In this first article about one of the most important strategic decisions in AI, I want to explore the basics of that decision. Later, we can go deeper. The first principle is well-known to any C-level executive or strategist: Cost vs. Benefit. Here, we will translate it into a distinct equation:
(Value - Investment) x Risk
Comparing these factors for a solution you buy versus one you build can help guide your decision. The value that AI will bring to your business is the "Benefit." The "Investment" is the cost. And the new and extremely important component in this equation is "Risk."
I don't pretend to fully clarify a topic that is so complex it remains open in both academic and business circles, but I want to share some experiences on how to estimate the value of AI for your business. In my framework, I think about two principles:
1 - How much value will this product/feature deliver to my business?
2 - How much of that value is dependent on the chosen solution (AI/GenAI, Bought vs. Built)?
I will give two examples from my experience in a large unicorn and as a solo entrepreneur.
Big Unicorn - Autonomous Pricing System: I was working at a leading unicorn, and the decision was made to create a pricing solution. The value here could be measured by two factors: the estimated revenue increase from adjusting prices with AI, and the flexibility of the solution to follow our desired strategy. Both were critical to the company, with a direct impact on revenue and growth. The dependence on a custom AI solution was easy to analyze: at that time, no existing solutions could do this. If we chose to create that product, we had to develop all the code in-house.
Solo Entrepreneur - AI Content Creator Agent: While developing my AI Content Creation Agent for Travel Therapy AI, I needed to choose between coding it in Python (accessing the GenAI system API directly) or using No-Code/Low-Code solutions. The value of the agent to me was huge, for two reasons: learning (I wanted to learn how to create these agents) and capacity (I believed that with AI agents, I could publish much more content and increase the web presence of Travel Therapy AI). However, I could achieve this result in several ways, from simple (using tools like ChatGPT agents or Google Gems) to complex (creating a detailed Python structure). The difference between the solutions was control. The Python solution gave me total control over how I adjusted my data and delivered it to the GenAI system. Here, we had options, and the most complex one (Python) delivered slightly more value because, while reaching the same objective, it gave me more control and flexibility.
Investment usually starts in C-level discussions about how the company will approach AI. For didactic purposes, I will focus on the act of deciding which projects an existing team will focus on, analyzing the opportunity cost of allocating an AI squad to a given project.
Unicorn - Pricing System: Fortunately, we had a team with state-of-the-art scientists, so we could invest in complex new solutions. To estimate the level of investment, we had to draft the solution and estimate the delivery time. My main tip here is to allow a lot of time to pivot, because if it is real innovation, you will. And if you are building something that goes beyond even the existing scientific literature, include time to think, discuss, and research.
Entrepreneur - AI Content Creator: Fortunately (or not), I learned Python before the rise of no-code solutions, so I already had the needed skills. But one investment helped me a lot without reducing the delivered value: using assisted code systems. My first two versions were built the old way—coding, asking GenAI for help, and searching on forums like Stack Overflow. My third version was made with a Google solution where you can prompt your solution, and the tool generates the code. With that solution, I did the work of two weeks in hours. I still needed to check and change a considerable part of the code, but it was a great solution. The main lesson here is that to use an existing tool effectively, you need to deeply understand the expected result and be able to evaluate and adjust it.
This is the most complex and valuable component of the decision.
Is my team able to develop the solution?
Will it be developed in time for Go-to-Market?
Can an in-house solution for this product generate a durable competitive advantage—a real moat?
The main point here is to identify which parts of your value proposition are critical to your company and, preferably, won't be made obsolete by Big Tech players like OpenAI, Google, and Deep Seek. It is a huge risk to develop something that they will probably offer to your same clients in six months.
Unicorn - Pricing: This was a highly specific solution for our market. It was highly probable that the only companies that would develop a similar system were our direct competitors. So, the risk of our solution becoming obsolete because another player created a packaged product was low. The higher risk was failing to deliver the solution, but this was offset by the expected return.
Entrepreneur - Content: My content creator was specific to Travel Therapy AI, a startup focused on uncommon travel. This was my biggest moat. Other companies might develop similar systems, but not with my niche focus. To develop that solution, I had to customize it deeply, so building was the only option. I always faced the high risks of not delivering and of a big tech company delivering a similar system, but my focus on a niche that was not interesting to them reduced that risk.
The main driver in the buy-or-build decision is how close the solution is to your core value proposition and competitive advantage.
If you buy a shelf solution, you have no advantage over your competitors in the solution itself (you can have an advantage in your data, but not the tool). In general, for the core value of your company, if there is a high dependence on AI, building is a great option because, in our equation, the Value is so high that it surpasses the Investment and Risk. But if it's a secondary layer of value, especially an established process, using an existing solution is often a good choice.
Of course, this is a simplified article. In reality, you will need to make this decision for each part of your solution. For both of my examples, it was okay to develop the main code, but it made no sense to internally develop a specific AI architecture or train a foundational LLM from scratch. Equilibrium is everything in these complex decisions.