Mike HeapMike is an experienced Product Manager who focuses on all the “non-development” areas of My AskAI, from finance and customer success to product design, copywriting, testing and more.
Table of Contents
- DTC hosiery AI customer service results on Gorgias at a glance
- What does the brand do?
- Which helpdesk does the brand use?
- How did the brand train their AI customer service agent?
- When did they decide to turn on 'direct replies' to customers?
- What was the biggest thing they did to improve their AI agent's resolution?
- How does the brand customize their AI agent setup to work for their business?
- Tone and communication style
- Escalation and handover
- Notion as the editable knowledge surface
- What impact is their AI customer service agent having now?
- Where does the brand go from here?
Created time
May 27, 2026 03:46 PM
Title length (<60)
Author
Last optimised
Sep 15, 2026
Ecomm?
Tags
Image
Publish date
May 27, 2026
Video
Slug
dtc-hosiery-ai-customer-support-case-study
Featured
Featured
Type
Case study
Ready to Publish
Ready to Publish
A DTC comfort-hosiery brand runs My AskAI as an internal Copilot inside Gorgias, drafting note replies on ~2,400 monthly tickets, with live Shopify + User Data API lookups behind every draft.
You probably run a DTC brand on Shopify and Gorgias and you're working out what AI inside the inbox actually looks like in practice. Or you've heard about teams running AI Copilot mode, with no direct customer replies, and you want to see what that kind of deployment looks like at real volume.
Either way, I've got you. This team (a DTC comfort-hosiery brand) sits at almost exactly this profile. Around 2,400 tickets a month landing in Gorgias, a short weekday support window, and a deliberate choice to put AI to work as an internal Copilot rather than as a customer-facing reply.
Today we draft an internal-note answer on each of those ~2,400 tickets. Every draft is grounded on three live data layers: the team's Notion playbook through our Notion connector, live order and product data through our Shopify connector, and live customer-level fields through the User Data API.
Agents send, edit, or discard. Direct replies to customers are off for now, and that's by design.
Here's how it came together.
DTC hosiery AI customer service results on Gorgias at a glance
Fact | A DTC comfort-hosiery brand |
Industry | eCommerce, comfort hosiery and compression socks (DTC on Shopify) |
Helpdesk | Gorgias (email tickets and on-site chat) |
Support volume | ~2,400 tickets/month |
Key features used | Notion connector, Shopify integration, User Data API, Communication Guidance, Handover & Escalation Guidance |
Knowledge sources | A Notion playbook via our Notion connector, live Shopify order and product data, live customer fields through the User Data API |
Go-live mode | Internal-note Copilot mode only, direct replies deliberately switched off |
What does the brand do?
The company is a direct-to-consumer brand built around comfort hosiery, with a range that spans everyday socks and more specialized lines.
The brand positions itself on comfort above all. Athletic performance and medical use are secondary at most.
They have a large customer base and an unusually high volume of reviews for their size, and that shows up in the kind of support questions the team gets.
Customer contact runs through email and phone, with the phone line open for a short weekday window.
Which helpdesk does the brand use?
They run customer support out of Gorgias, the Shopify-native helpdesk most DTC brands in this part of the market default to (yes, including the ones we onboard most weeks). Email tickets and on-site chat both flow into the same Gorgias inbox.
The question every Gorgias DTC team faces at some point is what to do about AI inside the inbox. Gorgias's own native option (Gorgias Automate, and the newer Agent AI) is the path of least resistance, and the install takes one click.
The trade-offs that send DTC apparel teams looking past it are well-known across the category:
- A per-resolution pricing model that charges on outcomes the customer can disagree with (we walk through the maths on Gorgias AI vs Intercom Fin pricing and on Gorgias AI vs Zendesk AI pricing in two separate breakdowns).
- Quality complaints that recur in DTC operator communities.
- A depth gap around live order and customer data, which means most AI replies are FAQ-shaped and know nothing about the customer's account.
This team chose to run us on top of Gorgias instead. We document the full integration on the My AskAI for Gorgias integration page, and the My AskAI app is listed on the Gorgias App Store.
The cost side lives on the Gorgias ROI calculator if you want to model it for your own volume. The helpdesk choice itself is unremarkable: Gorgias because the storefront is Shopify, and the rest of the support stack is built around Shopify too.
