Case study • Harry's

How Harry’s replaced rigid workflows with AI that actually resolves customer issues with follows through, at scale.

75%
of subscription tickets automated
360%
ROI on customer experience
99%
accuracy

Challenge

Harry’s set a clear goal for automation: reduce human handling of routine requests while making sure every resolution was fully executed and clearly auditable. The team wanted something they could configure by use case, that respected brand tone, and that followed through reliably without extensive flow building or ongoing maintenance.

The tools they’d tried relied on rigid workflows that broke on anything outside the script. AI replies could say an action had happened when it hadn’t. At the same time, Harry’s was moving from a homegrown subscription platform to Shopify Subscriptions, which brought a wave of login and account questions.

For a brand built on direct relationships with its customers, the stakes were clear. These responses would go to real customers, so the AI couldn’t hallucinate. After evaluating and piloting multiple products, Harry’s chose Kodif.

Solution

Phase 1: Early wins without custom training

Before any configuration, Kodif resolved routine account access and login issues using its ecommerce pre-training. That quick win gave the team confidence in a resolution approach that didn’t depend on mapped workflows.

Phase 2: Full-cycle resolution for subscriptions

Kodif’s AI agents were deployed to intercept and resolve subscription cases on their own:

Retaining cancellations, with automatic follow-ups when customers didn’t reply
Resolving login issues during the migration to Shopify Subscriptions
Freeing human agents to focus on complex edge cases and high-value tickets

Phase 3: AI supervision that verifies every action

A supervision layer checks that the agent ran the correct SOP and confirms the action was actually performed. If the AI says it updated an address, the system verifies it did. If it says it replied, the supervisor confirms the message went out. If it missed a follow-up, the gap is flagged and fixed automatically.

Phase 4: Brand voice and human handoff

Harry’s designed its agent to speak in the brand’s voice, de-escalate frustration, handle cancellations and refunds with care, and leave an audit trail for every action. When a case reaches the edge of what the agent should handle, Kodif passes it to a human specialist along with the transcript, intent, order details, and verified actions, so the customer never has to repeat themselves.

Results

Kodif now owns subscription tickets for Harry’s from start to finish, with verified follow-through on every action and a clean handoff to the human team when it’s needed.

75%

subscription tickets resolved autonomously

360%

ROI on customer experience

99%

accuracy

“We focused on the highest-volume use cases, where we could both improve speed and give time back to our CX team to put the care and attention to more unique and complex customer pain points.” (Faheem Uddin, Global Product Manager)

Takeaway

Most AI support tools are judged by whether they reply. Harry’s judged theirs by whether the action actually happened. That’s the right test for any brand where subscriptions, cancellations, and account changes drive ticket volume. A reply that says “done” when nothing changed in the backend creates a second ticket and costs customer trust.

For CX leaders evaluating AI, the useful question isn’t the deflection rate. It’s this: can you prove every action the AI claims to have taken? Verified actions, audit trails, and full-context handoffs are what let a brand with a high bar hand real tickets to AI.