In May 2026, Klaviyo turned on Custom Skills for its Customer Agent. The framing Klaviyo used in the partner announcement was specific and worth quoting in full: "Brands need AI agents that reflect how they actually operate. That's a build. Agent Orchestration is the service opportunity, and it sits on Klaviyo's CRM foundation, not a helpdesk."

That last clause is the whole story. Customer Agents in 2026 are a different category of thing from the support chatbots most brands have been running for the last three years. The difference is structural, the implications are commercial, and the founder led DTC brands that figure out what to build first will have a real advantage by Q4.

Here's what changed, what it means, and the use cases worth thinking about for a brand running $5M to $20M.

What Custom Skills actually are

Custom Skills let you build an AI agent that reflects how your specific brand actually operates. Your voice. Your service rules. Your escalation logic. Your product knowledge. Your customer history. It runs on Klaviyo's CRM data, which means it has access to the same picture your retention team uses to make decisions.

Functionally, you can think of Custom Skills as the set of capabilities the agent has and the rules that govern how it uses them. A skill might be "answer questions about shipping for any order placed in the last 30 days, in the brand voice, escalating to a human if the customer is a VIP". Another might be "respond to product enquiries by referencing the live catalogue and only recommend products in stock". A third might be "initiate a win back conversation with profiles that haven't purchased in 90 days, with offers that match the customer's previous purchase pattern".

The agent uses these skills to do work. The work isn't generic. It is configured to match how a particular brand wants to operate, with access to the data needed to do it properly.

Why this isn't a helpdesk

Most of the AI customer experience tooling that's been sold to DTC brands since 2023 has lived inside helpdesk software. Zendesk, Gorgias, Front, all of which have shipped AI features. Those features are useful for what they do, which is help a support agent answer tickets faster. They are limited by what they can see, which is the support ticket and maybe an order history.

Klaviyo's framing is that the Customer Agent shouldn't sit inside the helpdesk. It should sit on the CRM. The reason matters.

A support ticket tells you what the customer asked about today. The CRM tells you who they are. What they've bought. How often. When. What they've ignored. What they've engaged with. Where they are in the lifecycle. Whether they're trending up or down in value. That is a very different set of information to bring to an agent's response than just the ticket.

The practical implication is that an agent built on Klaviyo's data layer can answer questions and take actions a helpdesk agent cannot. It can say "I see you bought the matte foundation in shade four last September, the new formula in your shade is in stock, would you like me to add it to your reorder". It can recognise that the customer asking about shipping is a top decile spender and route the conversation differently. It can spot that a profile sending a complaint is in the post purchase flow and trigger a different path.

That isn't customer service. That is retention work, automated, with full context.

What this means for the brands that should care

Custom Skills are not for every DTC brand. The brands that should be thinking about them have three things in common.

First, enough volume to matter. If you have under 500 customer conversations a month, the build cost will exceed the operational gain. Custom Skills make sense at scale, which for most brands means $5M in revenue and up.

Second, a real brand voice. The whole point of Custom Skills is that the agent reflects how you actually operate. If your brand voice is generic, you get a generic agent. The brands that get the most out of this are the brands with real tone, real personality, and real service standards. The wellness, fashion, and lifestyle brands at the premium end of the market are the obvious examples.

Third, a willingness to think operationally. Custom Skills are not a plug and play feature. They are a build. The brands that succeed treat it as a process design problem. What do we want the agent to do, what do we want it to escalate, what do we want it to never touch, how do we want it to sound, what is the data it can read from, what is the data it can write to. The brands that don't think this through end up with an agent that says nice things and accomplishes nothing.

Use cases worth building first

Three use cases that pay back fastest for a DTC brand at $5M to $20M.

Post purchase support and reorder. The single highest leverage use case for most brands. A profile lands in the post purchase window. The agent reaches out via email or chat, references the specific product, asks how it's going, and offers a reorder path tied to the customer's purchase cadence. Done well, this lifts repeat rate without lifting discount spend. Done badly, it sounds like a robot. The difference is in the skills configuration and the voice training.

VIP and loyalty conversations. The top 10% of customers in a DTC brand usually account for 40% or more of revenue, give or take, depending on the category. They deserve a different conversation than the rest of the base. A Custom Skills agent can be configured to recognise VIP profiles, respond differently, offer access first rather than discount first, and route to a human for anything that needs personal attention. Most brands today do this with manual segmentation and template emails. The opportunity is to do it with an agent that responds in real time.

Win back and reactivation. Profiles that haven't bought in 90 to 180 days are usually contacted with a discount sequence and then dropped. The 30% of those profiles that would have come back without the discount cost margin. A Custom Skills agent can be configured to read the specific customer's previous behaviour, identify what kind of reactivation message is likely to land, and try a sequence of approaches rather than a single template. The margin opportunity is meaningful.

What it costs to build (loosely)

The honest answer is that it depends on what you're building. A simple post purchase reorder skill, configured by a team that knows what they're doing, is probably a two to four week build. A full Custom Skills programme across post purchase, VIP, and win back is closer to six to ten weeks, with ongoing refinement after that.

The bigger cost is the thinking. Most of the work is in the configuration. What rules, what voice, what escalation, what data, what guardrails. Brands that try to skip this stage end up paying for it later in agents that say weird things or take actions they shouldn't.

If you don't have a retention team that can think about this work properly, the right move is to bring in a partner. The work compounds, but it has to be designed by someone who understands both the platform and the brand.

Why this is the service opportunity Klaviyo called out

When Klaviyo writes "Agent Orchestration is the service opportunity, and it sits on Klaviyo's CRM foundation, not a helpdesk", they are saying something specific to partners. The work to build, configure, and run Custom Skills agents on top of Klaviyo's CRM is a service that didn't really exist twelve months ago. It does now.

For DTC brands, the practical implication is that the conversation with your retention agency or internal team should change. Six months ago, the conversation was about which flows to build and which campaigns to run. The conversation now should also include which Custom Skills agents to build, what data they should read from, and what work they should be doing on the brand's behalf in the background. That work compounds in a way that campaigns do not.

For brands without a current retention partner, the question is sharper. The agencies that are building this work right now will have a six to twelve month lead by the start of next year. The agencies that aren't will be behind.

What founders should do this quarter

Three questions worth answering before the end of June.

One, where in your customer lifecycle is the biggest gap between what you should be earning and what you are earning? For most brands at $5M to $20M, the answer is somewhere in post purchase or VIP. That is the use case to build first.

Two, what is your team's capacity to design a Custom Skills agent properly? If the answer is "we don't have one", the choice is between hiring or partnering. Both are valid. Drifting is not.

Three, what is your current Customer Agent doing in Klaviyo, if anything? If you have it switched on with the default skills, that's a starting point. If you don't, the first task is to turn it on and configure the basics before you build anything custom.

This is the kind of work that pays back through the rest of 2026. The brands that build it first will see the operational benefit by Q3 and the commercial benefit by Q4. The brands that wait will spend H2 watching their competitors compound.

If you want to talk through what a Klaviyo Custom Skills build could look like for your brand, get in touch with Webtopia and Oaks. We're already running this work and we'd be happy to walk through specifics.