The Integration-Automation AI Framework

Hey there,

I've seen hundreds of different use cases for AI in different flavors. Every use case that failed failed in its own way. But all that succeeded had one thing in common: They started augmented.

I can't say it enough: Knowing which AI use case to start with is probably the most critical piece of your AI journey.

That's why I've created a framework to help you navigate this space.

Today, I want to share it with you.

Let’s dive in!

AI and automation are two different things

Most people getting started with AI think about automation first:

• “AI” can automatically reply to my emails

• “AI” can automatically analyze my budgets

• “AI” can automatically optimize my lead flow

• etc.

But wait a second.

Starting your AI journey with any of the use cases above is probably a horrible idea.

The use cases above are all examples of highly integrated and highly automated use case, which are the most difficult.

But where should we start?

To find this out let’s take a look at the following matrix which I call the Integration-Automation AI (IA-AI) framework:

The Integration-Automation AI Framework

This framework is based on a simple principle: We classify use cases based on two criteria - how integrated and automated they are.

Integration: Refers to how well the AI system blends or works in synergy with your existing system landscape or business workflows.

Automation: Refers to how much the AI system can perform tasks with minimal human intervention.

Think about this for a second.

Most people would assume that integration drives automation and vice versa. But this isn’t necessarily true.

For instance, you can create a use case that includes some automation but is not fully integrated into your system environment.

Let me show you how by walking through the different fields.

Type 1: Assistants

The first type of use cases is what I call “assistants”.

In these projects, you typically work with an external AI application that requires you to handle inputs and outputs manually (like copy/pasting text or uploading/downloading a file).

The typical degree of integration is low and so is the automation.

ChatGPT is the classic example for this.

Make no mistake - these apps can be increasingly powerful. Especially ChatGPT is probably the most capable external AI assistant you’ll likely encounter (and it’s only getting better).

Other examples include:

• Poe.com

• Google’s Gemini

• Microsoft’s AI-Powered Bing

• AutoML tools like Obviously.ai when you upload/download files manually

Type 2: Copilots

Similar to assistants, Copilots require you to do the heavily lifting (i.e. their degree of automation is low), but they are more tightly integrated into your system landscape or business workflow.

This is currently the hottest field in AI and every software company you can think of is currently looking to include these “Copilot” features into their product:

• GitHub Copilot

• Microsoft Copilot

• Einstein Copilot in Tableau

• Duet AI in Google Workspace

• AI features in Slack, Notion, and co.

• etc.

Copilots are great because they know what you're doing at the moment.

For instance, if you use Outlook Copilot to reply to an email, it will already know the previous conversation history from that email thread.

No copy / paste necessary.

It also eliminates the friction of signing up for a new app.

However - and that’s the biggest difference compared to the next use case type - the AI suggestions cannot be implemented without your approval. So you have the ability (and responsibility) to review and correct the AI outputs as needed.

That's why this use case type is so popular right now. It offers great value, even if the AI models aren't 100% perfect (which they probably never will be).

Type 3: Autopilots

Autopilots shows a high degree of automaton, but a low degree of integration.

So what’s that?

Imagine you trained an AI-powered chatbot on your documents to answer first-level customer support queries.

Once deployed, this chatbot runs fully automatically, answering customer questions 24/7 without manual intervention.

At the same time, this chatbot isn't really integrated. It usually boils down to putting a small HTML or Javascript snippet on your website (I recently did this with my book. You can chat with it here.)

For example, if you tell my book chatbot “I want to buy the book” it will return an Amazon link, but it won’t be able to take the order and put it into a booking system.

If you want to do that (which is technically feasible), you’d need to increase the degree of integration.

This would move the project up to an "Agent" use case.

But before we discuss “Agents” in more detail, let’s review some more examples of “Autopilot” use cases:

• A social media tool that does automated content moderation

• An automated surveillance system that sends email notifications

• A robotic vacuum cleaner in your house cleaning on schedule

• etc.

Type 4: Agents

Agents are the holy grail of AI projects. Everyone wants them.

For instance, imagine having a customer support chatbot that can answer common questions and also perform tasks like resetting passwords, giving refunds, and taking orders.

To be clear, these use cases do exist, but they are the hardest nuts to crack.

Because at this stage, you're not only dealing with the challenges inherent in AI (inaccuracies, hallucinations, performance issues, to name a few), but also good old-fashioned IT integration issues involving a bunch of legacy applications.

Further examples include:

• Automatically processing & forwarding incoming documents

• Serving personalized recommendations on your website

• Flagging quality issues in a manufacturing line

• etc.

I'm a big fan of these use cases, and my goal in working with customers is to get them there as quickly as possible.

But they are not the best way to get started.

So what's the best way?

Where to start and where to go

There are two driving forces in the IA-AI framework:

• You either proceed from left to right (low automation to high automation) and increase the complexity of use cases by automating them more.

• Or you go from bottom to top (low integration to high integration) where the main challenge is to integrate what you have even further into your systems or workflows.

• It’s typically not a good idea to to both at the same time (i.e. moving diagonally and increase the level of integration and automation at once - so don’t jump from Assistants to Agents.)

From my experience working with different AI projects across multiple domains, the best approach is to typically tackle use cases in the order I introduced them in this newsletter:

Assistants → Copilots → Autopilots → Agents

The key is really to start with so-called "augmented" use cases (Assistants and Copilots) first, because they always keep a human in the loop.

This gives you two major advantages:

1. You can control the AI output and ensure quality results

2. You learn more and more about how the AI works over time

These two factors are critical components for ensuring long-lasting AI success. They will increase your AI maturity - a concept which I’ll explain in a future newsletter.

And the beauty of these augmented use cases is that they are typically the least complex which makes them great candidates for an Atomic AI use case.

Conclusion

I hope the IA-AI framework helps you navigate the intricacies of getting started with AI.

First, make sure there's a human in the loop before you give up control. No one would hire a stranger off the street to fly their jumbo jet.

Don't let that happen to your business.

Stay in control until you're ready to sit back and relax.

Enjoy your Friday and see you next week!

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.