVR) suggest words based on your initial input to help you communicate your intent. AARP does this very effectively. Don’t be afraid to have a two-tier FAQ list, as long as each cluster has no more than seven subjects in a level. Have three or even four columns on a web page rather than making the customer scroll down.
• To find a human, assure the customer you will help them get to a human if the AI/self-service is not successful. When I had an internet access problem, Xfinity assured me that if rebooting the system did not solve the problem, they would quickly transfer me to a human tech rep. Given that offer, I was willing to try the reboot, which solved the problem.

Response

Customers want three things:

• An apology
• A fair remedy to their problem.
• Assurance that the problem will be reported so that it will not be encountered again.

Let’s look at these in turn:

On apologies, most are canned and many use language that does not admit blame and sound artificial, e.g., “I do apologize for your inconvenience.”

The word “inconvenience” belittles most disasters customers are calling about. If the event being called about is a flight cancelation, an unfulfilled hotel or car rental reservation, or a missed connection, it is much more than an inconvenience.

Apologies are most effective when the rep is using their own language rather than artificial scripted language.

Moreover, and all too often, the response is a legalistic “Gotcha,” e.g., “Sorry, paragraph 6 excludes your situation from the guarantee.” But it makes the company look sleazy and the customer feel mistreated.

As to remedies, too often they don’t cover all the needed bases, leaving customers dissatisfied and less likely to buy from you again.

The customer’s confidence in your brand should be restored as much as possible without extra cost. If it was a company error, there should be some compensation, e.g., an extra 5,000 miles, and there should be assurance that the issue will be conveyed to the quality improvement process. Get Finance to do the math: if the customer is worth $1,000, does it make sense to skimp on a $50 compensatory remedy?

If the customer was at error, they should still be made whole, possibly with a note to the effect that “this one time we’ve got you covered, but please be aware that you had a responsibility to do X which wasn’t done.” It is ok to educate the customer nicely, noting that next time they should act accordingly.

The remedy should solve the basic problem, even if the customer may have misunderstood some of the rules of product use.

The solutions to response rage are doable and they won’t break the bank, or worse yet, annoy shareholders.

• Apologies are most effective when the rep is using their own language rather than artificial scripted language. Think how you would apologize to your significant other or your child when you truly screwed up. Use common words like mess and hassle rather than problem and issue which sound legalistic.
• Remedies are a fair resolution and compensation for the hassle. After having said that, you will not be able to compensate for the lost time and effort (especially if the customer is a lawyer who normally bills at $600 an hour). Explain what you can do and offer to advocate for the customer up the line if that is needed.

For assurance of being heard, this can be accomplished in 10 seconds by the rep saying, “Your issue will be reported to our Quality (or Continuous Improvement) Department.” That gives the customer hope and a feeling of impact.

The major point here is that most reps do not have assured input. It can be a very simple mechanism: at Toyota, we established a simple email box with messages directed to the department root cause analyst where CSRs could direct a short description with a case number. The Toyota reps were also told quarterly what had been done with their input, so they sounded much more confident when saying the incident had been reported.

Escalation

Escalation out of AI is where most rage occurs. Ideally, AI should identify when escalation is appropriate and offer it before the customer starts looking – what I call “Psychic Pizza” – here is the pizza you were about to order. Text and speech analytics should be able to tell when a customer is dissatisfied.

But most companies hide the 800 number, or if available, have long wait times, and have no transfer of data describing the customers’ recent interactions. This is the corporate equivalent of a bratty kid covering their ears and yelling back at a parent or sibling, “Na, na, na, I can’t hear you!”

Struggle and anger cause customers’ blood to drain from their brain and go to their muscles (fight or flight). Even loyal, lucid customers become somewhat irrational when faced with multiple aspects of struggle.

Your empowerment of AI should mirror the empowerment you give your CSRs. Test every solution space with the CSRs first and program the same solutions and explanations...

