Support is becoming the layer that translates the business: Conversation with Sarah

This interview is part of the SparrowDesk Spotlight series, where we sit down with the people rethinking how customer support actually works.

Sarah has spent over a decade building and running customer support and CX operations. Most recently, she set up support and operations at an AI image evaluation company, having previously served as Director of Support and CX Operations at Alyce.

Today, she's exploring fractional executive work, helping businesses understand the operational problems and bottlenecks that often go unnoticed until they start affecting customers and revenue.

Her perspective on AI in support is practical. The technology can take repetitive work off a team's plate, but it also changes the nature of the work humans are left with. And while many companies are busy asking how many tickets AI can deflect, Sarah thinks there are more fundamental questions they should be asking first.

We sat down with her to discuss how AI is changing support teams, why documentation matters more than ever, and what customer support needs to become next. Here's the conversation, lightly edited for clarity.

On why customer support sees problems before everyone else

Q: You've spent more than a decade in customer support, and your work now extends into operations and broader organisational problems. What led you from support to this wider view of the business?

"Support is in an interesting position to see everything that comes in. We hear all the problems and requests from customers. We start to recognise even things that aren't documented, the ways people talk about things, and where the actual friction is.

We're also uniquely positioned to have close relationships with product, marketing, sales, engineering, finance, and pretty much everybody in the company.

What I started seeing was that so many things that land on support's plate aren't necessarily things support can control. We see a particular product friction causing problems, and leadership might be choosing to prioritise something else.

But we can see that fixing this problem could save a significant amount of money and prevent a lot of churn.

That's what pushed me to move further up the line and tell executives: here's what to look for, here's where your valuable resources are, and here's how you can dig yourself out of a hole you don't realise is coming."

The point is that support doesn't just see individual complaints. It sees patterns across customers, products, and teams. The challenge is getting that information in front of the people who can actually do something about it.

On why deflection shouldn't be the starting point for AI

Q: When founders and support leaders start thinking about AI, the conversation often revolves around ticket deflection. What do you think they're missing?

"I think we still have some growth to do here. When I'm talking to founders, they're often asking, 'How many tickets can I deflect?' And I'm like, that's not where you need to start.

I spoke to a founder who wanted to bring me on for a one-year contract to set up a support team. She wanted to focus on which tool to use, how many tickets to deflect, and how much to reduce.

So I asked, 'How many tickets do you have now?'

They hadn't launched yet.

I asked about the product, and they didn't have those questions figured out either. Until you have that data, I can't give you numbers."

For Sarah, the first step is understanding the problem before deciding how AI should solve it. A team that hasn't established its support volume, common customer issues, or product friction has very little basis for setting meaningful automation targets.

AI can absolutely take repetitive work off agents' plates. But deciding what to automate should come after understanding what customers need and where the business is struggling.

On the documentation problem nobody should ignore

Q: If a company is just getting started with customer support, where should it begin? And what about an established enterprise looking to introduce AI?

"For a startup, I think they should start with collecting the data of what they have. Where are all the customer inquiries coming from? Are people calling, chatting, emailing, or asking questions on Reddit? Are they coming to your website from AI tools?

You should definitely have a knowledge base. People are asking AI before they're going to your website, so you need to make sure there's a presence out there answering questions the way you want them answered.

Then the next thing I'd say is tagging. Any inquiries or information you have, tag it. What's the product area? What problem did they face? Was it a bug, a feature request, a general complaint, or simply that they didn't know how to do something?

That's how you build the knowledge of what you need to do next."

Her advice for established companies is slightly different. They probably already have documentation, but years of accumulated articles, internal guides, and bot-generated content can leave them with a different problem: too much information, some of it outdated or contradictory.

"Start with documentation because they probably have levels of documentation. If you've been doing this for a while, people have been writing articles, bots have been writing articles. There's so much information.

Start with a full audit before you release any bots. The bot will find things that are six years old and give an answer, and you're like, 'That was 40 versions ago. Why is that even out there?'"

Sarah also recommends deploying AI internally before putting it in front of customers. Let the support team ask the bot questions, identify gaps, and fix the documentation before customers depend on it.

And there's a good reason to involve the people closest to the work.

"Everybody who interacts with support knows the product and the company better than anybody else. They are the most authoritative people to make those fixes and corrections."

