Building and Scaling a Customer Support Function

All right, we are officially live and I’m excited to talk customer support, customer experience with Steve Ketting. Steve, great to have you here today.

Thanks so much, Kate. It’s great to be here.

Cool. Can we go ahead and kick things off with your intro and tell everybody a little bit about your background and what you do today?

Yeah, of course. Well, again, thanks everyone for joining. A very quick auctioneer speed background on myself. So I am a recovering management consultant. I was at Bain for about seven or eight years, mostly in its private equity practice. I had a long and circuitous route into early stage tech startups here in the Bay Area. And that journey landed me in 2016 at DoorDash, which at the time was a sort of somewhat unknown fledgling- That’s funny to think about. Third- party delivery company that was almost bankrupt.

And I joined as their GM of its Austin market. That led me through what I call a battlefield promotion into triaging and then running its customer experience and customer service organization, which was and is massive. So in that five years I was at the company, from 2016, 2017 through IPO in 2021, grew that organization from about 400 folks to 20, 000, putting it into the upper stratus of large CX orgs.

And so got a crash course in all the nuances of how you build and structure and scale customer service, certainly, especially in the consumer space, but across all of its consumers, drivers, and restaurants. Since then, I have been building my own business. So I founded Alpine Partners, which is an advisory and angel investment firm. I jokingly call it Alpine Partner because it is just myself. And so in that role, I do a whole bunch of things like this, advising CX leaders and helping growth companies organize for scale.

Awesome. So today’s event is set up as an AMA. We have a bunch of pre- submitted questions that we’ll funnel into the conversation. As an attendee, please also ask questions as we go through the material and we’ll weave those into the discussion. So we’ll get started with some of the prepared materials and then weave in questions along the way.

Sounds great, Kate. Very roughly on the agenda today, I mean, we will touch on a variety of questions, but some main themes will be, how large should your customer service team be? How should it be organized? How should you think about measurement and Northstar metrics for that team to understand whether it’s doing a good job? Should it use AI? And more importantly, how should it do that? And so wanna touch on these things as we go.

But before we do that, I think it’s important to start from first principles because I think that we all are users of customer service in our lives. We’ve had hundreds, if not thousands of those interactions. And so we have an idea of what it is. And so often skip ahead to just like, here’s how it should be or feel. And I like to start from first principles of what are you trying to achieve with a customer service organization?

And I think about it in this way and the dichotomy between customer service and customer experience thusly, which is we all get customer service, right? You’re a customer, you get acquired, you have a good experience. At some point you have a defective experience at DoorDash that would be like your order is late. You call customer service, the human agent picks up, says, I’m so sorry, gives you a credit, end of transaction. That is a very narrow and linear version of customer service.

When I think of customer experience and what’s happened in the last 10 years of how it’s expanded, it looks a little more like this, which is you’re starting to get into, can you proactively get that consumer or customer back on the happy path of their customer journey proactively before they even have to talk to somebody? Once they have talked to somebody, if you can’t, then very importantly, there is a feedback loop that goes to the business such that those defects are increasingly eliminated over time.

And this is what I think a lot of companies get wrong about customer experience, that it is a team that, at DoorDash we jokingly call it the end of the river of garbage. The team that just handles all the trash and when bad things happen, it’s on that team.

The reality is customer experience, it is a mirror back to the business of how good your product is overall and great companies make sure that it’s a feedback loop such that over time you’re eliminating those defects at their root and not having them such that the customer is ultimately retained and continues to purchase and use.

So why does customer service exist?

Seems like a straightforward question, but I think that there’s some nuance here, that we get wrong sometimes, and I want to touch on why it exists and why it doesn’t. So I think one is: your customers have paid for it in their purchase in some way, shape or form, and so if they have a bad experience, you owe it to them to get them back on the happy path. They should be able to contact you in some way and get either an automated or human resolution to their problem. Like that is the table stakes of running a, you know, baseline business.

The second is to protect your brand. That you know to the extent to which your brand relies on, you know, five star service, or even, in like an Amazon case, let’s say, three stars, whatever you want to call it. You’ve made a commitment and a promise to your customers, and CX exists to protect your brand.

And then, most importantly- and this is where I think some people don’t spend enough time to put itself out of a job- that great CX teams shrink themselves over time, you know ensuring that they are solving defects and getting smaller and more efficient by doing so. Why doesn’t it exist? And so these are, I think, maybe more controversial: retain your customers- maybe the most controversial.

At DoorDash we found over time that that the impact to a customer’s retention from the initial defect just massively overshadowed anything we can do downstream in customer service. So once someone’s had a late order or whatever it is in the case of your business, the cake is somewhat baked. You, you can do your best to deliver. You know good service because it’s owed, because it’s good for the brand. But no matter what we spent on that resolution, you know a really great agent or a massive refund and credit, whatever it was.

We never showed an ability to drive differential impact or attention. So if you’re looking for customer service to drive incremental retention and goaling them thusly, it will be very challenging. Second of all, surprise and delight. Every CEO thinks: oh, I want you know customer service, send cards on people’s birthdays or send- you know, send- a cake. If we do something really really really terrible. This is a marketing activity. It is not typically successful within the customer service org. And then finally, be cheap.

