Atomic AI Use Cases

Hey there,

When building a strategic AI use case roadmap you need a strong mix of smaller and larger projects.

However, some use cases that sound great on paper often turn out to be a disaster in practice. I know companies that dumped 6-figures in the prototyping stage alone. How do we avoid that?

I found the concept of so-called “Atomic AI use cases” work really well.

Let’s find out what they are and how to do this!

When you start - go small, or go home

So let’s say you came up with a list of potential AI use cases, for example by running an AI Design Sprint™ in your organization.

As a result of this process, you could have a use case matrix which looks similar to this:

In this example, three participants of the sprint voted that this use case has high potential benefit and relatively low costs.

Logic tells us we should jump right into prototyping with this use cases because it offers the best ROI - right?

The problem is that these champions rarely exist in practice- especially if you're just starting out.

What really happens during most brainstorming sessions is that people fall in love with a certain solution, judging the business value too high or blending out complexities. The bad awakening comes when actually building something and then people say things like "AI doesn't work for us", or "I think we're not ready for this yet".

Note: A similar thing could happen to potential quick-wins (bottom left of the matrix above), too. In practice, they might turn out way more complex than we thought.

So for both of these "low cost/high feasibility" use cases, we need another layer or filter so that they are not only feasible in a production scenario, but that we can easily prototype them to validate our assumptions.

That's where Atomic AI use cases come in.

What are Atomic AI use cases?

In a single sentence:

Atomic AI use cases are small, focused projects that promise tangible value and are very easy to prototype.

In some cases this might involve breaking down a potential quick-win or champion into even smaller stepping stones.

Wait a minute, you might say - isn’t every use case that that has high feasibility in the classical sense not automatically easy to protoype?

Not necessarily.

Even a quick win that seems relatively easy to scale in the organization might be really hard to prototype.

For example, consider a use case involving an enterprise AutoML platform that does not offer a free trial, has a minimum contract period, or has a complicated internal IT setup.

Or consider a use case that involves personal identifiable information (PII). While this may not sound like an insurmountable challenge in production, it will likely kill a quick prototype.

Therefore, we need to find use cases that are both potentially feasible to scale and feasible to prototype.

So how do we find them?

Identifying Atomic AI Use Cases

First, take all your use cases that would pass as a potential champion or quick-win.

Now, with that list, ask your internal IT department or your favorite AI vendor the following question:

Which of these can you ship in less than 20 days and for less than 20k?

Note: The recommendation of a 20 days / 20k limit isn't derived from formal research but is based on my personal experience working with SMBs. Over time, I've observed that these thresholds strike a good balance between feasibility and investment, allowing businesses to test AI initiatives without overcommitting resources.

To stay within these limits, you typically force the tech side to leverage existing tools, platforms or frameworks for building the prototype - leaning towards a "buy" instead of a "make" decision here. Another trait is that these use cases are typically not integrated in your existing IT environment.

That's exactly what we're looking for with Atomic AI use cases!

Aligning Atomic AI Use Cases

Having a single atomic AI use case is nice, but the real value is realized when you have a few of them compound to help you achieve a larger strategic goal.

For instance, a B2B market research company might want to learn how to use AI to extract useful information from messy documents (Champion use case). This could be an important skill for them to grow their business in the coming years.

Here are a few ways they could break this up into multiple atomic AI use cases leveraging a powerful AI model like GPT-4:

• PDF summary generation use case

• Chat with your documents use case

• Document quality assessment use case

• Semantic analysis and sentiment extraction use case

All of these use cases build on a similar type of data set and technology capability. Even if they aren't perfectly aligned, they help you work toward a larger strategic goal.

In another example for a financial company, their main goal might be to improve their predictive analytics capabilities with AI so they can stay more competitive.

Some atomic opportunities in this area may involve supervised machine learning capabilities based on relatively high-quality data:

• Forecast financial metrics from BI data with AutoML

• Automatically categorize portfolios into different risk groups

• Detect anomalies in financial reporting data with an AI service

All of these could be scoped as atomic use cases that would converge toward building a capability to leverage BI data and use supervised learning techniques to make smarter decisions overall.

Conclusion

Atomic AI use cases are a great way to get your foot in the door with AI. They help you build momentum and reduce risk.

You'll also learn a lot, and they often lead to unexpected rewards.

In these smaller projects, you will often stumble upon new opportunities or ideas that can help you improve or grow your current business.

So start small, learn fast, and set yourself up for success when you're ready to go big.

Thinking about implementing an AI use case?

Let me know if you need any help!

See you next Friday,

Tobias

Want to learn more?

1. Book a meeting: Let's find out how I can help you over a coffee chat (use code FREEFLOW to book the call for free).

2. Read my book: Improve your AI/ML skills and apply them to real-world use cases with AI-Powered Business Intelligence (O'Reilly).

3. Follow me: I regularly share free content on LinkedIn and Twitter.