On the work AI makes harder for support agents

Q: AI can take care of routine queries, but what are you hearing from support teams about how their jobs are changing?

"They like the support AI does to help them do their job. A lot of them are very worried about burning out, because the tickets are harder. The work they do is more emotional and takes a lot more strategy, thinking, and planning."

Sarah describes a change in the composition of support queues. When AI handles familiar, repetitive questions, agents are left with the issues that require more investigation, context, and judgement.

That affects even experienced agents.

"If you're tier two or tier three, you have tickets that you see a lot and are easy to answer. You see them a lot, and AI answers them now. So there aren't the tickets that are familiar. Nothing is familiar, which takes a lot more context switching and creative thinking.

You have fewer tickets, and they're harder. It used to be like, 'I had a rough day, but I knocked out 30 tickets. I feel pretty good about that.' Now you're like, 'I have three tickets. They're all pending.' It just isn't as satisfying."

It's an important distinction. A reduction in ticket volume doesn't automatically mean a reduction in workload or stress. Agents may be handling fewer conversations while carrying more cognitive and emotional weight in each one.

For Sarah, there's also a longer-term concern: what happens when new support professionals no longer get the opportunity to learn through those simpler, everyday problems?

On why new agents still need to learn the hard stuff

Q: If AI handles the easier queries, how do new support agents develop the skills they'll need to solve more complicated problems?

Sarah compares the situation to learning mathematics.

"When you're learning maths as a child, you might have complicated calculators and computers, but if you give that calculator to a seven-year-old, they don't learn the mental process of how to add, subtract, multiply, or think about algebraic equations.

We have to practise the hard stuff in order to do the hard stuff."

Her concern is that companies are so excited about deploying AI that they're skipping the learning process that helps people build problem-solving skills.

"We're not building those muscles. We're just saying, 'I'm going to give you the answer that seven plus seven is 14. Don't think about it.'

But you have to have seven squares and seven squares, count them all up, and understand how seven plus seven is 14."

In support, that practice often comes from investigating recurring questions, navigating systems, and working through problems without an immediate answer. If AI removes those opportunities entirely, newer agents may struggle when they encounter situations that automation can't handle.

Sarah doesn't offer an easy fix, and that's part of what makes the challenge worth paying attention to. Automating today's repetitive work is one thing. Making sure tomorrow's support professionals can handle the complex work is another.

On using AI to make customer insights more useful across the business

Q: Beyond answering customer queries, where do you see AI making a real difference to support operations?

"With AI, we can get so much more custom in what we review in our own products. We can really dig into tickets at a volume that no human eye could do before."

Sarah sees particular value in using sentiment analysis and large-scale ticket analysis to understand customer problems, coach teams, and improve how information is shared across departments.

But the opportunity goes beyond analysing more data. It's also about communicating that data in a way each team can use.

"Every specialty has its own language. Now I can say, 'Here's what I've been getting from my product team. What are they actually focusing on based on what they're writing?'

Then I can tailor my reports to product to what they're looking for and the language they speak. I can tailor my response to marketing based on what they're really focused on.

It allows us to do some translation that we wouldn't be able to do otherwise unless we'd spent significant time with that team."

This is where support can become much more valuable to the organisation. Rather than simply reporting ticket volumes or passing along individual complaints, teams can use AI to identify recurring themes, organise customer feedback, and turn it into insights that product, marketing, and leadership can act on.

On the metrics that actually tell you whether support is working

Q: CSAT, first response time, and ticket volume have been standard support metrics for years. Do they still matter in an AI-driven world?

Sarah has long been sceptical of CSAT as a standalone measure of support quality.

"That score, that rating, is influenced by people's satisfaction with the product and their satisfaction with the policies. It's not just the agent. So I never score people on human review."

She doesn't suggest that response speed is irrelevant. Instead, she believes teams need to understand what speed means for their particular customers.

"I still think it's important to ask whether somebody got a reply quickly. You have to do your own maths on how fast is important to your customers.

I've had customers who want everything resolved within an hour because it's a high-urgency product. Other customers have a low-urgency product, and as long as you get to them this week, they're happy."

The same principle applies to choosing broader performance metrics. What matters depends on the product, customer behaviour, and the outcome the business is trying to achieve.

