Behind the Scenes: How to Get VC Funding for Machine Learning R&D – Part 1
VC-funded startups and AI are a match made in heaven. The former aims to disrupt its field with an innovative solution, the latter is the…
VC-funded startups and AI are a match made in heaven. The former aims to disrupt its field with an innovative solution, the latter is the innovation the world can’t get enough of – according to a Q4 2018 report from PwC and CB Insights, VC funding for AI startups topped a record $9.3 billion (a 72% spike from last year).
That sum doesn’t even account for AI-powered VCs like In-Reach Ventures who this year announced a new €53m fund, exceeding the original fund target of €50 million.
You get the point: AI is smoking hot right now. And if you’re reading this, there’s a fair chance you’re looking to Venture Capital to finance your AI solution; but where things get hot, people get burned.
However, since we’ve successfully navigated several startups through the VC funding minefield – securing capital for their ML R&D projects – we thought you might enjoy a peek behind the scenes at how it’s done.
So, welcome to our mini-series on how to get VC funding where we’ll take you on the journey, step-by-step, and reveal the elements of an ML R&D project plan that gain traction among early-stage investors.
We’ve written about the rush on AI solutions, and how not every business will benefit from ML or deep learning. Yet despite our words of caution, many companies can’t wait to sprinkle a little ML magic dust on their product or process – regardless of whether it’s needed; or, even feasible.
And it’s due to this lack of rigor that the quality of ideas that startups have when it comes to ML range from game-changing to game-over. Of course, the company matters as much as the idea, but what makes for the perfect combo?
Looking back on our experience – these four characteristics contributed to funding success.
The startups we worked with were already operating and generating revenue.
The functionalities in development were not only answering a specific business need; they had a sound business case backed by a plan for monetization.
Validation: customers are willing to pay for the features; other companies do have similar features; you’re doing this process manually and can confirm (with numbers) that using ML will save a concrete amount of time and/or money
Each of the projects was delivered within 6-9 months, meaning the entire investment concluded within a year. That’s not to say that long-term projects can’t get financing. But in our experience, the relatively short-term nature of these projects was a significant plus.
This section wouldn’t be complete if we didn’t shed light on bad ideas. And yes, we’ve seen our fair share. Typically, bad ideas are of the forced variety: where the startup doesn’t have a concrete plan or application for AI but has decided to get on the ML train, nonetheless – even though it’s a road to nowhere.
In most of these cases, data is the culprit. Either the startup doesn’t have the appropriate data or the data is of sub-par quality or quantity. Still, the team insists on funding their data science project anyway – whether the data is there, or not.
Where it’s not an issue with the data, it’s a question of unrealistic expectations – coupled with a lack of basic understanding of how ML works or what it can deliver – that results in a forced idea.
One example of this was with a startup using deep neural networks to verify whether changing an element on the front-end would impact its functionality; that’s not how deep neural networks work, so the concept simply isn’t feasible – still, the facts didn’t halt the effort.
If the concept looks promising, then it’s time to get to know the client, their business, and their idea. We’re talking an in-depth, bare-all kind of affair, where we get to feel what it’s like in the startup’s shoes.
Yes, we’ve considered a few of these items a step earlier – but here, we dive even deeper.
So, in order:
Identify exactly what your client wants and verify whether the concept is feasible. Understand their business objective and think of efficient ways of getting there. Typically, at this stage, you want to find out what your client’s budget looks like – as this will dictate what’s available to you as you plan the solution.
Once you’ve identified a way of satisfying your client’s objective, find out if any existing solutions could work. It may well be that there’s an available technology you can adapt, then implement. If so, the scope of your project drastically changes as it stops being R&D – and you should report the findings to your client immediately, then get on with integrating the solution.
If your research of existing solutions comes back fruitless, it’s time for … yes, more research.
That’s right; only this time, you need to focus on the client’s market and their industry. Study the competition, dive into industry trends – look far and wide as your goal is to better understand what’s missing across the spectrum to be able to suggest a solution that’s both unique and innovative….
As these are the characteristics, you’ll need to secure financing.
You’ll need to research publications that cover similar projects. And pay particular attention to the methods, results, and problems faced by projects, as well as the data used to power solutions.
Google Scholar is a valuable resource; especially the “cited by” search function, which surfaces articles that are harder to find using keywords.
Always remember to save your sources as you’ll need to show your reference material when you come to present your plan.
Yes, this can all take a lot of time and effort upfront, but without the early investment, you’ll not only strike out on financing, but you could also embarrass yourself by chasing a rabbit down a hole. So, give it all you’ve got at the outset and give yourself the best chance of success.
In the next segment, we’ll cover how to set KPIs, then get to writing your plan.
This Is Why Machine Learning Is So Hard. Off-the-shelf models are a sound basis for enhancing custom-built ML solutions, but a little more. It’s rare…