Every CEO, and CFO in particular, looks at CX and just sees a bag of money and wants to strangle out as many dollars from that team as possible and by all means be efficient. My journey at DoorDash was was almost entirely an efficiency journey, but not at the expense of the customer experience. Very oftentimes that that, that eye of Sauron, should be going back to the business itself, where the defects are created, not downstream at your, your CX team.

So what extent? But on that side, to what extent can support shore up? You know gaps in your product or service offering. You know if it’s a software product and there are some pieces of it that you know are buggy, to what extent can support help smooth that over? Or you know if, operationally, you’re having some challenges, what is the role of support in making up for deficiencies elsewhere in your product?

It’s a great question and the answer there is: there’s. There’s sort of two, two buckets of things that CX does. One is what I call sort of enduring issues, and these are issues like: take an airline, an airline will always have late flights and so you need to build a team that can do a really good job handling those things forever.

And then there’s the like of the product gaps and certainly over like especially in growth companies, you will have product gaps, and those are sort of the acute issues where something doesn’t work properly and, yes, you need a backstop there. It’s the latter bucket that needs to be continually addressed and fixed.

And I will say one thing: you know, one of the best wins we had at at DoorDash was every product brief we would get would have a part of the bottom of that product group that said, okay, what will we do with edge cases, because this product will not work in a hundred percent of situations. Oh, just send it to customer service. That culture is something that over time will overwhelm your business if you don’t start forcing craftsmanship upstream.

So yes, by all means, like customer service, literally by definition, exists to handle edge cases, but it cannot do so, you know, at huge scale and in perpetuity.

One more question in that vein: should CX being able to actually fix things? Did you staff any developers to the CX team at DoorDash?

Good question: yes and no. Yes, we had our own product and engineering team to build our own tools. So, for example, we built a custom workflow tool for our agents to get them to proficiency and make them very good at their job delivering service. Did we have engineers? I’ll give you an example: there was a bug or an issue at DoorDash where customers would mistakenly order deliveries to an address they were not currently at, so they’d be at work, they’d order it would go to their house and that would cause problems. That was not our job to fix.

Nor I don’t think it should be an engineering squad within CX to fix. That needs to be the engineering pod that owns, in our case. Like the consumer growth side of the business, like that needs to rest with the team. You know so you break it. You bought it that if you own the consumer journey, that you need to solve those issues. Now to the point of empowered to actually fix things. Yes, so I’ll get to the org chart in a second.

On my team that the customer experience side or branch of our org chart was people who mapped upstream to different teams whose job it was to go and hammer on desks to make sure that the people who cause those problems went and fix them. And we had very strong buy- in from our C level team to have them go do that.

Awesome, let’s get into the orchard, sure you?

So, it’s been an eye chart, and I’ll go through individual pieces here in turn, but in short, we broke customer experience down into really, for all intents and purposes, like two big buckets. Let’s put trust and safety aside for a second because I think, you know, you must have it. We all know what it is.

It’s not terribly strategic in this case.

Those two buckets were support operations and customer experience. And the metaphor we used was like, think about like a cruise ship or an oil tanker. Customer experience is the bridge. It’s the strategy. It’s pointing out where we’re going to go. We’re going to take the customer experience. Support operations is the engine room that makes the massive oil tanker move and run.

And that was the case for us.

So support operations, their job effectively was to make sure that when someone had a problem, there was an agent available at that exact moment in time at the cost or the budget that we could afford or at our budget target.

And so all the pieces to do that were as follows. So as I was saying, I think I’ll just take you here.

So support operations, balancing service level, i. e.

speed of answer and cost.

And that’s really what that team focused on over time. Customer experience was more quality. So CSAT, quality of the resolution, NPS, feeling about, you know, the overall journey, et cetera. To get into, as I was saying, the support operations side, there are a few different functions, you know, I’d call them inside baseball functions of customer service. Those are in order.

Workforce management, that’s the team that’s literally looking at the arrival pattern of all your tickets and figuring out, you know, who needs to be staffed, where to answer quickly.

Learning and development, how do you get people up to proficiency who are either new agents or existing agents learning a new thing? Quality assurance, how do you measure whether the, you know, when an agent does something they shouldn’t or provides a bad experience, how do you measure that, identify it and coach to fix it? And that’s increasingly more and more automated with AI, but has been a large team in the past. Tools and technology, self- explanatory. What is your tech stack? How is that driving down the handle time of your contacts?

And then a final, which is agent operations, and this is really, if you have an internal team, it’s your, you know, internal contact center. If you have external outsource vendors, it’s managing those vendors. So really, it’s like, this is the people side of the equation.

And so imagine a small company that has, you know, a head of CX and five support agents. The head of CX is doing all of this stuff?

Yes, and it’s no less simple than just yes. What I’ve seen happen is this tree just gets deeper and deeper the larger you get. So if you’re small, it’s your point, if you have five agents, you are doing everything until you can’t do everything anymore, and then you start breaking it down. And I think that you then break it down into areas of importance.

