Why Motorway's CEO wants less AI output

“I think we've got to the point where no one's impressed that you've been able to use AI to create a presentation.”

Tom Leathes of Motorway told me about a workplace phenomenon that I think many will relate to. We’re moving past the stage where some people show off their AI skills to create impressive documents at speed.

And when you graduate beyond that, that’s when AI can really make an impact on your business and affect your bottom line.

Inside Motorway's AI-first culture

Motorway is very much a UK tech success story. It’s a digital platform that lets people sell used cars online. It was founded in 2017 and achieved unicorn status in 2021. Last year it reported revenues of £78m and it has more than 400 employees.

Motorway doesn’t have a separate AI team. Leathes expects everyone to be using the technology. Their main tool is Anthropic’s Claude.

Leathes recounted a phenomenon to me when we spoke at his offices in central London, which happened around a year ago.

Many people joined the business from Amazon. They brought with them the idea of ‘silent read-ins’ at the start of meetings, favoured by Bezos, where people take time to read key documents before discussion starts.

Some early adopters twigged the power of Anthropic to quickly draw up powerful presentations to wow colleagues at meetings like this. (I think we’ve all seen this in action!)

There are two main problems with this...

The hidden cost of AI-generated confidence

Firstly, people don’t fully understand what’s been written in their name. “A well-written document isn't just good grammar,” points out Leathes. “It's the structure, the thinking, the ‘what if this doesn't work?’ All of those things that you have to think about to get a really good plan in place. The problem now is that anyone can talk to Claude for five minutes and it spits it out!”

I’m sure many of us have seen people generate impressive documents which they then lack the confidence to discuss or elaborate on, because they haven’t put the brainpower into creating them in the first place!

This approach to AI can also create content overload internally at companies, cautions Leathes.

“We went through a period,” he recalled, “when we actually had to sit down with teams and say ‘enough, this is too much.’ We can't pick the signal from the noise because there's so much work being created.”

This wasn’t just in documents for meetings, but in thought pieces shared in groups on Slack, for example.

“I think it actually drives you back to the old school where you have to sit down and talk to that person,” Leathes reckons.

So ‘AI misuse’, if we can call it that, was actively addressed by the leadership. But the company is motoring ahead thanks to its use of AI across the business.

How AI is fixing Motorway's own backlog

The software team use AWS, which has Kira as its central platform, accessing a variety of AI tools. This has led to so many new feature developments – like improving bidding logic in car auctions, for example – that it initially led to a pipeline problem.

“The backlog was thousands of releases long, going through human review,” according to Leathes. So they now use AI to triage the suggested improvements created by AI, identifying the low-risk ones that can go live quickly, while other ones are flagged for human review.

Motorway has also invested the time to pipe all of its relevant data into one “lake”. This reservoir includes product metrics, customer funnel data and, importantly, all the customer reviews it can access, collated from places like Trustpilot, but also culled from transcripts with customer service. All this data has been piped into Claude.

“So, I can now sit down and go, what's happening in the call centre today?” Leathes told me. This allows him to investigate the small failures and inefficiencies that are happening in the business every day, in a forensic way. AI can help him to spot themes that would have been like finding a needle in a haystack.

He gave me an example. “It's possible there may be a bug in our photo-taking app because there have been six mentions of the shutter not working on the Android app.” In the past, this kind of failing might have eventually filtered its way through to the executive team after a series of cascading meetings. But now it can be picked up much faster.

“It’s an amazing unblocker for finding problems faster and working out what to fix,” says Leathes.