For a consumer business selling water filters, subscription rates, repeat purchases, and upsells might be useful indicators. For an image analysis software company, success could mean more users adopting the product, sharing their work, or expanding usage across the organisation.

The common thread is that the metrics should reflect what customers are trying to accomplish, rather than simply how many interactions a team processes.

On customer effort as a better measure of success

Q: If traditional support metrics don't tell the whole story, what would you prioritise instead?

Sarah sees considerable potential in customer effort: how difficult it is for someone to move from a problem to a resolution.

"How hard was it for a customer to go from problem to resolution? And that can include how many escalations, how many departments of support they had to talk to, whether they talked to the bot, whether they started at the knowledge base, how many different people they talked to, and how many times they opened a ticket on that particular problem."

She sums up her approach to support with a straightforward principle:

"Basically, my North Star for my teams, since I work in enterprise software, has been: get the customer back to work. They bought us to do a job, not to talk to us.

How quickly can we get them back to doing their job?"

It's a useful way to reframe support performance. A fast reply is only valuable if it helps the customer move forward. A ticket closed by AI is only a success if the underlying problem is actually resolved. And a good customer experience shouldn't require someone to spend their day navigating a company's support process.

On why support needs to be involved before a product launches

Q: Are companies becoming better at recognising the strategic value of support?

Sarah has seen progress. More executives now recognise support as a specialised function, and some are involving support leaders much earlier in product development.

"I have executives in my network now who will come to me and say, 'Okay, I have this idea for a product, and I want to make it supportable.' And that was a question that anybody asked before. Never. That was something that you got as an afterthought, usually."

She sees this as part of a broader shift in the profession. Support has evolved into more specialised functions, from onboarding and customer success to customer marketing, billing, and AI operations.

AI is also creating new responsibilities around knowledge management, tooling, and integrations. These roles bring support closer to product and engineering, because successful automation depends on reliable documentation and the ability to connect systems.

Sarah believes support teams are increasingly recognised as experts in the pitfalls of product development because they've seen how products fail customers in the real world.

The next step is for more businesses to act on that expertise before problems reach the customer.

On where companies should actually start with AI

Q: If a company wants to introduce AI into its support operations, what should it focus on first?

Sarah's answer comes back to a question that applies to startups and established enterprises alike: where is the pain?

"Where is your pain? Go after that. What's the data? Tag, collect data, find out where your team is spinning their wheels. Where are they getting stuck? Where are they doing the same thing over and over again? Talk to your team. What are they sick of doing? Those are good places to start."

She also recommends involving newer team members in documentation audits. People who are learning a product for the first time can spot confusing instructions and broken workflows that experienced employees have learned to work around.

"When we're experts at something, we patch a lot of flaws in our head. We overlook things very easily because we translate and correct in our head as we go.

But somebody new who's going through the flow for the first time doesn't have all that knowledge. They don't know what to correct. And so they are the experts at finding the beginner stuff."

It's a practical way to approach AI adoption without turning it into a technology-first exercise. Understand the work, find the recurring friction, improve the information people rely on, and then identify where AI can help.

On what customer support might look like in five years

Q: If we meet again three to five years from now, what do you think will look completely different about customer support?

Sarah offers two possible futures, one considerably less optimistic than the other.

"My nihilist answer is we find out that turning everything over to the bots was a complete and utter disaster, and we lost touch with humanity and everything crashed out. Because if you stop having jobs for people to do in software, nobody's there to buy software because they're not employed."

Her more hopeful vision is one where AI takes care of the work that doesn't need a human, allowing support to become much more personal.

"The happier answer is that we have found a way to let the bots do all of the stuff that doesn't need a human, and we've moved over into very human support.

So I know my customers very well. I understand exactly why they're using my product. I can customise my product. I think AI will allow us to do a lot more customisation."

She imagines products becoming more adaptable to individual customers and their needs, rather than giving everyone the same interface and expecting them to work around it.

For Sarah, that future also means a closer relationship between support, customer success, sales, and marketing. Understanding why customers buy a product, what they're trying to accomplish, and what success looks like for them could become just as important as resolving their immediate problems.

As more interactions become automated, the human relationship may become even more valuable.

More from Sarah

Sarah is a customer support and CX operations leader who has worked across AI and SaaS businesses, including Alyce and an AI image evaluation company. She is currently on a sabbatical and exploring fractional executive work focused on support operations and organisational bottlenecks.