From my perspective, you know, as I, you know, if I was growing a small team, I would want somebody on probably like workforce first, because like, that’s an important deliverable or metric that needs to be had around like, who’s on managing your service levels. But it depends on the situation. And it’s up to each individual leader, the leader to figure that out.

Awesome. We have a couple of other questions that have come in that these two are related. So one is, should CX make promises on features to customers? And the other is saying no, it’s really hard for CX for a number of reasons. Have you found some clever ways to break out the crisis cycle to handle upstream issues and allow CX to optimize and focus on scale?

If I understand the question, saying no is difficult in CX for providers and the customer will churn or it’s urgent. I think this, you know, certainly this is a, you know, very applicable in B2B. I think it’s hard, I’ll say in two parts. The first part is great CX leaders are incredibly empathetic. That’s certainly in my case. I never want, like, I take it personally when someone has a bad experience or we deliver a bad resolution for someone.

That is like certainly the case. I gave a talk last week on, you know, hyperscale and what it feels like to be in hyperscale for a business.

And the reality is that great businesses prioritize ruthlessly. DoorDash is 10 years old.

You can believe that there’s still tens of thousands of people every single day that have bad experiences.

It is not because they don’t care about that, but it’s because they are prioritizing their scarce resources to solve the biggest issues in order.

And so I, if I’m answering the question correctly, I think, you know, you need to be, one of the things that was very helpful for us is when we go through a planning cycle, we’d be very explicit within the company about, like, will and won’t do.

So we would have a list of, like, a zillion things you want fixed. We prioritize five product teams across the company would say, we’ll do these. And we are, we’re not going to do the following ones.

So everyone was aware when people inbounded, screaming and yelling and kicking and screaming.

that we’d be like, yep, we decided not to do that. It sucks, it’s bad, but Rome was not built in a day and it’s unreasonable that you can have a flawless experience for an early stage growth company. I wanna pause and see if that answers the questions because it might or it might not have.

So Ben and Shan, feel free to clarify in the comments. I guess, what about Shan’s question around should CX make promises on features? So if there’s a feature that you’ve agreed is in development, can you tell customers or is that dangerous territory?

I think I would say it’s dangerous just because I don’t know. I think it’s fine to say that and we certainly would say like, if it’s something, we will take that feedback and relay it to our product teams. Certainly, if that’s true, like pulling back, one thing that I learned very, very aggressively is never lie. And I mean that down at the even the agent level, it might feel good in the instance, but that is never helpful, especially with restaurants at DoorDash that live on the platform, they’re highly engaged.

I think it’s fine to say, we take this feedback and we relay it to our merchant product team, if that’s true. In our case, it was true and so we would say that. Product roadmaps change frequently and so I would never wanna commit to something that I was not directly in control of.

I think that makes sense. And I think that your point around, hey, how do we align internally on, here are all the challenges we’re having, these ones are the most important to us, can we get these ones fixed is probably the only way through. Cool, so no additional comments there, so we can keep going back to sort of walking through the org chart.

For sure. So stepping in the other half of the org chart on the customer experience side, it’s more limited, but it really breaks down into, as we were just discussing, two main pieces. There’s the customer experience of customer service and that’s things like resolutions and workflows. So what do I mean by that? At DoorDash, on the consumer side, we were constantly rolling out new consumer products. I’ll use a good example, DashPass, the subscription product at DoorDash.

Well, when the consumer team rolls that out, you need to have someone who gets deep on that product and figures out how you’re gonna, well, what are the problems gonna be that you’re gonna get things like, how do you refund a subscription? When does subscription start, et cetera? My subscription wasn’t applied. My Chase rewards card was not recognized for the promotional, yada, yada, yada. So someone needs to dig into those things and figure out what the resolution or workflow’s gonna be.

So that’s one, and that’s the person who’s responsible for the CSAT or the satisfaction of that resolution. The second piece is like, okay, hey, look, we have all these defects with DashPass. That person then needs to go back upstream across the office to the head of DashPass product and say like, Doug, these three things are ridiculous and they’re broken. Please fix them by the end of the quarter and then stay on Doug to make sure that he does it. So those two things are of incredible importance.

If I was to break these two pieces down around like, who do you hire? If you’re gonna hire experts, they probably are gonna go to the support operations side. If you’re gonna hire smart, talented generalists, they go to the customer experience side. Wanted to step into just a quick rundown of CX math. How do you goal a team on efficiency and what are those metrics? So this is what we spent most of our time on for our entire five- year stint at DoorDash, I’ll break this number down.

Total cost is a function of basically five main inputs and I’ll step through them to give you a window into each. So those are business volume, defect rate, labor costs per hour, productive minutes per hour and average handle time. So I’ll step through and show how these ladder up to get your total cost of customer experience or customer service. So very simply, you start with your business volume times your defect rate. So let’s just say you have a hundred orders and your defect rate or your contact rate is 20%.

And I use those somewhat interchangeably. You’ve got 20 contacts. Makes sense so far.

