Building An AI-enabled Startup In 3 Steps When You’re Not An AI Startup

Ever get the feeling that you can apply artificial intelligence in a meaningful and value-adding way to your startup but you are sitting on the sidelines? 

If the answer is yes, you are in the majority. The AI hype has become almost unavoidable at this point, especially for anyone working in tech. But many who are not at the cutting edge or at the nexus of obvious use cases may not see how AI will affect their business.

Yet the headlines persist, from doom and gloom articles about the coming job apocalypse caused by AI automation to reports of how AI will be our savior by helping us tackle massive problems like climate change and rising health care costs. And for those with little or no exposure to the fundamentals of machine learning — by far the most common technologies we refer to when we casually say ‘A.I.’ — it can be difficult to see how you should be using AI or if you should bother with it at all.

While AI will probably leak into every organization sooner or later, either through the tools we use or by the services we purchase, the reality is that it doesn’t need to be difficult or complex to get started, and there may even be solutions out there you can start testing on some of your problems.


Understand when you should (not) use AI

AI is not right for every problem, but if aligned properly to the right problem it can be enormously useful. There are three things, in particular, AI is good at: automation, scalability and flexibility. 

Automation helps speed up repetitive or difficult tasks. Sintecsys, a Brazilian startup, joined our AI incubator to create a machine-learning model to detect wildfires in the Amazon rainforest. The company had a system of cameras deployed to detect fires and had already been quite successful at reducing time to detection. With its first foray into machine learning, it was able to implement a model that further reduced detection time from 40 minutes down to five minutes. What makes the company’s story particularly inspiring is that it had no prior experience with AI. But because of the good data it had and the nature of its problem, AI ended up being a natural fit.

There is a common misconception, however, that you need a perfect data set to get things started. That is not only wrong, it would also mean that the vast majority of organizations will never be able to do any meaningful work with AI. In practice, the work required to find, curate and manage the data is often the biggest and hardest part of the problem. If your data is sparse, you still have several options, with one promising direction being leveraging the growing amount of open-source datasets. You may even be able to assemble your own data through web-scraping or other techniques.

Scalability is another great asset of AI. This means you can easily expand something you’re already doing. If the conditions are right, you can add more data and continue the same analysis with little or no effort. Child Growth Monitor, a startup tackling child malnutrition, found that health practitioners on the ground had to carry cumbersome and often difficult to use equipment to measure a child’s growth rate in order to detect malnutrition. Not only was it difficult to measure a child, but measurements were often done wrong due to poor training and led to bad data. By designing an app it could use on a mobile device, it could reach more children more accurately and scale it very quickly because it only required a mobile phone. 

And finally, the flexibility of AI and digital tools can allow you to design solutions that suit your particular needs, whether that’s creative experimentation or standardized reporting. You can customize a visualization to allow you to see a particular analysis without having to repeat the process.



But how do you actually get started? Anyone running a lean startup knows to start small. As Brant Cooper, author of The Lean Entrepreneur said, “The idea around lean is the elimination of waste. Don’t waste your time, money, resources, creativity, inspiration in building products nobody wants.” This applies to products designed for customers or internally for your employees. 

You should know that people are going to use the solution as soon as possible. That means starting small and experimenting. Pick a problem that would be a good fit for machine learning and apply a variety of approaches to it. Take an audit of the data you have. If you don’t have any data, try to identify good data sources in the public domain. 

The World Resources Institute’s climate team used this approach with their online partners. They needed to understand the narrative around nature-based climate solutions and were manually going through their partners’ websites one at a time — a very tedious and inefficient process. After employing a wide range of natural language processing approaches, the team created a dataset of 700,000 PDFs from websites in 14 hours. This resulted in a network graph, a knowledge graph, a recommender system, and a number of other tools that WRI could then evaluate, show to their partners, and greenlight for further development. 


Go at your own pace

Experimenting is especially useful if you are new to AI. The temptation can be to play catch-up with other organizations, maybe even organizations in your industry. Resist playing catch-up and go at your own pace. Validate your findings and move forward with the right approach. 

AI is an iterative process but it’s also cumulative. The smarter you and your machines get, the more benefit you will reap down the line. This is one of the key factors that makes AI different from other technologies. It continuously gets better as you get more and better data, and you refine your methodologies bit by bit. 


Create a strategy

After you’ve had some success, and a direction becomes more apparent, you’ll want to start thinking about longer-term measures. Depending on the size of your startup, bringing AI in may be a significant task. There are, however, other considerations beyond uptake and implementation such as: What platforms and tools are best for your organization? Should you go open-sourced or not? Do you need computing resources? Do you have the necessary talent in-house or do you need to hire? 

But alongside all those practical considerations are the bigger questions: How secure does your data need to be? Does it contain private or confidential information and if so what are the laws around storing and processing that data? There also may be ethical considerations you need to make to prevent bias or critical mistakes. All this needs to be factored into your strategy as you expand your operations and process more data.  

Terry Maher is the Startup Partnership Manager at Omdena. He enables impact-driven startups to join Omdena´s fast-growing AI incubator. Apply to join their program at

  • Originally published March 3, 2021