How UKLASH Automated 1 in 5 Subscription Cancellations End-to-End With Actions

The Challenge: An AI Agent that couldn’t yet make decisions

When UKLASH first introduced AI Agent two years ago, the wins came fast on volume. Their lean CX team automated repetitive tickets like order status and shipping policies through peak seasons like BFCM, clearing backlogs that could stretch to five days.

The tickets that mattered most were the edge cases. A refund where the policy depends on what was bought and when. A subscription question mid-cycle. Those need a decision, and back then an AI agent didn't make decisions; it followed the instructions you gave it. So the team wrote instructions, then more of them as new cases surfaced, and revised them again whenever policy moved.

“At first, it was almost a burden to coach AI Agent and ensure our knowledge was robust, accurate, and up-to-date. It was a big lift that our team couldn’t manage.”

— Tamara Preece, Customer Experience Manager at UKLASH 

Instructions can only cover the cases you anticipate, so the tickets that fell outside them came back to the team. For UKLASH to lean in further, AI Agent needed to be more than a fast responder. It needed to be able to act.

The Turn: A stronger AI Agent to come back to

AI Agent 3.0 changed what was on the table. The rebuild gave UKLASH a more capable agent than the one they'd tested. Actions gave it something it had never had: the ability to complete a request, not just answer one.

UKLASH joined as alpha and beta testers, pushing Gorgias on the two things they cared most about: brand voice control and subscription workflows. If Grace, UKLASH’s AI Agent, was going to handle a cancellation, it had to sound like the brand and follow the correct policies. 

The Solution: Actions turned answering tickets into end-to-end resolutions 

Today, UKLASH runs 8 Action configurations across subscription management and order operations. 

The Actions carrying the most weight:

• Cancel subscription (via Loop): verifies the customer's email, confirms which subscription, and cancels.
• Pause and skip subscription: offers a pause or a skipped shipment instead of a cancellation, turning a would-be loss into a retained customer.
• Update shipping address: the Recheck app holds an order in Shopify before it releases to the warehouse, giving AI Agent a safe window to change an address that used to always need a human.
• Order edits, reships, and refunds: resolved directly within the guardrails UKLASH sets, instead of escalating every post-purchase change.
• Fulfillment holds: pause an order so nothing ships while an issue is reviewed.

Turning cancellation requests into retention opportunities

The first Action UKLASH built was subscription cancellation through Loop, set up around BFCM 2025. They started conservatively. Canceling a subscription is irreversible, so the team relied on a robust condition list that Grace had to meet before proceeding. The result was safe and slow: most requests still handed off to a human.

Using the event logs, the team audited where the workflow was stopping and found most conditions were acting as an extra step rather than a guardrail. 

UKLASH cut it down to two conditions:

1. The customer's email
2. Confirmation of which subscription to cancel

That was enough. The workflow became reliable enough to run without human intervention, with the customer confirming every cancellation before it executes.

Today, roughly one in five cancellation requests is completed by Grace end-to-end. And because the agent can now offer a pause or a skip instead of a straight cancel, about 13% of those requests end in a save rather than a loss.

“AI Agent isn’t limited to cancellation anymore. It now offers customers the option to skip or pause, giving us another chance to retain the relationship.”
‍
— Tamara Preece, Customer Experience Manager at UKLASH

That's the shift Actions makes concrete. A cancellation request used to be a ticket to process. Now it's a conversation with something at stake, and Grace is equipped to handle it.

Address changes, before the order leaves the warehouse 

Address updates always needed a human, for a reason that had nothing to do with AI: orders reached UKLASH's warehouse system in about seven minutes, with no sync back to Shopify. If they missed that window, the change was impossible.

Adding Recheck created a holding window. Orders now pause in Shopify before release, giving Grace time to update an address before fulfillment starts, eliminating the need for human intervention. 

“Introducing Actions reduced handovers where there didn’t need to be one, dropping by 6%.”

— Tamara Preece, Customer Experience Manager at UKLASH

Automating order edits, reships, refunds, and fulfillment holds

Beyond subscriptions and address updates, AI Agent now handles approved order edits, item changes, reships, refunds, and fulfillment holds within the guardrails UKLASH sets. Instead of escalating every post-purchase change, Grace resolves it directly or places a hold so nothing ships while the team reviews. 

For example, when Grace detects a ticket with the order/missing intent, she follows UKLASH’s policy. She determines which order the customer is referring to, checks the shipping status, addresses the customer’s concern about the missing item, and sends the Missing item form so the customer can confirm the issue. 

The tuning layer: How UKLASH keeps expanding what Grace can do

For UKLASH, AI Agent 3.0 gave the team a reason to reinvest in automation: a stronger, more reliable version of the agent they had tested before.

Giving an AI agent the ability to act raises the bar on knowing what it's doing. Gaia and MCP are how UKLASH holds that bar.

Gaia acts as a QA coach. In one case, it traced an incorrect return answer back to a specific gap in the knowledge base and drafted the fix. A process that used to require reading transcripts and scanning ticket logs is now a single conversation. 

MCP turns support data into operational insight, pulling tickets by country and contact reason for an Ops courier review in seconds instead of hours. 

Together they gave the team the confidence to keep expanding what AI Agent is allowed to action.

"Before MCP and Gaia, I was hesitant to even touch the guidance because there were so many variables. Now I can be specific about what I need, review what it creates, approve it, and monitor what happens."

— Jakub Lukasiewicz, Senior Customer Care Lead at UKLASH

That confidence is what lets UKLASH keep widening the set of things Grace is allowed to do. Each Action they add is a decision they can see the results of.