If further escalation to a supervisor is needed, the wait and access are often worse. At one car rental company in the late evening, I was told by a preferred CSR that no supervisors would be available for three hours, when a new shift came on. The rep also was demoralized when I pointed out that there might be an altercation with dozens of waiting customers, resulting in the police being called to the airport before that time.

The solution to escalation rage requires quantifying its costs.

Step 1: Establish value of customers

Quantify conservative value of each customer i.e., one to three years of revenue. I usually ask Finance for the number and cut it in half. The calculation still works and when you present your business case to management, you get criticized by Finance for being too conservative. Perfect!

Step 2: Estimate customers lost if they abandon or struggle to escalate

We have found that up to 50% of customers with failed escalations are lost. Moreover, one out of five non-complaining (20%) are lost if they encounter a problem but don’t complain. If the customer complains and gets a poor response, the damage grows to at least 30%. Yes, one out of three customers is at risk.

A failed escalation results in more damage, up to 50% reduction in loyalty, while negative WOM doubled, which we saw in one case. Yes, one out of two customers is lost. Not only did the company have a bad product, but the struggle showed it didn’t care (also see BOX).

Step 3: Demonstrate that the ROI> 100%

When Finance objects to escalation to a human costing $25, place that within the context of a customer worth $500 or even thousands of dollars.

Help Finance to do the math. Is it worth saving $25 while losing $250 (50% X $500)? At minimum, the ROI of easy escalation is 200%, and in this case, $83/$25=330%, assuming a 33% gross margin.

Fix Your AI The Same Way You (Hopefully) Fixed Your IVR

I’ve discussed the causes of rage that we’ve found with IVR and are now seeing with AI and have provided solutions that can be applied to both. But now let’s deep dive into them, especially those that apply to AI.

1. Provide Easy Access

• Highlight help and be transparent it is an AI application.
• Minimize front-end messages based on what you know about the customer. Push back on Legal and Marketing that want to front-load frustrating messages.
• Assist the customer in communicating their intent. Think how you used example words in IVR menus.
• Test and continuously improve your AI tool using all the reasons for escalation which signify failures. Never stop measuring and assign a reason to every escalation beyond original intent.

2. Assure Effective Response

• For any problem, always have a natural sounding apology – for questions, always thank the customer for asking – convey enthusiasm via language. At Zappos, a CSR would say, “I love this type of problem.”
• Create flexible solution spaces for every difficult issue based on internal corporate data: do not trust external sources. (This one action will reduce your probability of disaster by at least 50%.)
• Your empowerment of AI should mirror the empowerment you give your CSRs. Test every solution space with the CSRs first and program the same solutions and explanations into the AI.
• Assure the customer the issue will be reported to Quality or Continuous Improvement [as I recommended earlier], and actually create a reporting channel.
• Ask (via popup survey or speech question) every customer receiving a problem remedy if they feel they were treated fairly and given a clear explanation. Make it part of the basic AI transaction, not a separate survey. These two dimensions are the best predictors of long-term loyalty and correlate with being “Easy to do business with.”

3. Proactively Offer Escalation

Remember, every escalation is a customer saved with a positive ROI.

• Provide an obvious channel for escalation to a human. Triggers to provide an offer of escalation can be use of negative language, more than three tries or an explicit request for “Human” or “Representative.”
• Provide an assured response time if it is not immediate; virtual queues with response within 20 minutes are satisfactory as long as the promise is kept.
• AI should transfer the data, or the customer should be given a case number to avoid retelling the story.
• The CSR should always stress the customer’s incident will be reported to Quality and identify the reason for escalation: separate from the initial problem. The reason for escalation is usually more important than the original issue in terms of process improvement.

Summary

AI tools are basically glorified IVRs with all the same problems. The payoffs of a good tool are huge for the 90% of issues that can be handled with self-service. The damage of a flawed tool is likewise as large. The key to success is to encourage escalation out of AI and capture the reason for escalation for every transaction.