Search

The Idiomatic Blog

Smarter Support

AI for Customer Support: The Idiomatic View

Artificial intelligence technology is about to change customer support, to the great benefit of customers and the companies that serve them.

This imminent evolution presents a major opportunity for customer support teams that are able to effectively leverage AI to craft consistently amazing customer experiences, at lower cost.  At Idiomatic, we are building a novel AI system that partners with customer support teams to make it easy to do exactly that.

Starting from the developments that have brought us to this point and a survey of the current state of AI for support, this post lays out the Idiomatic view of what AI for customer support should be.

Why Now

Customers have ever-higher expectations for the product and service experiences that companies deliver.  The companies that win are those that create great customer experiences, and customer support teams are on the front lines of doing so.

However, consistently creating amazing customer experiences is currently difficult and costly, particularly at scale. We’ve seen this in both our own team’s experience and our many conversations with great support teams at beloved companies.  It requires being able to consistently delight customers, rapidly solving even the most complex customer problems and empathetically connecting with and helping even the angriest customers.  Great support teams also communicate actionable insights based on all of those customer interactions to the rest of the company to improve the entire customer experience.  These tasks are non-trivial to execute (see our previous blog posts for some workflow best practices), even for a large and experienced support team.

Fortunately, even as customer expectations reach greater and greater heights, rapid advances in AI technology are creating new state-of-the-art capabilities for language understanding, speech recognition, and many other domains.  These AI advances, intelligently applied, promise to make support better, easier, and less expensive.

The Current State of AI for Support

Unfortunately, existing applications of AI to support do not deliver on its promise. All too often, existing systems create poor customer experiences in order to yield cost savings and are limited in their usefulness.

chatbotAnyone who has interacted with chatbots knows how quickly, and frustratingly, they reach their limits.  As a result, some companies have instead adopted a human-plus-chatbot model in which a chatbot suggests responses and a human agent can override them.  This approach better avoids the poor customer experiences that occur when fully autonomous chatbots fall short, but it only helps with one component of a great support team’s workflow: drafting the text of short chat snippets for customer issues that have been observed in the past.

portal-deflection-onTicket “deflection” tools take a different approach by attempting to divert customers away from actually speaking with support, generally by sending them impersonal prefabricated responses or links to knowledge base articles.  These systems’ shortcomings are immediately clear even from their name: their goal is not to create great customer experiences but rather to reduce the number of customer contacts by deflecting customers away from conversations with support, without actually driving fixes to the root causes of those contacts.  Additionally, this approach is only feasible for some types of customer issues and frequently results in frustrating customer experiences.

The Idiomatic View

We believe that it is time to actually deliver on the promise of AI for support: consistently amazing customer experiences, at lower cost.  So, we are building a novel AI system that is designed from the ground up to partner with support teams and empower them to be much smarter about how they accomplish both of those things.

We believe that AI systems for support should not result in poor customer experiences in order to reduce costs, nor should they avoid solving the hard problems inherent in fundamentally improving the customer experience and customer support teams’ workflows.  We are building Idiomatic’s AI system accordingly.

In order to accomplish all of this, an AI system for support must have a number of key characteristics (some of which have been similarly recognized in recent efforts in other domains to create AI systems that operate in the wild in close contact with humans).  Our system, in addition to satisfying standard infrastructural requirements such as high availability, is designed to be

  • Customer-centric: As discussed above, many existing AI systems for support produce poor customer experiences, often focusing on optimizing only low-level statistical metrics, such as prediction accuracy, for standard machine learning tasks.  However, an AI system for support should focus on the mission of support: consistently creating amazing customer experiences at or beyond the level of a veteran human support agent, while also reducing costs.
  • Workflow-aware: An AI system for support should meaningfully address the core tasks of a great support team and integrate seamlessly with its workflow.  Those tasks include routing tickets to the right agents, diagnosing customer issues, actually resolving those issues rapidly and consistently, responding to customers empathetically and delightfully, monitoring performance on all of the above, creating and deploying self-service resources such as knowledge base articles, and driving fixes for the problems that cause customers to contact support in the first place.
  • Easy-to-use: Support teams should not need to go to great lengths to clean, organize, or label data in order to use an AI system, nor should they require a team of engineers or data scientists.  An AI system for support should function well, and continue to function well, out of the box.
  • Robust: Customer issues are varied and complex, and new issues continually arise, particularly as products change and new features are released.  An AI system for support should provide consistently great customer experiences even as it encounters this variability and complexity.
  • Transparent: An AI system for support should be able to explain itself and its actions.  For example, such a system should be able to tell you how many customers have complained about a given issue, or why it routed a particular ticket to a particular agent.  This capability ensures that valuable insights are not lost and that the system can be trusted and effectively controlled.
  • Controllable: Much as a support leader can rapidly change the way in which their team handles different customer issues, so too should they be able to rapidly and reliably change the way in which their AI system does so.

