Portfolio & Experience

Like any subscription business, this client has a meaningful share of monthly payments that fail — expired cards, insufficient funds, etc. I built a system that captures these and handles them automatically.

Every morning it pulls the last 24 hours of failed payments, works out why each one failed, and takes the right action, such as retrying against other saved cards, rescheduling to the customer's next likely payday, or opening a Zendesk support ticket for anything that genuinely needs a person. The team can see everything through a dashboard showing what's been handled and how recovery rates are trending month-on-month.

It processes 9,000 failed invoices a month and successfully recovers about 3,000 of those.

The client offers a range of services to customers if they report a pet missing, such as outreach to nearby vets, a poster and posts on their social media channels.

I automated this whole process, including AI-powered content moderation, poster generation, social media posting and SMS and email outreach to nearby vets and rescue centres. I also added features such as automatic follow-up after a week, including reading the owner's reply to work out whether the pet is still missing or safely home, and a web interface for the team to view the process. This process used to require about 30 to 45 minutes of work per report, and only available during office hours, but now runs 24/7 and scales without needing extra customer service staff.

The system processes roughly 500 reports a month, and it frees up roughly 30 hours a week of operator time.

This project tackled two problems in one service. The first was Transfer of Keepership (the legal process when a UK microchipped pet changes hands) which requires getting formal consent from the previous keeper. This includes chasing them if they don't respond, escalating refusals, and timing things out after 28 days. The team were running all of that manually. The second was the volume of routine customer query emails landing in Zendesk.

The solution handled the TOK workflow end-to-end by emailing the previous keeper for consent, falling back to SMS if there's no email on file, routing letter-required cases to a real person, and resolving or escalating in Zendesk. For the email triage side, every new ticket gets classified by AI into one of about 12 categories, and routine ones get the right reply and help article sent automatically. The support team's queue ends up with only the tickets that actually need a human.

Between the two, it handles about 5,000 emails and 2,000 TOK requests a month, with a full audit trail for the regulated stuff.