Buying an ‘AI’ company? MathLabs will tell you what it’s [not] worth

Legal IT Insider editor Caroline Hill meets the company that can tell you whether the tech at the centre of your latest M&A deal is scaleable AI comprising valuable IP, or standard technology with a pretty window dressing. We speak to Fujitsu about how MathLabs has become indispensable during the due diligence process. Could this be a white-labelled law firm differentiator?

We’re all familiar with unsubstantiated brags about all-singing-all-dancing AI technology (aka ‘my AI is better than yours.’) But what happens when you are tasked with investing substantial amounts of money to buy the company that develops that AI? How do you – particularly given the constraints of COVID19 – work out if the company is really worth what it says it is? After all, just the top 10 enterprise M&A deals in 2020 were worth a staggering $165.2bn, according to TechCrunch.

Enter 20-strong London-based computational mathematics and AI lab MathLabs, which is now relied on heavily by global companies including Fujitsu to, speaking frankly, cut through the BS and assess the value of a company’s tech stack, including any IP. Not only that, but it can help the purchasing company to work out how to best leverage the tech going forward.

Fujitsu has moved from a historical hardware and services business to a digital business – a strategy that its CEO has been vocal about. As part of that strategy, the Japanese-headquartered global IT equipment and services company is growing organically and inorganically, with corporate executive officer Nicholas Fraser brought in to spearhead that growth.

Speaking to Legal IT Insider, Fraser, who is also ex-McKinsey and joined Fujitsu in March 2020, said: “The tech companies we look at are typically in emerging areas. Quite often people position their company as digital and having a wonderful platform or IP, and it all sounds great, but what is the reality? What are the skillsets of the people? Do they own the IP? We need to be able to work with a specialist partner to do a technical assessment, especially in COVID times when traveling is difficult. If we do decide to acquire a company, then there is the question of how to get the most value from the acquisition. MathLabs provides support in both the assessment of a company’s technical capabilities and also in how to integrate and operate the acquired company post-acquisition. We’ve used them a number of times in the past and plan to use them in the future.”

MathLabs’ valuation has on at least one occasion led to Fujitsu changing the amount offered for a company, which, incidentally, it ended up not acquiring. Fraser says: “You’ve got to be skilled to be able to make these technical assessments and give a reliable opinion. We were looking at one company, for example, and MathLabs said ‘what they have isn’t as good as they make out, it’s very standard – we advise you to think about the valuation.’ This caused us to temper our valuation range to reflect this more balanced view of the target company’s technology solutions and skillsets.”

MathLabs is filling a skills gap that in many cases has led companies to buy assets that are worth less than anticipated. It speaks to a wider trend observed within everything from the court system to government regulation, where the development of AI far outpaces an understanding of how it works.

Fraser says: “I have worked at good companies and it would be nice to say that mistakes don’t happen but there are inevitably transactions where you find out that what is being considered as a great acquisition target is actually not as good as first thought.” He adds: “Part of the problem has to do with the definition of digital. Today it is used very broadly almost becoming ubiquitous in how a company describes themselves.”

So, who are MathLabs?

It’s probably relevant to mention at the outset that the senior team are McKinsey alumni. Director Erez Raanan was global head of analytics ventures at McKinsey Ventures; chief scientist Prodipto Binayak Ghosh was lead AI scientist at McKinsey.

Raanan says: “We saw at McKinsey that the amount of transactions with AI at the core is increasing, making it hard even for a company like McKinsey to understand the value of it. Around 10% of transactions now have AI at the core. It makes law firm services difficult, where even the acquirer has difficulty understanding the value, and that is what MathLabs is working on.”

The approach

Being completely honest, there’s little point for the majority of us in taking a deep dive into the methodology used by MathLabs (therein lies the problem). But suffice to say, they provide high level insights and hands-on testing to ascertain whether the asset has competitive advantage. In one recent project, MathLabs notably pitched for a deal alongside McKinsey.

The potential for law firms is huge and Raanan says: “If a law firm offers this as a white labelling service in partnership with MathLabs, it becomes a differentiator.”

Types of deal

MathLabs works on any deal where there is enterprise software. Raanan says: “Whenever you have enterprise software, half are going to involve AI. If I’m a big company, I probably already have a fund to invest in fintech companies, most of which are powered by AI.

Case Studies

MathLabs case studies could also be called cautionary tales. Ghosh says: “In one case we worked on, the target company had a quantum mathematics and AI powered platform to shorten the drug discovery process – they had created their own AI. The buyer did their due diligence and weren’t sure of the valuation or what the IP was, and the company claimed massive IP. We found a few things very quickly. All the AI they were doing was open source, and it was none of their own IP. Secondly, there were a lot of claims of AI but what they were doing was using fairly publicly known algorithms and they didn’t have the team to do AI. The real value of the company was in two or three staff with experience in computational computing, but there were no provisions to prevent those guys from leaving.”

The acquiring company ended up not going through with the deal.

In another instance, the company was marketing itself as an AI platform with a hefty price tag, but Ghosh says: “The deal wasn’t just about IP but a platform that was transferrable. We discovered the IP was a thin wrapper around off the shelf machine learning services. It had a high=quality interface, but the knowledge wasn’t there.”

The company ended up with a significantly lower valuation.

A truly valuable company will comprise a combination of IP and talent, and both Raanan and Ghosh stress the need to think about how scaleable the platform is. How easy is it to replicate? How defensible is it?

The honest and slightly concerning conclusion – and a conversation to be pursued further another day – is that the majority of people working in the corporate sector do not and will not have a clue. For now, it’s very much MathLabs time in the sun. Fraser concludes: “There are many consulting firms that provide technical opinions, but they are not as specialist as MathLabs with the same level of talent. We’ve used them a number of times in the past and plan to use them in the future.”