In developing the AI techniques and technology needed to deliver on all of the above, we’ve designed our system to include humans in the loop as a key ingredient.  The hybrid intelligence that results from this partnership between humans and AI allows both to do what they do best, each enabling that which cannot be done by the other in isolation.  Machines can act rapidly and consistently, automate simpler repetitive tasks, and effortlessly assimilate (and optimize based on) large amounts of information and past experience.  Where they fall short, they rely on human supervisors to add knowledge of business context and an empathetic touch, as well as a robust ability to handle previously unseen or complex customer issues and language.  In turn, humans are freed to focus on more complex and valuable issues and tasks, aided by machines to the extent possible.  Machine intelligence and human intelligence and empathy each elevate and assist the other.

The first version of our system is already providing our clients with unprecedented continuous insight into what their customers are saying at scale, in a precise and human-understandable way; this insight is a powerful lever for improving support operations and fixing the root causes of customer contacts.  And, we’re hard at work on bringing AI-driven capabilities even more directly to bear throughout support teams’ workflows.

AI is coming to customer support, and we’re solving the hard problem of making it easy for companies to seize that opportunity.  Stay tuned for more dispatches from the journey!

Simple KPIs to Transform Your Customer Support From a Cost Center to a Strategic Asset

Most Founders and CEOs grow a Customer Support team because they have to not because they want to. No matter how well a product is designed, customers wind up having problems and when they do, they prefer a company that provides them with the support they need instead of letting them struggle to solve their own problems. In fact, study after study shows that high quality support can be more impactful than innovative products.

To fill this need, companies start by asking employees to answer customer questions in their free time. The “free time” model never scales; therefore, a dedicated Customer Support team almost always soon follows. The mindset of “someone has to answer the emails” leads companies to use productivity-based metrics to judge the performance of their Customer Support teams. Reporting things like the number of emails answered, time to resolution, and the costs of the headcount give a basic sense for how costly answering all the emails has become. It’s a cost-based mindset; the goal is the most production for the least investment.

But, a properly run Customer Support team has loads more strategic value to offer. Specifically, talking to thousands of customers every week is a secret weapon that allows your company to make customer-driven decisions. That is, decisions resulting in happier customers due to improved experiences using your products or services.  

…talking to thousands of customers every week is a secret weapon that allows your company to make customer driven decisions.

How do you give your teams the ability to make customer-driven decisions? How you do you unleash your secret weapon? After all, you can’t just forward thousands of emails each month to your product, marketing and other teams who need to know this information. Companies try things like “all-hands support,” where every employee takes some time each week to answer Customer Support emails. In this case, each employee only learns what they anecdotally see. Anecdotes do not provide the conclusive insights necessary to be confident in your decisions.  

Similarly, the thousands of conversations you regularly have with customers are dispersed across many Customer Support agents. Agents are not normally empowered within the organization and even if they are, there is the problem of the dispersed information. How do you make sure the right agents are speaking to the right people within your team? Furthermore, agents are often not the best people at your company to be doing this work. Hence the relatively recent rise of “Voice of Customer” or “User Research” teams.

You must discard the cost-based mindset to unleash the value in your organization. Idiomatic was designed to combat the problem of dispersed information, but even if you aren’t ready to adopt our tool you can read our post about how to get started tagging your Customer Support cases as they come to create a shared set of structured data for your entire company to learn from. Instead of using productivity metrics to judge your Customer Support team success, you need to measure and judge the things that matter for collecting and communicating insight.

Metric 1: Tagging Accuracy.

Start by measuring your tagging accuracy to measure your team’s ability to collect data accurately. The best approach is to manually measure the consistency of your tagging by doing audits. Choose a random sample (start with 1,000 each week) and reassess the tags on those cases. Whatever percentage are properly tagged is your accuracy percentage. Use this percentage as your KPI and track it over time to see if your tagging accuracy improves. A good rule of thumb is that this number should be over 85% if your tagging is well thought out.

Metric 2: Product Specific Contact Rate

Measure the contact rate being produced by each feature or area of your product. In order to do this, you need to make sure your tags are specific enough to answer the question: “have more or less people contacted me about this feature this week?” Customer-driven improvements should measurably reduce these numbers because a better product experience that produces less friction and reasons for contact from your customers. If your product or service improvements do not reduce your contact rate then you have identified an error in your attempts to be customer-driven. This means that for each product feature / area at your company you will have a separate KPI known as “contact rate”.

Start by measuring your tagging accuracy to measure your team’s ability to collect data accurately…

Replacing your traditional productivity KPIs with these metrics will allow your team to focus on higher value goals. Instead of rushing to solve as many cases as possible, your team can focus on learning from the interactions, reducing contact rates and eliminating future efforts. It is far more efficient to not have cases then to solve the faster. As such, you must choose the right KPIs to incentivize your team to improve on this learning.

The benefits of being customer-driven are improved retention and growth through improved customer experiences. One of our customers is Slack, an incredibly successful company that attributes a lot of their success to customer-driven product decisions. If you want your customers to love you as much as folks love Slack, change your mindset and don’t set KPIs around cost; choose KPIs that incentivize your Customer Support team to help your company learn from each interaction. Only then will you be able to make customer-driven improvements to your products and services.