How they deployed it is where the rest of this post lives.
How did the brand train their AI customer service agent?
Training started with the team's own playbook. They do have a public help surface on the site, but the AI's knowledge base is a Notion page connected through our Notion connector.
The Notion page covers the things an apparel comfort brand has to get right on every reply: the return policy specifics, the sizing rules for the specialized lines, fabric care guidance, the shipping windows by region. Ops can edit the Notion page directly and we re-train on the change automatically (no separate help-center build, no engineering involvement).
The second source is our Shopify connector. Gorgias doesn't have a traditional help center the way Zendesk or Freshdesk do, so the Notion page does the work a Help Center connector would do elsewhere. Shopify, meanwhile, handles everything order- and product-shaped: live order status, product detail (size charts, material composition), variant availability.
When a customer asks where their order is or what fiber the ankle socks are made of, the answer grounds on the live Shopify record.
The third source (and the one that does the heavy lifting on draft quality) is the User Data API. We use the User Data API to pull live customer-level context into every reply, beyond what the Shopify connector exposes by default.
Whether a customer is inside their 30-day return window, whether this is their first compression-sock pair or their third, whether the order was a gift, whether there is an open return or refund in flight: all of that arrives in the AI's context before it drafts the reply. It's the single highest-leverage thing on the training stack for this account.
The four sources stack into one Copilot setup: the Notion playbook for what the company knows about its own products and policies, Shopify for what the platform knows about every order, the User Data API for what the team knows about every customer, and the Guidance layer (covered in the next-but-one section) that decides how all of that gets used.
When did they decide to turn on 'direct replies' to customers?
They haven't, and that's deliberate. We're deployed in Gorgias in Internal Notes / Copilot mode only.
Every AI reply lands as a private draft note on the ticket, ready for a human agent to read, send, edit, or discard. No customer ever sees a reply that hasn't been touched by an agent first.
For an apparel comfort brand, the reasoning is simple. Sizing and comfort questions carry judgment, and the cost of a wrong recommendation on a compression-sock sizing question is a return cycle (and a customer who didn't get the comfort they bought).
Some of the range sits closer to the medical end of comfort than the rest. Even so, the language people use around circulation, leg fatigue, and sensitive feet is exactly the language a team will want an agent to check before it ships.
Brand voice is part of the product on a comfort brand. The team wants every reply agent-checked until they trust us on tone as well as content.
The short weekday support window is part of the picture too. The Copilot's job is to make sure those hours go as far as possible.
Every agent signs in to an inbox where we've already drafted the answer on every open ticket, using the live Shopify order, the live customer record, and the Notion playbook. The agent's work becomes review-and-send.
Going direct to customers is the obvious next step. Sequencing knowledge depth first is a deliberate choice. Barn Owl, another Gorgias brand, made the opposite call and went direct from day one.
What was the biggest thing they did to improve their AI agent's resolution?
If we had to point at one lever that did the heavy lifting on draft quality here, it's connecting the User Data API on top of the Shopify connector.
Shopify on its own gives us the order. The User Data API gives us the customer. Together they turn a generic "your order shipped on the 14th" draft into one the agent can send in a single click, because we already know whether the customer is inside their return window, whether this is the second compression-sock pair they've ordered, whether the order was a gift, and whether there's an open return or refund in flight.
The alternative is what most AI replies inside a Gorgias inbox look like: FAQ-shaped retrieval that answers the literal question without the customer context.
(We've seen the same pattern enough times now to spot it in the first ten drafts of a rollout.) A customer asks where their order is and the AI explains how shipping windows work. A customer asks about a size-up and the AI quotes the size chart.
Both technically correct, both miss the point. Once Shopify plus the User Data API are plugged in, we answer the same questions with the customer's actual order, their actual past purchases, their actual return window.
Luckily, that difference is one config change. The draft is closer to what an experienced agent would write themselves.
Measuring this in Copilot mode needs a reframe. The standard case-study metric (AI resolution rate) assumes the AI is replying to customers.
In Copilot mode, the right thing to measure is different. We look at three things instead:
- Draft acceptance: what fraction of drafts the agent sends without editing.
- Handle-time delta: how much faster a ticket gets resolved when the agent starts from a draft.
- Source-grounding rate: how often the draft was built on the live data layer.
