How Healf lifted its Trustpilot rating from 4.2 to 4.8 with a Claude-powered support agent

About Healf

Healf is a UK health and wellness e-commerce platform, ranked by the Financial Times as the #1 fastest-growing company in Europe. Thousands of orders, subscriptions and wellness products mean a constant stream of customer questions: where is my order, change my subscription, is this product right for me.

The challenge

Support volume was growing as fast as revenue. Response times stretched from minutes to hours at peak, resolution quality was inconsistent, and it showed where it hurts most for an e-commerce brand: the Trustpilot score sat at 4.2 and was drifting down, dragging conversion with it.

Hiring more agents would have scaled the cost line with the problem. Healf wanted the opposite: support quality that improves as volume grows.

The solution: an autonomous support agent built on Claude

We built a production support agent with Anthropic's Claude as the reasoning engine. Claude was chosen for the qualities that matter in front of customers: reliable instruction-following, a natural and on-brand tone, and strong judgment about when not to act.

For every inbound contact, the agent assembles full context before it acts: the customer's order history, subscription status, delivery tracking and past conversations. Claude then classifies the intent and either resolves the case end to end or routes it to a human with a drafted response and its reasoning attached.

• Autonomous resolution of refunds, re-ships, order changes and subscription updates, executed against Healf's commerce systems
• Context enrichment on every ticket: order, subscription and customer history injected before Claude reasons about the case
• Tiered autonomy: routine actions run fully autonomously; sensitive cases go to a human with Claude's draft and rationale
• Full audit trail of every decision, so any response can be traced, reviewed and improved
• Weekly tuning loop: misclassifications and escalations feed back into prompts and policies

The results

The agent now handles the majority of support contacts autonomously, around the clock. Typical response times fell from hours to minutes, resolution quality became consistent, and the support team's time shifted to the genuinely difficult cases.

The number that matters most moved with it: Healf's Trustpilot rating climbed from 4.2 to 4.8 over the engagement, and support capacity no longer scales with headcount.

Why Claude

Customer support is an unforgiving place for an LLM: every mistake is public and every response carries the brand's voice. Claude's instruction-following and judgment let us give the agent real autonomy, executing refunds and account changes, not just drafting replies, while keeping a clean line between what it may do alone and what needs a human. That balance is what made autonomous support safe enough to run in production at this scale.