You then go down the equation and look at your labor cost per hour of the people who are handling those contacts times how many, you’re paying them 25 bucks an hour, how much of that hour are they actually working on tickets versus doing other things, having lunch, having a break, getting trained, whatever. And so you get down to let’s say 45 minutes of that hour, you’re actually working on tickets. So you get to your cost per productive minute.

You then take the average handle time, so that’s on average how long does it take an agent to handle a ticket, let’s say in this case it’s 10 minutes, so you get to your $ 5. 56 per contact. You multiply through 20 contacts by this, you get to $ 111. 11 in this made up example. So this is just like the basic overview of CX math, and it’s probably more helpful to folks who are outside of the CX world to just get an idea of how you might want to break these things down.

And you can break them into more detail, which we won’t do here, but happy to take questions if helpful.

We do have one efficiency question that was pre- submitted. How do we get more efficient, particularly on the call handle time? Discourage 15 to 30 minute long chat engagements with customers, which is common at our company. Any advice there?

It’s a really good question. I would start by, I’ve worked in CX across three different businesses, and I’ve advised probably a dozen. And one thing I’ve learned to be true is that outside of the long tail, no one wants to stay on with customer service more than they do. And so I would answer this with questions around, is it because the agent is just running the clock? Is it because it’s a nice old lady or a nice old man that just wants to have a chat? If it’s happening at scale, I want to dig into the why.

If it is that people are just moving slowly, I will tell you one thing without having to be able to have a back and forth to dig into this one. One thing I will tell you is this, a huge mistake that we made. If you go back to the math here, average handle time, if you reduce average handle time by 50%, you reduce your cost by 50%. If you go from, like we did, from 14 minutes down to seven minutes, your cost gets cut in half. That’s great.

The problem is, we made the mistake of going to all of our agents and all of our outsourcers and all of our internal team and say, we said, go faster. The problem is this, they will. You tell an outsourcer or you tell a team to go faster, they’ll do it, but they’ll do it in ways that suck. They’ll end chats quicker. They will be ruder. They will take shortcuts. They will, like, hit your goal that you give them and probably damage the customer experience.

So in this particular example, without knowing more, what I would say is unpack the root cause of why this is happening and solve there rather than finding a way to, like, negatively discourage or penalize people for having long conversations. Because unless it’s, like, fraud, I wouldn’t expect that it’s something, you know, that you wouldn’t want to tackle at the source.

Cool. Feel free to chime in with more questions related to efficiency, but we can keep cruising.

Sounds great. CX math on quality is the challenge here, and for folks on the call who are deep in CX might relate to this. Measuring the quality of CX and customer experience is challenging because unlike most customer- facing teams, the North Star is not revenue. So if it’s not revenue, which is a very hard and fast and easily measurable thing, how do you measure it? And I’ve sort of come to four measures of quality in customer service that you can look at as a basket. So those are defect rate, I think, is enormous.

Is that you, like, I’ve had clients who were at a 4. 95 out of 5 CSAT, you know, saying, hey, we want our CSAT to be higher. And I’d ask, like, what’s your defect rate? They’re like, oh, 50%. Great businesses have a defect or a contact rate as low as possible so that people just don’t have to contact them. Have you ever contacted, like, McDonald’s customer service number? I haven’t.

I have. Their app actually kind of sucks. But that’s a different problem.

That is a wild twist. I never thought I’d get pushed back on that one piece. It’s a really good point. Defect severity. So this is where retention does actually enter, which is that, like, you know, I work for apparel companies where the defects are, hey, how do I do a return? And you tell them. They’re like, great, thanks. That’s a very, you know, unsevere defect. It doesn’t really matter. Then there’s, like, hey, I’m a dasher at DoorDash and I haven’t been paid for the last month. That’s extremely severe.

And it’s going to have a huge impact on their retention. So there’s some sort of measure of, like, you know, how frequent but then how severe as well. Then you get on the resolution side. There’s, like, resolution effort and quality. So resolution effort is, like, what’s your how many contacts or, like, what percent of your contacts are you answering effectively instantly, which is, like, I would say the expectation today. What is your abandon rate? How many people are just saying forget this?

I’ve waited for, like, you know, at DoorDash it was 20 seconds. If it took you longer than that, you’d bounce. And then finally resolution quality. So basic measures of CSAT, MPS. Did you do a good job once you got the person in front of you? And the sum total of those, you know, basket into the collective total of your customer experience.

So I have another org question that kind of goes into this batch, which is, what are the pros and cons of organizing support teams by customer type or business segment or function?

Such a good question. I’ll tell you how we did it and then I’ll get to this question. We ran a matrix. So if I pull back for a second, the entire DoorDash structure and operating playbook was ripped right from Amazon, because it was all a bunch of Amazon executives. So the cornerstone of the entire organization was the concept of ownership. That if there was a problem, one person owned it and then they could just figure out how to achieve their objective. So you end up with a pretty complex matrix and that’s what we did in CX for us.

