A guide for startups in the AI space trying to impress investors during the due diligence phase
Over the last few years, there has been a rapid growth of new startups using AI to solve different kinds of business challenges. This rise has led to an acceleration of micro-VCs (or angel syndicates) interested in early investments in these startups. One of the most common challenges these groups run into during the due diligence period is performing the “diligence” on the AI, the product, the team, etc.
Typically, investors (at any stage) perform several due diligence efforts: business, financial, legal, technical, people, etc. and come up with a solid thesis to justify their investments. This is also true for smaller angel funds, syndicates, and micro-VCs. For the few target companies in the deal pipeline that tout AI as a value-differentiator, technical due diligence becomes a key input to the funding decisions. It helps in the proper allocation of the funds (once the round is settled).
As for the nascent stage startups who are busy working on code and algorithms and implementing solutions for their first few fledgling customers, due diligence seems a bit alien and scary. The most common brave-enough response is usually,
“We are not ready for this but happy to help with the information you need. We have some processes and there are lots of rooms for improvement. That is why we need to raise money and use it for structuring the product engineering teams.”
Calculai is a boutique firm that provides advisory services to VC/PE firms, startup ventures, global multilateral organizations at the intersection of innovation, AI, and data strategy. One of the first things Calculai did as a company (when we started) was to look at different industries and domains and understand the impact of AI.
How are predictive intelligence, computer vision, NLP, and robotics used across the spectrum of progressive companies in different verticals like retail, education, agriculture, ad-delivery, and media? This provided us with a broad view of the AI marketplace and how it was evolving, leading to high-quality conversations with our clients on how AI is shaping the way we eat, play, shop, work, and more and what they need to get started and stay on the course.
Based on such conversations, we had the opportunity to do technical due diligence before millions of dollars (seed investments) were pumped into a few target AI startups in the deal flow.
We scoured the internet and found a few downloadable technical due diligence questionnaires. While all of them are good in some ways, none of them fit the needs we had for the startups we were looking at. We discussed with a few different VC fund analysts and dug deeper to understand what exactly did they care about as a result of the due diligence process.
We realized that most VCs are somewhat interested in a 360-degree assessment but what they really care about is the defensibility of their investment (scalability of existing processes and the presence of a tangible competitive advantage) to their LPs (limited partners).
Based on this core input and using our team’s prior experience of doing due diligence and building AI products, we created a simple but robust framework to guide our work. We wanted to see extemporaneous responses from the startup founders to our challenging questions. We gave the founders at these target companies minimal indication of what we would ask ahead of time in the interviews so that we could judge the authenticity of their responses and the adaptability of their platform and processes. Few of the startups we engaged with did get multi-million dollar seed rounds and happily thanked us for the process. The rest didn’t make the cut and thanked us, too — they realized the glaring holes that needed urgent patching up to be ready for investments.
Here are a few of the salient elements and associated questions that helped us with a robust understanding of any startup company involved in building products using AI.
AI companies are non-functional without data. It’s essential to understand the pipeline of high-quality data that helps with training, inferencing and storytelling of successful AI solutions for different stakeholders. Some of the questions we asked and scored were
- How many different data sources? Are there any logical groupings?
- How is the data stored? Is there data versioning?
- Is there synthetic data? How does the team qualify the synthetic data vs. real data?
- How to add a new data source? Or remove an existing one? How to suspend a data source in production?
- What volume is necessary to create a new algorithm or upgrade the existing one?
No AI company can survive without a decent infrastructure. Hardware and software are needed to run continual experiments and build better solutions while keeping the teams productive. Here are some questions to consider for understanding the maturity of the company in question in terms of infra they put to use
- Are the data pipelines infrastructure vendor agnostic? (can they easily switch from one cloud vendor to another to manage costs and performance?)
- Who has access to the infrastructure for development and production?
- What kind of infrastructure is used for experimentation?
- What kind of infrastructure is used in production for scaling?
- How are the micro-services managed? (if any)
The structure, behavior, and governance are needed to assure fairness, mitigate discrimination, and maintain the integrity of data, algorithms and solutions. This is a touchy subject and highlights how the company thinks about the ethics of the algorithms they put out in the market.
- Do they have regular ethics reviews of the data and inferences?
- Who participates in such governance and reviews?
- How many cases have happened where they had to give a negative rating to data or algorithms?
- Can they explain the inferences from algorithms with some degree of confidence?
- Can they debug the algorithm? Can they roll back a deployment to an older model?
AI companies exist to solve business problems with algorithms. Some of the questions we asked and scored were
- How many algorithms are in production?
- How do you version algorithms?
- Are there custom business rules incorporated in the post-processing of algorithms to suit them to customer needs?
- How is the deployment of algorithms managed?
- How will the algorithm function in cloud vs edge?
- How are the endpoints for accessing the algorithms exposed?
The most important asset of any startup is its team — the humans involved in conceiving, designing, building, deploying and refining AI-based solutions. It is critical to understand how they work as individuals and as teams, how they manage their processes and adapt to changes.
- Is the team diverse (age/gender/ethnicity/degree)?
- How frequently do they review their processes?
- How are the teams structured?
- How is the career growth planned within the teams?
- Do they bring in customer members into their teams when implementing a solution for customers?
6. Change Management
This one is slightly different and focuses on how the entire team and their technology are oriented towards the customer. In most cases, an AI-powered product or service works within the scope of a bigger customer solution (or business process), which means lots of interactions with the customer stakeholders.
- How frequently do they receive change requests from their customers?
- Do they help the customer in any change processes to better fit their AI product?
- How are such change requests bundled into the product or the algorithms?
Of course, this is not an exhaustive list, and there is no perfect answer to any of these questions.
This is a sample list to dig deeper if you are a fund investing in early-stage startups working on AI-powered solutions. Deep tech startups need an early infusion of cash, and we hope asking some intelligent questions will help you better grow the breadth and depth of the product and the team.
This is a sample list to start thinking and planning for if you are a startup trying to impress your investors during the due-diligence phase. Investors always look for the right data-driven venture to invest in and celebrate in a few years. However, the tech, the team, and the concept may be difficult to validate for success.
We hope the above guidelines help.