Luke
Harper

Problem

Dealers were receiving buyer enquiries through four separate, disconnected inboxes — each built by a different team, in a different era, with a different design language. There was no unified view, no way to track what happened to a lead after it arrived, and no way for the business to know if its leads were actually resulting in sales. It was a fragmented, seller-hostile experience that nobody had properly named or owned.

Approach

I led the LMS from day zero: making the internal case, running discovery with dealers, scoping the MVP, and driving delivery across engineering, design, and commercial stakeholders. The brief was clear — put dealers first, consolidate the inboxes, and build the infrastructure to track lead outcomes.

The most strategically significant decision was introducing Lead Status — a mechanism for dealers to mark whether a lead resulted in a sale. Simple in concept, transformational in practice. It established the business's first ever Lead-to-Sale (L2S) benchmark, creating a new measurement layer and opening the door to outcome-based conversations with commercial partners.

With four legacy inboxes to replace and stakeholders across five teams, scope control was critical. I made fast, decisive prioritisation calls throughout — constantly pressure-testing what was truly MVP and what could wait — managing competing stakeholder opinions to keep the project on track and on time.

Problem

Speed of response is the single biggest predictor of a lead converting to a sale. Dealers are time-poor — managing a forecourt, walk-ins, and a lead inbox simultaneously. Many leads went unanswered, or received generic replies that failed to address the buyer's actual query. Both outcomes were damaging the Lead-to-Sale scores the LMS had made visible for the first time.

Approach

Using the Gemini API, we built an AI Response tool directly inside the LMS. When a dealer opens a lead, the feature generates a ready-to-send, personalised written response — informed by the full conversation history, the dealer's own inventory, and the buyer's behavioural signals from the platform.

A deliberate design decision: the response is editable before sending, not a one-click auto-send. Dealers are professionals and trust in AI output had to be earned incrementally. The goal was to augment dealer judgement, not replace it. This was an intentionally constrained first release — the right scope for a new AI surface in a high-stakes professional tool.

Problem

The platform had a retention problem disguised as an engagement problem. Users would arrive, find a course, and leave — never to return. This was damaging the quality of referrals sent to university partners (the core revenue source), because universities paying for leads want engaged, informed applicants — not one-and-done visitors. Without registration, there was also no way to track user behaviour meaningfully or pass valuable insights back to partners.

Approach

I led everything hands-on — from making the internal case for a mobile-first product to owning the vision, strategy, and delivery. This was the company's first ever mobile app, which brought significant organisational challenges alongside the usual product ones.

Scope was the biggest risk. Without clear boundaries the project was heading towards a delayed launch — I stepped in, made the hard prioritisation calls, and cut what wasn't truly MVP. That decisiveness is what kept delivery on track and got the app shipped on time across both iOS and Android.

A key strategic decision was making registration central to the experience — giving us the behavioural data to track user journeys and surface meaningful insights for university partners for the first time.

Problem

University Clearing is the UK's results day — the academic equivalent of Black Friday. For students who didn't get into their first choice university, it's a last chance to secure a place before vacancies fill up within hours. Already dealing with the disappointment of missed grades, these students then had to phone each university individually, managing multiple stressful conversations simultaneously under extreme time pressure — all while terrified of missing out entirely. On the other side, universities were spending heavily staffing call centres to handle the surge — a costly, inefficient operation that still left students frustrated and underserved. Everyone was losing, and no digital alternative existed.

Approach

I identified the opportunity, made the case internally, and led the project end-to-end — working closely with a team across design, engineering, and commercial to take it from concept to launch.

The commercial model was novel: institutions would pay per application received, creating a fresh revenue stream that didn't cannibalise the existing referral business. Getting internal buy-in for an entirely new revenue category required as much stakeholder management as the product work itself.

The hardest constraint was the deadline. Clearing happens on a single day each year — miss it and you wait another twelve months. Working with the team I kept scope tight and focus sharp, shipping a BETA first to run A/B tests and iron out issues before the high-stakes final launch. We hit the date.

Problem

A counter-intuitive problem: users said in research they wanted personalised results, and the product had a sophisticated filter system to deliver them — but filter usage was falling. The interface wasn't the issue. The behaviour was. Filters require users to articulate their own preferences in the product's language — a significant cognitive ask for someone uncertain about what they want.

Approach

I led the project end-to-end, working closely with the user research team and engineering to take it from concept to launch across both the website and mobile app.

The insight was simple: filters require users to already know what they want and express it in the product's language. A chatbot inverts that — it asks the right questions in plain language and does the translation automatically. Getting the logic right was the easy part.

The harder challenge was tone. After launch we collected feedback directly in the product and ran dedicated user feedback sessions with the research team. Users were clear — the chatbot felt robotic, cold, and off-putting. We iterated fast: the language changed, the tone changed, and even the product name changed based on what users told us. The principle I took from it — that copy and personality are as product-critical as functionality — is one I've applied to every conversational and AI-facing product I've worked on since.