So we had a director who owned our consumer line, our dashboard line, and our restaurant line, and then one who owned international businesses, and then matrixed into that, we had our support operations team that actually ran the corresponding pieces for each of those people. So in that way, we had people who you could own very clearly, something like your SLAs, that was our head of workforce, but then also someone who could own, for example, the defect rate on the consumer side, that would be our consumer head for CX. So we did both.

Here’s what I’ll say though, is where this goes wrong, and what I’ve seen is that you have your operations. Let’s say you have a team of 100 agents, and then those agents have leads, those leads have managers, and their managers report to the head of CX. Well, you have not broken down your metric tree whatsoever, because every single lead and every manager owns a little bit of SLA, a little bit of CSAT, a little bit of yada, yada, yada, which means that you as the head of the team own everything. That’s the worst model, I would say.

So just so long as you’ve taken the problems that you want to solve, we look at those metrics of efficiency and of quality. As long as you have clear ownership of that, I think you can organize by your business needs.

So in which of these roles might you need deep product experience or industry experience, for example, and where would you split things out if that were the case?

Yeah, for sure. So on the support operations side here, you’re going to need someone who has some idea about workforce. It’s like a relatively esoteric discipline. Learning and development, you’re going to want someone who has experience like training agents, because it’s somewhat unique. That’s where, I mean, it’s less…

You know, I’m thinking, so let’s say I sell a cybersecurity

Yeah, where might…

Okay, hypothetically, I sell a product, and some of my customers are in cybersecurity, and they require this extra special knowledge, and some of them aren’t. Where might I split up my team if that were the case?

I think that you might want to solve that in terms of like level of escalation, and not by breakdown, but like at level. And so if you contact tier one support, and yeah, they can handle a password reset, but you call in and you say, you know, I don’t even know what the thing would be, but it’s like very deep in that product, you probably get escalated to, you know, a tier three, or even tier four sort of pseudo- engineering team to solve that.

That makes sense. One other, while we’re talking org, and this is from very much a B2B lens. Oops, sorry, not this one. We had a question on, I will find it, the balance of support, CSMs, professional services, onboarding. So say you have a large post- sale team that includes a bunch of different functions. How do you think about, I mean, the ratio is probably going to be, and it depends, but where does support fit in? Can you opine a little bit on the sort of broader post- sales team?

Yeah, well, you beat me to it, that the ratio is, it depends, and I can speak to… I think that, I’ll give an example. So at.

At DoorDash, the back- office support for restaurants, the team that would set up the menus of these restaurants, send out a tablet, send out a printer, et cetera, you’ll get your banking information set up, working in Phoenix, they bounced between orgs constantly. We couldn’t figure out where they fit, and ultimately, they were under my team, and we ultimately moved them under our head of merchant, and their goal was literally merchant revenue, so the total dollars that we generated from restaurants on orders.

The reason for that was that we ultimately got to a place where the question was, whose ultimate end goal is this team most aligned to? Who will succeed or fail most based on whether this team does its job or not? That’s what dictated the org structure. To your point here about onboarding, support, CSMs, I think where I would say is I’ve seen success in putting sales, onboarding, and CSMs all under the same CRO because they are all necessary.

You need onboarding so that sales gets paid, you need the CSM so that the clients are retained and that you grow their revenue, your NRR grows at goal, whatever. So thematically, those all lie under the same thing, and so I think that that’s a good home if that answers the question.

I think so. Thank you for indulging me and going back to org. We can go back to efficiency and then get into AI where there are a bunch more questions.

Of course. Here we go. Some thoughts on AI from a very grumpy 42- year- old. There’s two thoughts I have at the highest level around AI. Look, we’re all very excited about AI, and I think that there’s a lot of potential and that it is unquestionable that it will radically transform CX in the next 10 to 20 years. Two things I would keep in mind. First of all, the customer case, experience versus efficiency. Don’t just build an AI resolution that’s worse for the customer if it’s just to save money. So everyone gets extremely pumped about chatbots.

I will be honest, I don’t know. I’m on the internet. I have not had any great number of experiences at this point interacting with a chatbot that was better than getting a human is one. Will that change? Definitely. But it is not there today is my grumpy opinion. Number two is, keep in mind, I’ve had a lot of clients in my consulting life that say, I want to automate 100% of our customer. If you’re a CFO, you’re just salivating at that idea.

We can just get rid of our entire CX team with AI. That is unreasonable, but also it’s not smart for the following reasons, which is, I would always say, there’s this idea of acute versus enduring defects. Enduring defects are like, yeah, my plane is late. That will always happen. Why do you want to automate with AI the resolution of a defect that’s acute that should just be fixed? That is the worst of both worlds for everyone.

I think that as we get excited about AI, we still want to think, the best customer journey doesn’t involve, I just bought a sweater. I don’t need AI to be part of my journey, unless I have a problem that needs to be solved, necessarily. That’s one preamble. A framework.

Quick question off of what you just said. Curious what trends you’ve observed in CX on which aspects customers prefer to resolve themselves via self- serve, perhaps AI versus things they prefer help from a human to resolve.

