MUST-HAVES BEFORE GETTING IN ON THE AI GAME — Notes for Founders and Investors
For those that don’t know, my work focuses on the business/financial and operational aspects of startups. I help founders build sustainable businesses based on their innovations. I help investors assess the potential business/financial applications of innovation.
My latest project has taken me deep down the rabbit hole of the world of AI, and I want to share with you a summary of all my findings.
Specifically, I want to share with you four key considerations for making AI sustainable as an actual business case — both for founders who are developing products, and for investors who are funding products.
Thoughts/feedback are welcome, and shares/likes are greatly appreciated.
I’m going to be blunt here — I fear the hype of AI. I have been exposed to too many situations of “AI for X”, where founders are using this latest buzzword haphazardly in their decks, and VCs are throwing money at bad ideas based on the latest trending buzzword.
What follows are 4 MUSTS for any startup dealing with AI to be tenable. These are items that must be confirmed beyond the tech, the team, and the extensive math.
For founders, if these things are not honed and solved, go back to the drawing board.
For investors, if these things are not honed and solved, reconsider.
(I use the term AI as a catch-all for Machine Learning, Deep Learning, NN, Cognitive Computing, etc…)
1. THE USE CASE MUST BE EXPLICITLY DEFINED
Here’s a story. Two young guys approached me for help with their startup, building AI for the HR/Staffing industry.
Me: So what aspect are you looking to improve?
Them: What do you mean?
Me: Screening? Matching? Retention?
Them: Um, all of it?
The application of AI is all about prediction — using information you have to generate information you don’t have — like predicting if a medical image is pre-cancerous, or if a car should slow down if there is a pedestrian. All of these predictions are on specific tasks, sometimes many at a time, that serve a unique, defined purpose.
In other words, AI is at is best when the use case is determinate, and based on a defined process. Broad use cases need to be broken down in to manageable steps in order to be successful.
If you are not able to have an effective “monopoly” on a specific use case, keep narrowing down your niche. Once one process is “monopolized” you can move on to the next.
Founders can have grand visions, which you may agree with, but if they are not able to describe the individual steps to get to that vision, and the solution for each step, then reconsider.
2. SPECIFIED, PROPRIETARY DATA
One of the key reasons why the latest AI push is coming from large companies like Google, Facebook, Amazon, etc. is because they own troves of data. This data is ready at their fingertips for training, testing, specification, and QA.
For any startup pursuing AI, once the Use Case is clearly defined, it is imperative to “own” the data for the niche or the vertical. There are plenty of open-source data sets available for testing, but this isn’t good enough. When everybody has access to the same information, there is no unique value proposition.
One option is to find the partners who have access to relevant data very early and lock them in. Zebra Medical, for example, built partnerships before building the product.
There are ways to get creative as well, so think outside the box. X.ai is a great example. They are using AI to create virtual personal assistants. Such interactions are more complicated than people think, and their product needs to understand preferences and be able to communicate seamlessly.
According to stories, after the product was developed, they needed real life data to train and test; effectively info on how existing personal assistants work. Begging a bunch of existing personal assistants to share their data is nearly impossible, so what did they do? The X.ai team worked as personal assistants themselves, building their own proprietary data for the machine to observe and learn from.
The first question you should be asking founders is: How are you going to own proprietary data for your vertical? Founders, especially in the earliest stages, may not have a concrete answer, but at the very least they need to recognize the importance of proprietary data. If they don’t, reconsider.
3. THE END USER/LAST-MILE SOLUTION MUST BE BUILT
Data is only as valuable as how it can be used to deduce actionable information that creates utility for an end-user. Once the use-case is defined and the data is secured, it must be packaged in a way to give insights that can be applied to tangible recommendations for action, otherwise it is effectively useless.
Without addressing how the end-user can effectively gain from the AI product, the startup has little chance to succeed. In other words, creating “middleware” is simply not good enough for the world of AI.
Understanding how an end-user can effectively utilize your output is the essence of “Product Market Fit” for AI products. This requires a deep understanding of what the end-user needs for each niche and vertical. This also requires building out the product all the way through to the end, including UI/UX considerations.
As stated in the first step, AI is great when used for a defined process. Your job is to make sure founders have solved for that process all the way to the end, and have not stopped in the middle. Founders need to demonstrate to you how a potential customer will benefit from the product, not just why. The why is step 1.
4. SECURED BUY-IN FROM END USERS
It is one thing to build an AI product for a high-tech multinational, it another thing to build an AI product for your accountant. One side probably already has the techniques/tools available to integrate AI products into the relevant workflow. The other side may still be sending faxes. One side has a pre-existing culture of “AI Trust” that is open and receptive to testing AI products. The other side doesn’t understand what all these kids are doing on Snapchat all day long.
For an AI product to succeed, it must have secured buy-in from the end users. This includes understand where the product integrates to the customer, what tools are available and necessary, and working with potential customers to bring them to a position where they understand the potential benefit and are willing to test products.
If your potential customer is unable or unwilling to use your product then what’s the point? AI is still in the realm of “scary-new” for the vast majority of the world. Therefore, directly following step 3, you must secure buy-in from potential customers as early as possible, which includes working with them on integration into their existing workflow.
This step is critical for any startup that is not targeting high-tech multinationals as a potential customer. For every deck you get where the founders pitch “AI for X”, you should confirm that founders understand what the end-customer of “X” requires for product adoption. Founders should have a solid understanding of what existing tools/workflows/limitations exist for “X”, and how to overcome any barriers in the process. Otherwise reconsider.
AI is an exciting new realm of endless applications, but instead of pursuing visions, lets stop for a second and consider pursuing actual sustainable businesses. In order for that to happen, and for AI to be successfully integrated into the world, we must look beyond the strength of teams and the developments of the technology.
Like all startups, AI-based startups must be accountable to business/financial considerations, both for their own validation and for the potential success of the product.