When Alex Dalyac first pitched Tractable, he called bringing deep learning to accident recovery their ‘billion dollar idea’. Six years on, he’s been proven right.

He and his co-founder Razvan had been looking for real-world image classification tasks to apply ‘deep learning’, a form of AI that was developing rapidly, fired by dramatic computer processing improvement. Having explored pipe welds, dermatology, and geology, accident recovery was finally the industry with enough data, and a large enough total addressable market, to set their sights on conquering.

From then on, their job was to build the technology, and to penetrate their market.

In this blog, Alex shares how the team did it – from persuading early adopters to take on their technology, to now, where they’re able to use their existing market penetration to explore new areas.

The solution

When I joined Entrepreneur First, I already knew what technology I was going to build a company around.

I first encountered Deep Learning on a Computer Science conversion Masters at Imperial College London. The technology was vastly accelerating thanks to improvements in computer processing, and only a few labs in the world were working on it. And I was lovestruck.

We were, for the first time, seeing algorithms able to work out what was in a picture better than humans. I was soon spending every spare moment on image classification with deep learning.

But back in 2012, no-one was talking about it, even within Imperial’s computer vision lab. While college labs will often sit on a lot of breakthrough technology, they don’t necessarily think about commercialisation and real-world applications. Yet I could clearly see the impact it could make – it was like being in on some great secret.

Before I’d even left Imperial, I began working on applying AI to water and gas utility inspection, which was my first step into seeing how this technology could be commercialised. This culminated in a PhD offer, but my goal still went beyond academia. I hadn’t lost sight of my passion for company building. But now I had the technical knowledge I needed to do it.

That’s when I came to EF, and where I met my co-founder Razvan. 6 years on, we’re applying that technology to make a real impact; first through accident recovery, and now car recycling, and rebuilding after natural disasters.

This is how we found our market, gained traction, and scaled to new areas.

Finding our market

Raz was smart. Really smart. Where I had some technical ability, and a passion for the impact deep learning could build in the real world, he had the skills to take it to the next level.

When we were working on the market potential for deep learning and an industry problem to solve, we had to meet two criteria:

1) There had to be a large enough market, and;

2) We had to work out who would care, if it was cheap.

Additionally for deep learning, we had the third problem of ensuring that there was sufficient training data to learn from.

While the pipe welding I’d begun working on at Imperial was a suitable application for our solution, 300k annual pipe weld inspections in the UK was too small a market.

We then explored utilities, geology, dermatology – and then we hit on accident recovery.

We found a pain point. Accidents disrupt lives. Once the initial shock has passed your next priority is getting back to normal. In a car accident, this will normally involve claiming on your insurance. But this process is long. An appraiser will have to come out and assess your vehicle, and generate a quote, before you can even take your car to the repair shop. In reality your car will only spend a matter of hours being fixed, but the process of getting it fixed can take weeks or even months.

“This was a trillion dollar market, and one that could make a genuine difference to people’s experiences.”

We were sure that deep learning could accelerate this process:

1) We put the users in charge of taking pictures of the accident damage with their phone. This meant a significant reduction in time and cost, as this could replace the need for an appraiser to come to the scene and assess the vehicle.

2) Our trained algorithms that had processed millions of images would be able to generate an instant cost to repair the damage, accurately.

With billions of images of damage caused by accidents every year, it was a deep learning gold mine. We could also convincingly show that we were faster and cheaper than the current solution, as an example, we demonstrated our ability to assess accidents in minutes, instead of the typical weeks and months.

From repairers to insurance companies, this was a trillion dollar market, and one that could make a genuine difference to people’s experiences. We had found what we were looking for.

Penetrating our market

Despite knowing we’d hit on a billion dollar idea, making it a reality and getting the first customers through the door was much more of a challenge. The industry enforced a brutally slow sales cycle on us: for example, it took us 12 months to get a one-month pilot with an insurer, and the revenue generated barely covered the travel costs incurred, let alone a profit. We even tried to get acquired for $25m by a tech giant (obviously, that didn’t work out).

It was the first million dollar contract that changed everything. By demonstrating to an insurer that Tractable could generate $50 in efficiency gains for each of their million claims, this contract was easily worth it. On top of this, there was a clear value-add in customer satisfaction.

How did we get this deal? Well, we ‘went camping’ – i.e. we identified potential targets that we thought would be suitable clients, and then spent as much time around them as possible.

“Once one large company announced it was adopting our technology, that gave us the leverage to sell into others, and from there the dominoes lined up quickly.”

We actually used to spend months in the same cities as our customers so we could simply call them and say ‘Hey, we’re in the area – why don’t we get together’ – when, actually, being in their city was the only reason we were there! We also spent a lot of time finding and developing internal champions – people whose careers will benefit from adopting your technology. When they realise that, they will ultimately sell your company for you and help get a deal over the line, and that’s what happened here.