Absolutely. A few things here. Let’s pull back from AI and flip the script and think about human beings. Right? When you are interacting with, when you have a problem with a business, the fundamental issue you have in your chest is, I am afraid that I am not going to be made right for this issue. Right? The best thing we’ve ever did at DoorDash was, and I’m sure if you’ve used DoorDash, you’ve experienced this, unfortunately. If you have a missing item, a late order, you go in the app, you click four buttons, and you get your money back.

People love it. Because they’re in control, it’s fast and easy, and it’s resolved really quickly. If you can mirror that experience with AI, God bless. But don’t sacrifice the anxiety of not knowing if I’ll get it. If I’m just sitting there trying to get a refund on my order, I’m futzing around with some AI bot, that is not good for me. And so if you can, great. If you can’t, then I would question it. But to your point of what trends.

One is, and this is true for both consumers and agents, every company has a massive knowledge base of a bunch of stuff, and maintaining the, it’s like, how do you set up your password? How do you onboard a new user, yada, yada, yada. This content grows and grows and grows, and it’s impossible to maintain, and it becomes the workflows for your agents, and it’s a mess. AI is incredible at distilling that, and a lot of great contact center software platforms will use that.

You know, other consumer- facing, consumer assist and chatbots, and then agent assist on the backend will, you know, scrub all of your internal knowledge- based content and spit out really good answers that are helpful. I’ve never been helped by an FAQ section, but when I get that information fed to me when I ask for it, it’s usually, like, pretty spot on.

I’ve had some exciting experiences recently. I use a cold email tool called Instantly, and they have a chatbot that has indexed all of their support articles and is pretty damn good, and it’s quick, so I empathize with that use case. It’d be sort of interesting. I also have been working with many, many chatbots, and, you know, human, agent, human, agent. It’d be almost interesting to look at an analysis of what queries end up having people just demand, please give me a person.

I think that’s exactly right. I think that the issues, like, where I see is that, you know, if it’s not a first contact, if it’s a second or third, that’s a problem. And if it’s a, you know, we saw this very much on the restaurant side at DoorDash. We had good success with AI on the consumer and dashboard side. When it came to restaurants, two things. One, those, like, a restaurant owner is not the, like, customer segment that’s going to give you any time with AI. Like, they want to pick up a phone and call and get a human being.

That is a cultural issue, not a technology issue. So, like, keep in mind, you live in the real world. Like, you know, technology is exciting, but, like, think about who the customer segment is you’re thinking about. And then two is the issues were far more acute and complex, and the likelihood that an AI bot was going to be able to solve them, especially for a physical world product, was low. Cool.

So, I’m curious to get these, can you walk us through these thoughts on AI? Yeah, of course.

So, I think, you know, such as it is, this is a, you know, a flow diagram of areas where I think that there are obvious and may not be obvious opportunities for AI usage. So, as I was saying, like, self- help and FAQs, yes. And I think that AI is especially valuable here because it transforms your FAQs from just like a, pardon the language, F off and do it yourself to something that’s more custom and bespoke, right? It feels like it’s actually being served to you, and that’s a better experience.

You then get into, okay, well, if you can’t be self- helped or, you know, self- served, what kind of an issue is it? I’ve said this twice now, if it’s enduring and going to be around forever, yes, by all means, build that AI because it’s going to pay dividends. If it’s just something stupid you should fix, just go fix it. If it’s, you know, we had an issue where like Discover cards wouldn’t work in Quebec. Don’t build it, just make it work. The next is pre- chat issue diagnostic. This is very, very helpful for a number of reasons.

And so, we did leverage AI through Salesforce to do this at DoorDash, issue diagnostic. So, if you’re an agent, not only can you take some time off the contact that you’re actually on it, which is efficiency gaining, but agents then start with an idea of who they’re talking to and what the problem is, which is really helpful for a human being to know what burning building they’re walking into, right? So, is this person angry? What is their issue? What is happening? So, that’s good. The human resolution, the human interaction, like, no.

And I think this goes back to the very first slide around, like, what is the point of customer service? The point of customer service is to, you know, both solve problems, but also demonstrate that as a company, you’re taking ownership of the problem and the resolution. And at some point saying you’re sorry, you know, taking a human out of the interactions where a human’s needed for that, basically demonstrate to your customer, like, think about it.

Like, if you’re angry about something and your product did not deliver, you want someone to look you in the eye and say that you’re sorry. And I think that there is a subset of contacts where that’s incredibly important, where you will never get rid of that human. And then finally, post- contact summary, obviously, right? Like, you have, you know, agents spending two, three, four minutes typing out notes. That is unnecessary with basically every tool you have today. And then agent quality assurance.

We had a team of, like, 500 people doing agent QA. I think the number is now zero. So, those are things that do not need to be done by humans.

So, in terms of, we had a question on establishing an automated first line of defense for incoming queries. Is that maybe, one, sometimes you can send them to the self- help, and two, maybe you can do a pre- chat issue diagnostic?

