Emre Girgin

Wrong-vehicle and poor-quality incidents hurt a premium ride experience. Vehicle reviews asks chauffeurs to photograph their car before selected rides and checks it against our quality standards. It began as an AI hackathon project and became a workstream, and I led the design.

V1: Shipped in two weeks

Before randomly selected rides, chauffeurs photograph the exterior and back seat. Big, immersive example photos explain the task without words, and short tips cover our standards. We launched in a few small districts, with every review done by hand while we trained the model. Submission and success rates were both around 25%.

Research

I interviewed chauffeurs. I found out that they understood vehicle reviews and already kept high standards, but the feature felt like a one-way obligation: they sent photos and never heard back. That insight shaped what came next, real-time feedback and an insights screen. view details

Vehicle reviews: Research

Blacklane · May 2026 · Chauffeur research

After a few months live, 94% of chauffeurs were submitting vehicle reviews, but only about 25% passed. Retakes did no better. I interviewed nine chauffeurs across London, Madrid and Singapore to find out why.

The headline

Chauffeurs understand what’s checked and keep high standards with or without this feature. The example photos help them know what to do and act on it. What’s missing is any sign that their effort matters.

It feels one-way

Chauffeurs submit photos and hear nothing back. One called it “an add-on feature that’s not doing anything.” Word has spread that submissions might not be reviewed at all, and some chauffeurs now submit low-effort photos as a result.

Timing is personal

Some moments make submitting hard, such as motorway transit, airport pickups or underground car parks with no signal. The “later” button solves this, but no chauffeur found it alone. Each learned about it from a colleague. There’s also no single natural moment: some prefer the start of a shift, others five minutes before pickup.

Effort already exists

All nine described careful cleaning routines that predate the feature, and the example photo was the most valued part of the flow. It shows the angle, framing and what to include, which makes it the best way to explain what good looks like. Beyond cleanliness and water, though, knowledge of the criteria was patchy.

What I recommended

1. Close the loop. Show a submission history so chauffeurs know their photos are seen.
2. Make “later” easy to find. Visible from the first encounter.
3. Explain what’s checked. Use photos and video rather than text.
4. Reframe as recognition. Treat submissions as a mark of professionalism, not an audit.

Longer-term ideas, such as incentives and in-person spot checks, need business and operations input.

V2: Real-time feedback

AI analysis now shows what looks good and what doesn’t, and asks for a retake when a photo misses the standard. Chauffeurs could see their photos were actually being reviewed, and success rose to 45%.

V3: History & insights

A history and insights screen gives chauffeurs an overview of their vehicle quality. The central idea was to focus on one single, digestible number (and not a dashboard), visible on their profile. Success rate rose from 45% to 70%.

Next: Mark on photo

Over 80% of failed submissions come down to a missing water bottle or phone charger, which are hard to spot in photos. Chauffeurs can now tap to mark them when the AI misses them, so a good submission doesn’t need a retake.

Chauffeur onboarding ran on a desktop-only web system split across several tools. It took about 11 days on average, around 5 of them spent on manual steps like scheduling the webinar. I designed an in-app version where partners invite chauffeurs, chauffeurs complete everything from their phone.

Sending an invitation

Partners see their chauffeurs and vehicles in one place and invite a new chauffeur by entering a few details. The chauffeur gets an email, sets a password, and lands in the app to start.

Starting onboarding

Tasks are grouped into three short sections: upload documents, complete your profile, and watch training videos. The language is actionable, and progress updates as each task is done. For documents, chauffeurs can take a photo or upload a file, and when a document has two sides, the flow guides them through the front and back.

Profile photo requirements are strict, and chauffeurs sometimes struggle to meet them. After taking or uploading a photo, three yes/no questions check for the most common reasons submissions get rejected. Approval rates doubled after release.

Starting with Blacklane

When everything is confirmed, they get a notification and view the welcome animation, a first moment as a Blacklane partner that I wanted to feel earned.

The problem

For a long time, chauffeurs saw only their overall rating. Guests’ written reviews stayed internal, used for quality scoring, to protect guest privacy. Chauffeurs were curious what guests said about their work.

The design

I proposed a solution on my own initiative: AI summarizes the written reviews, so chauffeurs get the substance without exposing any individual guest. I defined the summary categories, which are consistent for every chauffeur everywhere, so feedback is comparable and fair. Company managers can filter by chauffeur to see their ratings and reviews.

Today

Currently in release process.

The problem

Customers trust reviews but can’t read thousands of them. Scientists had built a model that grouped reviews by topic and sentiment, but earlier concepts were too complex for design leadership. They were built around the model, not the customer. I started over from what customers need.

The design

Past research showed customers look for topics that matter to them, read mentions to understand why, and think in pros and cons. I explored three directions and chose topics with progressive disclosure: pills open a sheet of mentions. It resonated in leadership reviews, and my user tests confirmed it. Customers trusted topics drawn from reviews and relied on mentions more than counts, so I kept numbers minimal, even when the science team wanted more. We then ran experiments on four pill and sheet combinations, and customers decided faster and kept coming back.

Today

This was the beginning of Amazon’s review summarization work. The experience is still live on Amazon’s app and website. It has evolved, but the UX is close to what I designed. Airbnb and Tripadvisor later shipped similar review summaries using the same pattern.