After that first big deal, we got into a stride. The contract made us credible to new customers and their execs. When they signed on, we always tried to announce it in the press. Where previously we might have been a risky bet, it was becoming riskier for big players to see their competition deploy our technology.

While getting this first customer to sign on was the hardest part, after that our customer acquisition process massively accelerated, especially as it generated FOMO in countries like Japan – one of the largest markets in the world, dominated by very large corporations. Once one large company announced it was adopting our technology, that gave us the leverage to sell into others, and from there the dominoes lined up quickly.

As well as Japan, we’ve also focused on the largest markets where we can scale quickly – the US, SE Asia, France. This isn’t just for practical reasons – the insurance companies are larger and the markets more competitive – but also important from a volume perspective too: more volumes means both more income and more data for us to train our AI with, making the solution even more accurate.

The last few years have seen us go from a small, scrappy company, to a fully automated AI service. Our customers trust us now; they know the value that we can bring to them, and to their customers’ experience. With this as our foundation, we’re well-positioned to focus on two new areas as part of our total addressable market expansion plans.

“We have huge capacity for impact - for our customers, and for the world. ”

The potential of deep learning and image classification is limitless. With our same technology, we are extending into new areas, including car recycling and natural disaster recovery.

These areas have huge capacity for impact – for our customers, and for the world.

We’ve already recycled as many cars as Tesla put on the road in 2019. When the typhoons hit in Japan later this year, as they’re sadly expected to, we can be a part of making sure households don’t go for months without roofs over their heads.

We wouldn’t be able to do this at all without our existing technology and market penetration – let alone at the speed we’ll do it at. The possibilities for Tractable, for deep learning and image classification, and the impact we can make, go on – and we’re just getting started.

The impact of typhoons in Japan — Car engines for recycling

Alex’s advice for other founders

As of June this year, Tractable is now worth $1B. It’s been an incredible journey to get there, and we’re still learning more every day. Here are just a few pieces of advice I’ve picked up along the way:

1. Find the technical magic that only you can wave

Finding a great addressable market was crucial to our success – but also important was that we had a solution that was almost unique to us at the time. We were into deep learning and image classification around 6 months before it kicked off, which gave us a great advantage – especially when competitors got onto it, but we were already ahead.

Don’t look in TechCrunch or what VCs are investing in for ideas – if they’re in there, it’s already too late. By deploying deep learning at the point of its breakthrough, we were able to leverage something that no-one else had.

To find this technology for yourself, read the journals. Speak to professors. If they get shy and embarrassed and say ‘you’ll probably find this really technical and boring’, there’s a good chance you’ve hit the jackpot.

2. Use FOMO to your advantage

As noted, it was when we got the first few big contracts that we really started to gather momentum as a business.

One thing we absolutely prioritised was trying where we could to announce our partnerships. This gave us that badge of credibility: we weren’t a small AI company, but were established experts trusted by other industry big players. And if you’re not adopting it soon, you’re going to miss out. This was critical in establishing us in our market, and increasing our penetration.

3. Utilise advisors – but make sure that they align with you
Advisors are invaluable in starting a company – especially if you’re less experienced or don’t have the rolodex available to make headway in a particular industry. In addition, they’re also going to ask for way less equity than it would cost you to get another VC on your cap table.

However, you have to make sure that they suit you, and what you need. For me, and for Tractable, that meant getting someone with a high appetite for risk, and a strong bias to action who would be making those introductions for us. If they’re not, then they’re wasting your time.

As well as advisors, ensure that your first hires are world-class. Shortly after our seed round our third co-founder, Adrien Cohen, came on board. As an experienced business executive and ‘trusted pair of hands’, he changed everything in making introductions, and teaching us how to sell ourselves and our pitch. If, like me, you have less experience in industry, and particularly if that industry is more conservative when it comes to change, find your Adrien.

4. The tech giants can only focus on a limited number of areas – be the best at a vertical that sits outside of this

With hindsight, one risk that we took with Tractable was by focussing on a particular vertical, rather than applying deep learning horizontally. This didn’t always attract VCs, because even with a vertical as large as insurance, there still wasn’t as big of a market as some wanted.

We chose to do this because we wanted to build an end-to-end solution for our customers. We wanted to connect everyone at every stage in the process – from the claimant, to the repairer, to external bodies, to the insurer. If we’d built a horizontal product, we wouldn’t have the resources, especially at the early stages, to build this full solution. What we’ve built is better for our customers.

The big tech companies have the engineers and resources to build what we’ve built. In fact, one of them tried to. What they don’t have, however, is AI for accident recovery at the top of their list of business priorities – so they can’t build those in-depth relationships.

Not only this, but, as we got there first with applying deep learning to that space, we had a head start. By the time they got to it, we were simply already doing it well, and could give customers more resources and focus than they could.