That’s exactly right. And if I think about most customer service teams, I think if you look at the, and this is unpaid, but, like, if you look at the off- the- rack software solutions like Intercom, Gladly, Customer, Zendesk, others, they have increased, either have or increasingly have good tools to do both of those things. So, you know, either.

answer with, you know, an FAQ answer that looks like it’s actually bespoke and made for you, or actually, you know, be able to ingest what your issues are and then kick it to an agent.

And so then another question, what’s the hype and what’s the reality of using AI to support issues? I think you shared some of the, hey, post- contact summary, really good use case. Quality assurance, really good use case. Can you share some of the, like, hype but the there’s not there yet that you’re seeing?

Yeah, I think that, and I would take all my thoughts with a grain of salt because, like, I think betting against AI as a rule is probably, just doesn’t feel like the right answer. But for today, I think that the hype is, as I was kind of getting at, is, look, everyone references the Klarna example of how Klarna has saved all this money and automated all these things. Well, Klarna is its own sort of like walled garden of, you know, sort of semi- financial consumer product.

And I would believe that it was able to build AI to ingest all of its internal information, churn through it and be able to provide automated resolutions to its consumers. I would not believe that that is the case for every business, and especially businesses that have increasing amounts of, like, real world operations. It’s just not, the physical world is too complex to have AI agents be able to, you know, handle every issue. So I think I’ve kind of touched on a lot of, you know, the hype and reality.

I’d say the reality is, you got to go back to, like, people who get excited about this just see it as the best application of AI because they’re excited about AI. I’d say from the customer experience lens, you got to remember, like, your goal is not to demonstrate how amazing AI is. It’s to solve consumers’ problems.

One more. We’ve got a lot of AI questions, and feel free to keep submitting them if you have questions that aren’t being answered. How do you leverage AI in high- service SaaS support situations, promoting customer enablement as a support model? And I think an interesting addition that I’ll add to this question is maybe how can you use AI to make support agents who are still humans more efficient if, like, high- touch, maybe kind of white- glove service is part of your brand?

Yes. So some, I think what’s interesting about AI in these, like, high, I mean, because the easy answer to say, well, it’s high- service. You want to have that relationship with your CSM or whomever who is managing your business.

One thing I think is really interesting is AI, in terms of being able to detect, like, declining satisfaction and churn risk, and so you start seeing, like, the, you know, whether it’s whatever CRM you’re using, a lot are building in the ability to combine all sorts of inputs, like, you know, your customer service contacts, you know, time between your MBRs, like, the transcripts of your MBRs, you know, whatever it is, all this information and giving you a score of, like, hey, among your clients, like, this subset is at higher churn risk because of these inputs.

And so I think that that’s particularly interesting. Yeah.

Actually, I come to think of it, customer health scores are going to change, like, there’s going to be so many more inputs and be able to crunch so many more inputs. One more new question coming in. What do you suggest when tickets or issues are received at your business are very technical and require field expertise to answer the questions, yet we’re being used by internal leadership to use AI? Yes.

It’s always a bit challenging to know without knowing the actual business itself, but what I would say is, I mean, if I understand the question, it almost comes back to, like, a very simple answer of, like, can you? It’s like, I’m surprised to learn that if you need detailed expertise and contextual information that AI would be able to do that. And so, you know, it might be, and I think thematically through this conversation, like, there is a huge push to, like, find new ways to leverage AI.

My thematic answer through a lot of this is humans are most effective at dealing with other humans. Everything else behind the scenes, to sort of use the duck example, like below the water line is an opportunity for AI. So in this case, without knowing what technical and field expertise you have, are you able to ingest and triage tickets based on what’s being asked to get it routed to your field team in a more efficient way? So maybe you need a smaller group of experts. I’m not sure based on this exact example, but I would go back to what is the point?

Like 90% of CX organizations, when I talk to them, so our budget at DoorDash was a billion dollars. And so, you know, you think about like a 1% change of that budget is like unbelievably material to the bottom line of the company. For 90% of CX teams, their budget is not material. Like it’s five or 10 people. And what I would say is like, if you have a client that even has a team of like 60, and I said, if you reduced your cost by 50%, would anyone notice? And she’s like, no.

So my point would be, yes, you’re getting pushed to use AI, probably because people are excited about it. And every company wants an AI story to tell, try to find an AI story that accomplishes your goal. And that might be not customer facing.

I think that makes sense. We’ve had another expert in more of a go- to- market context, create kind of a two- by- two of where, you know, is this internal or external facing, and is the data internal or external? And, you know, internal facing use cases are much safer to start with, because you’re not putting that, like you’ve got a line of human judgment in between the AI and your customer, which makes a ton of sense.

Exactly right.

Do you have other slides or should we go into open Q& A? I think that’s it.

There’s some appendix slides that are kind of eye charts around, you know, tech stack and things, but we can go to Q& A. Cool.

So I will keep going through some of the pre- submitted questions, but I’ll, you know, prioritize anything that’s coming in live. So let’s look at, sorry, one second.

Oh, is there a new one?

Perfect. Thank you. How do you advocate for CX optimization tools when the roadmap takes priority?

