What Technical Questions an Investor can ask to AI Startup Founders before Financing?
Success of an AI startup company depends upon multiple factors among which the ‘AI technology’ plays a major role. Here's list of technical things an investor should consider before financing an AI startup.
- Secret Ingredient: No one can claim that they own an idea. Hence ideas cannot be copyrighted. A novel and interesting idea of a startup can be duplicated by an alternative emerging startup. So, the only factor that differentiates a startup from another (both working on same idea) is the ‘solution’ part. Does the startup has a unique AI Model(s) which is publicly not available where no other competitor can duplicate it? As there is no single solution in the AI domain, how many AI models are getting developed by the technical team?
- Scale of Data: Is the scale of the data (number of samples) for AI model training enough to cover different distribution of future data coming-in? Is the number of samples for training in the scale of thousands or millions? What is the cost of acquiring more data? More data is better. In fact, more data with good variety is much better. But too much data is hard to manage sometimes.
- Data Source: Does the data is aggregated from single source or multiple sources? Does the single data source cover good variety of samples? Do the data from different sources have similar samples or variety of samples? Do the data cover variety of Geographic (country, region, state, city, etc), Demographic (socio-economic factors such as class, gender, race, age, occupation, income etc) Cultural (education, cultural origin, language, religion, race), Behavioural (attitude towards something, usage rate, and performance), Psychographic (activities, interests, and opinions) factors.
- Quality of Data: Is the data used for AI model development is of good quality? Is there any error in human labelling or annotation in the data? Do the data have any noise or human bias (related to race, religion, language, or gender)?
- Expected Future Data: What is the equipment or hardware that generated the training data (used for AI model training) and the hardware that will be used by customers to generate the data? Where is the end user application going to be deployed? What is the nature of expected data from customers? Is it going to be looking similar to what you have used in the development? How much is the similarity between the ‘data used for AI model development’ and the ‘data that will be coming from the users’? It is much recommended to have training data (used for AI model development) similar to future incoming data (from customers).
- Adaptive Learning: What's the strategy for improving the AI product performance when new data comes in? How often the AI model will be retrained or fine-tuned? How the AI model will be evolving over time? How the model will be trained based on failure cases reported by the customers in real-time?
- API Usage: How much external AI related APIs (or readymade solution) are to be used in the product? What will be the extra cost involved? What is the performance (speed and accuracy) of the APIs to be used in the product? If the startup idea can be solely implemented using ready-made APIs available, then it’s not worth to continue the idea.
- Competitor Status: What is the performance (accuracy and speed) of the competitor’s AI solution available in the market. It’s good to have few competitors in the game as it signifies that there is a requirement for such product in the market. However, the technical team of the startup should be working on improving their own model performance. Is there any alternative technology evolving under research which has potential future to be a better solution?
- Development Timeline: How long the research will take for the AI model development? What are the required resources for storage and computation (and cost involved) if the AI models have to be trained over cloud (or on premises)? What is the cost involved in
- Customer Engagement: What is the role of the customers’ feedback in AI model training? Will the model provide generic results or personalized results to the customers? If the AI model is based on user’s activity or personal information, how the privacy of the user is preserved in the whole process? How AI model will be secured over cloud/phone/premise?
- Market Demand: Whether the product is currently required in the market or not? If so the analysis of the survey taken for market demand. Due to lack of understanding of the customer and market requirements, many of the startups fail to gain revenue. There is always a clear distinction between fascinating ideas vs the solutions which customers are ready to pay for it. An idea can be novel but might not be demanded in the market by customers.
The founders of an AI startup can also consider the above as a checklist to deliver a cost effective & perfect product in the market.
Thanks for your time!
Bye from Dr. Sridhar Swaminathan
Published on 08 September 2021
Share
About the Author