The Key to Scaling Customer Service the Smart Way

Before starting Idiomatic, I founded EAT Club, now the country’s largest online corporate lunch delivery company. In our first year, I personally fielded every customer service email and call, showering each customer with attention.

Along the way, I got to know hundreds of our early customers. I learned their names, where they worked, and their favorite meals. Importantly, I learned what they loved and hated about our experience, allowing my co-founder and I to focus on fixing the largest pain points.

Our happy customers spread the word, and the business grew. But, we quickly learned that my original strategy “do whatever it takes” doesn’t scale for large customer support teams. The dual challenges we faced were maintaining high support quality while controlling costs.

cost-final
Continue reading “The Key to Scaling Customer Service the Smart Way”

Getting started with tagging: a step-by-step guide to mapping the customer experience

Tagging can be a powerful tool for any customer support team. When done correctly, tags serve as a common language to communicate the customer journey at scale—when having one person keep in touch with every single customer is no longer an option.

When it was just me, I knew everything, so I was the entire voice of the customer.

Now we’re bigger. When we give feedback, folks ask, “How much of a problem was that last week? Is this just what you remember or is this a real trend? Are all the agents seeing the same trends?

– Recent Idiomatic customer interview

Tagging can provide the quantification necessary to answer these questions by aggregating data across the entire support team. But, smart support leaders know tagging will only be effective if the tags are designed properly, the agents are trained consistently, and someone is held accountable for the accuracy of the tagging. So, whether you are developing a tagging process for the first time or redesigning one that has become unwieldy, here are some practical considerations to help you get started.

Continue reading “Getting started with tagging: a step-by-step guide to mapping the customer experience”

Idiomatic Bootcamp #2: Uber & Lyft Cancellation Fees (infographic)

My wife and I recently took a red-eye flight into Boston for our college reunion. We landed at 6am and were exhausted, so I decided to call an Uber rather than taking the subway.

Six minutes later, my phone buzzed and informed me that our Uber had arrived, but the promised red Prius was nowhere in sight. So, I called the driver, and found out that he was on the departures level whereas we were upstairs at arrivals (as specified by the app). The driver said that he’d circle around and come get us. 10 minutes went by and still no Prius, so I checked the app and saw that the driver had left the airport altogether and was halfway to downtown Boston. I called the driver again, who assured us that he was coming back, but the app clearly showed that he was continuing to speed away from the airport.

At this point, a reasonable person would have cancelled the ride to find an alternative. Had I done that, I would have been hit with a cancellation fee, because Uber’s policy is to charge $5 if a rider cancels a ride after 5 minutes (2 minutes in some cities). Continue reading “Idiomatic Bootcamp #2: Uber & Lyft Cancellation Fees (infographic)”

Idiomatic Bootcamp #1: Why Uber’s new feature misses the mark (infographic)

Idiomatic helps companies identify pain points and friction for their users by structuring their customer feedback (support emails, reviews, survey comments). The data below comes from our platform. If you want to explore the Uber data in Idiomatic’s dashboard, request a login.

Without a doubt, Uber is a business and social phenomenon. Despite being a magnet for controversy (e.g., cities forcing them outlawsuits by drivers), Uber continues to thrive because it has created a magical experience for a large swath of riders. But, Uber’s path to world domination / IPO requires continued growth, which means winning over more riders and fending off competitors. And the only way it can do that is to make sure an ever larger fraction of those who try Uber fall in love with the experience.

Continue reading “Idiomatic Bootcamp #1: Why Uber’s new feature misses the mark (infographic)”

Why we solve Pied Piper’s product/market fit problem

I am the stereotypical startup founder HBO’s Silicon Valley was born to mock, but I love the show. I went to Stanford, live in Palo Alto, and am working on my second company right now. In the most recent episode of Silicon Valley, Pied Piper (the fictional company around which the show revolves) struggles to gain product/market fit. Last night, when I finally got around to watching the show, in my head I was shouting “We could totally solve this!” I had to write a post explaining why. Continue reading “Why we solve Pied Piper’s product/market fit problem”

Hello, world!

Welcome to the very first blog post by your friends here at Idiomatic. We’ve chosen a blog to announce ourselves instead of your run-of-the-mill funding announcement because, frankly, we don’t currently have plans to raise any money. We’ve already accomplished some amazing things, and we’re very excited about what’s next. We’re working on important problems, we’re overcoming big challenges, and we’re having fun doing it. But first things first, how did we get here?

What we believe.

“What is an important truth that very few people agree with?”
-Peter Thiel

This is a question that Paypal founder and early Facebook investor Peter Thiel often asks startup founders. Here’s what we believe:

Products that win feel magical. Customer feedback provides the path to eliminating the friction disrupting that magic.

Continue reading “Hello, world!”

Blog at WordPress.com.

Up ↑