The team doesn't report draft-acceptance or handle-time numbers as customer-facing metrics today. The qualitative read from the team: agents start every ticket with context they used to look up themselves (across Shopify, the customer record, and the Notion page).
That change is what they felt first.
Guidance plays the supporting role here. Communication Guidance sets how the draft is written. Handover & Escalation Guidance marks the ticket categories where we skip the draft and tell the agent to step in.
The two work alongside the live-data stack. Shopify and the User Data API supply the facts on the ticket, and Guidance decides what we do with them.
How does the brand customize their AI agent setup to work for their business?
Beyond the four-source training stack, the team has tuned us in three specific ways.
Tone and communication style
This is a comfort brand, and comfort brands sound a particular way. Communication Guidance is where the team sets the rules for how we write: sentence rhythm, sign-offs, when to use lists, when to keep replies short.
There's also a specific instruction to keep the language on the more specialized lines out of clinical / medical-advice territory. They are consumer comfort products, and we're guided to talk about them that way: warm, helpful, and a long way from white-coated.
The reader on the other end is buying socks that feel good.
Escalation and handover
Some tickets should never get an AI draft, regardless of how confident we are. Handover & Escalation Guidance is where the team draws those lines.
The categories they've drawn lines around include:
- Refunds beyond the 30-day return window.
- Suspected quality issues on a specific batch.
- Anything that reads as a health concern around the compression products.
- Gift-order ownership disputes.
- Sizing complaints that imply a product defect rather than a fit problem.
All of these route to a human with our context attached but no draft attempted. The agent picks up the ticket already knowing what the customer asked, what their order history looks like, and what flagged the escalation in the first place.
Notion as the editable knowledge surface
The Notion connector turns the team's internal playbook into our training source. Ops edits a Notion page; we re-train on the change.
That arrangement does more than save a knowledge-base migration. In Copilot mode specifically, it changes how agents relate to the AI (luckily, in the right direction): agents trust the drafts more because the source of truth is a document they themselves edited.
When something in a draft is wrong, the fix is to open the Notion page and correct it, and the next draft on the same topic is better. The feedback loop is short, visible, and lives inside the tools the team already uses.
What impact is their AI customer service agent having now?
The impact today, in short:
- ~2,400 tickets a month receive an AI Copilot draft note inside Gorgias.
- Every draft is grounded on three live data layers (the Shopify order record, the customer's User Data API fields, and the team's Notion playbook) before it lands in the ticket.
The lift the team felt first is the one that's hardest to put a percentage on. Agents start every ticket with context they used to look up themselves across Shopify, the customer record, and the Notion page.
The short weekday support window goes further than it did before, because we do the assembly work in advance and the agent does the judgment work on top.
Where does the brand go from here?
The next move is already clear. The narrow first step into direct replies is WISMO (where is my order), because it's the lowest-judgment ticket category and Shopify already gives us everything a customer needs to answer it.
WISMO is also the category most relevant outside the support window. Orders ship and customers ask about them at every hour, and a direct AI reply on a WISMO question outside agent hours is a win for both sides.
After WISMO, the obvious beats are Tools and Tasks on top of the existing data layer:
- Refunds for orders inside the 30-day return window through Shopify Refund.
- Address changes on unshipped orders through Shopify Edit Shipping Address.
- The gift-order ownership-transfer flow the team handles by hand today.
Each of those is a ticket category where we already have the data. Turning us from drafter into doer is the upgrade.
Further out, an apparel comfort brand's most valuable AI assistance is around the repeat-buyer cycle. A customer who returned their first compression pair for being too tight, then comes back asking what to try next, is the kind of conversation we should be actively guiding: walking the customer through what worked for similar shoppers.
That sits on the same training stack the team already has in place (we're not asking them to build anything new for it).
What 100% automation looks like for a brand like this isn't 100% direct replies on every ticket. Sizing, comfort, returns, fit: these are the conversations that drive a second purchase and the word-of-mouth the brand is built on.
The goal we share with the team is freeing that support window for those conversations specifically. Direct replies on WISMO, Tools for refunds and address changes, AI-guided repeat-buyer flows: each of those gives agents more of the window back for the work that compounds.
For another team that leaned hard on the User Data API specifically, Edel Optics jumped from 25% to 79% AI resolution on Zendesk after the same connection. Browse the full case studies tag page for more, or read about how the Gorgias integration works.