Great question. And this is probably relevant to every CX org on the planet. There’s some thematic things and some tactical things. I think that also the tactical, because it’s the easiest. I think that a lot of CX teams go wrong by asking, like they’ll get in front of like a product leader or a CEO and then ask for the world. And so what was very successful for us at DoorDash, because you can imagine we had the exact same issue, right? Where like, you know, spending money on my tools versus spending money on like a consumer facing product.

The ROI like was never there for me in the early days. So what I would do is like ask for one thing, just like, you know, yes, the world’s on fire and nothing works, but just like ask for one thing. One thing that’s been very well thought of and prioritized is pretty hard to say no to. 50 things really easy to say no to. And then the second piece I would say is a lot of times, if I’ve seen it once, I’ve seen it a hundred times where the first order of business of every new CX leader is we need a new contact center software platform.

It’s like literally as old as time. And so what I would say is spend the time framing the ask in terms of the business need and why this is important and why either on the cost side, you can put real dollars to it. Like, hey, we get the software, we can reduce our handle time by 40% and thus our costs by 40%. That’s easy. Or you can say, hey, CEO, you know, the emails you send me every day from angry customers, this one thing will make that get cut in half. So again, I think through the entire answer, it’s just like focus of what the ask is.

Awesome. One more. Our customer mainly reached out to their CSM instead of using self- help options, increasing support tickets. How can we drive better adoption of self- service resources?

Yes, this was our problem as well at DoorDash on the restaurant side. So restaurants would reach out to their account manager and I’ll give an answer that’s like going to make nobody happy, it’s like don’t, like don’t try to have them adopt self- service resources because they don’t want to.

Again, it goes back to the thing of like, you know, whether it was a new product or anything within our customer experience team, like the question that Tony, founder of DoorDash, would pepper us with every meeting, you know, DashPass or new country launch or new feature around, you know, ASAP estimate, whatever it was, why is this better for the customer? And if there wasn’t a coherent answer for why it was better for the customer, like we wouldn’t do it.

And so what we did on the restaurant side in this exact example is we just made a team of people that supported our CSMs that restaurants could reach out to directly because they wanted to. And so I think if you can make a self- serve resource that’s so good that it’s a better experience then like go for it. That’s a good solution. I’m skeptical you can.

And so in that case, I would say, encourage them to reach out to like, you know, their CSM or, you know, have a couple direct lines to like, you know, CSM partners who can help with support issues and then get rid of those issues so they don’t have to contact in the first place.

Here’s another one that’s kind of fun. So if you were to grow a CX function from zero through series D at a B2B software company, how would you do it? So I guess who are your first hires and how does it grow?

30 seconds on that?

Yeah, exactly.

I’ll give you a couple things that I would do. It’s not a playbook, but things that are, I think, important. One, hire generalists.

I would say that, you know, if you’re building, yeah, if you’re going to go from zero through series D and you’re going to have a lot of complex problems to navigate, I would say either hire people from the CX world who have extremely strong foundational toolkits around like analytical problem solving and, you know, strategic thinking, or just go get people who have that and then teach them the CX side so that, you know, those people, those people will scale. Like, you know, we always said hire for slope, not Y intercept.

So hire people that have a lower starting base, but have a higher slope, and they will be able to scale that team and probably find better first principles solutions. Second, I would start doing all support internal before I even consider outsourcing because you have to like, you know, have, you’ve got to have the answer to the test before you, you know, give other people the test.

And then finally, I would choose a really good software platform that was extensible and would carry us through that period because, you know, rinsing and replacing your tech stack every couple of years is just a hot nightmare that everyone should avoid.

And how, I guess, how, I know that you can’t give, it depends upon the product, but how quickly as you’re adding customers, do you add support reps? And do you ever, like, early on, do you ever want to over index on, you know, hiring more people, onboarding more support to make sure that you’re in a good spot for scale?

Yes. So the answer is always, it depends because what we, I mean, DoorDash was in hyper growth for like three plus years where we were growing like five to 10% per week. And, you know, you can imagine that you can turn the, turn on the funnel of customer acquisition faster than you can go and hire new customer support teammates and train them, right? So the key, you know, there’s a key there, which was like aligning with your, you know, your finance team’s forecast, planning and executing against that and doing your best to adjust as you go.

It comes back to the business model, right? For us, these are live issues. People are angry and you need to answer like within five seconds. And so to your point, Kate, like, yes, we absolutely overstaffed. And we also overstaffed on chat, which you could get to concurrent chats. So we can always flex up from doing one to two to even three chats at peak volume to try to like flex and absorb that. And it goes back to like your Northstar metrics, right?

Like our Northstar metric we thought about the most was that SLA, like 95% of contacts answered within like 10 seconds. And so we would say to hit this number based on the massive volatility we have, we do need to have like 10% more people and it will cost five cents more per delivery to do that.

Cool. Well, we are almost out of time. That was a meaty last question. Well handled in three minutes or less. Thanks so much for taking the time today. We will share Steve’s information in the recap along with the recording. We’ll also mention your website. So if you want to get in touch with Steve, we’re happy to help make that connection. Thanks everybody for joining today.

Thank you very much. And thanks, Kate.