1. Scan

Scan for capabilities and technologies that may be applicable in your domain. These may come from startups, publicly available prototypes, or academic papers. In parallel to that, look for opportunities in your context that are unlocked by new AI capabilities. This should be an ongoing activity, or one that you repeat in short cycles (at least every 3 months).

2. Explore

For each potential opportunity, explore how good "state of the art" AI may work for it, by doing cheap and fast experiments. Use human guidance, e.g., try to get ChatGPT Pro to solve an example problem in the course of a longer chat session. Don't worry about the unit cost for processing each instance of your application. The goal here is to get an idea about the state of the art and if it may be at all capable enough for your application. Additionally, you'll gain enough clarity to be able to prioritize opportunities for subsequent steps.

3. Evaluate

Sometimes the Explore phase yields an off-the-shelf product that may be good enough at solving your problem. Congratulations! It's a buying decision now. If that product is not quite good enough, you have a make / buy / wait decision. Evaluating the solution is a bit different from what you're used to in classical software projects or products. For example, these methods are not 100% deterministic, there is an unfamiliar surface area for misuse, and checking a vendor's feature list is not enough.

4. Deploy

Deploying AI solutions involves more than just integrating new technology into existing systems. Specific challenges in this phase include establishing a data pipeline, ensuring adequate observability, and ways to measure the business impact.

5. Optimize

Operating in a live environment reveals new insights and challenges. Observing the system in action allows you to gather valuable feedback on its performance and impact, leading to opportunities for optimization.

Not every idea will progress smoothly through the SEEDO framework. Some may hit a roadblock at any phase, deemed 'impossible' for the current state of technology or your business context. However, this doesn't mean these ideas are discarded forever. Instead, they become part of a Portfolio of Opportunities — a collection of potential projects that are periodically revisited and reassessed.