Principles of Effective Workflow Augmentation

When was the last time you moved?

I'm moving this week, and in case you forgot what moving looks like, this is essentially it:

Moving is really a big decluttering exercise. And letting go of things can be tough - even if it's just temporarily. Our youngest used to unpack everything we packed. "Why are you putting the gloves away? What if it snows?!!" he commented. In July.

Just like in our personal lives, business processes clutter over time for "reasons". And right now, a lot of businesses find "reasons" to clutter their processes by sprinkling "Use AI ✨" over them.

However, the most effective way for AI workflow augmentation isn't to add more complexity, but to take complexity away and put less burden on the user, allowing them to get things done better and faster.

Today, I'm going to show you an example of how this can look like in practice as well as the augmentation principles behind it.

Let's dive in.

See Effective AI Augmentation in Action

The video below showcases the new Math Notes feature on Apple iPads. If you watched their WWDC 2024 keynote, this shouldn't come as a surprise. I found it to be the most exciting feature announced, not because I use the calculator frequently, but because I believe it's a perfect example of effective workflow augmentation.

With 1,000+ reactions and 100+ comments on LinkedIn, it seems that many people agree:

The interesting thing here is that Apple doesn't even mention AI even though it's a big game changer. While being able to write on a screen and have it recognize handwriting to perform fancy tasks isn't new, AI and machine learning take this to a whole new level.

So before we dive in and explore what makes this augmentation so special, let's briefly review the use of AI in this example:

1. Track the pencil movements: Notice how fast the pencil moves and how accurately the display returns a smooth, steady line almost in real-time. It's likely that machine learning is used to optimize this capture.

2. Recognize the handwriting: Observe how messy some digits are written, especially the 7. Thanks to AI/ML, the software can still recognize it.

3. Keep track of the equations: There's no form capture here, so the software needs to determine which numbers belong to which equation, both horizontally and vertically.

4. Return the output to the user: While the actual computation of the equation does not require AI, figuring out what the user actually wants certainly does. When the user writes an equation, they want the result. When they draw a line under a block of numbers, they want a summation. If they declare a variable, they want to keep track of it. AI can help classify the user's intent accordingly.

5. Recreate the handwriting: The app will create a font on the fly that looks like your own handwriting, allowing it to blend seamlessly with the existing writing.

6. Find and update previous entries: When the user starts writing longer, more complex equations with variables, AI can track which variable belongs to which equation and organize information under the hood, enabling the calculator to perform the necessary calculations.

Full disclosure - I don't have any inside knowledge of the processes running internally here. Only Apple does. But if I were to leverage AI for this given use case, that's how I would approach it. And that's not even all – note that we didn't use Generative AI here.

TRICUS Augmentation Principles - Why This Feels Like Magic

Let's explore the principles behind this augmentation that make it so effective, almost magical. In chapter 5 of our recent book "Augmented Analytics," Willi Weber and I describe the key characteristics of effective workflow augmentation: TRICUS.

TRICUS stands for:

• Timely: Give users insights just at the right time

• Relevant: Align with the problem the user wants to solve

• Insightful: Provide meaningful and actionable information

• Credible: Users can trust and rely on the insights

• Unobtrusive: The user's workflow is not disturbed

• Specific: Deliver precise information tailored to the user

If you want to dive deeper, check out our book. For now, let's see these principles in action within the Math Notes feature.

How the TRICUS Principles Apply in This Example

The calculator augmentation works so well because it checks off (almost) all of the TRICUS boxes:

✅ Timeliness

As you can see from the video, the responses are coming in super fast, almost instantaneously, but not so fast that it disrupts the user's writing process (we'll come back to that in a bit).

✅ Relevancy

This one is super important. The calculator app "magically" recognizes what the user wants. For example, see how the moment the user writes a mathematical equation, the app suggests drawing that function as a graph.

✅ Insightful

The workflow is able to spot relevant context and suggest actions to the user. For instance, it might recommend adding a new function to the same plot so they can compare both functions side by side.

❗Credible

Can the user trust and rely on the outputs? I think this is where this augmentation could be improved. I don't know about you, but at the beginning, I double-checked if these calculations were actually correct. If the app recognizes handwriting incorrectly, the result will be wrong too. We have to see how the final implementation looks, but it seems the UI has some ways to highlight low-confidence recognition to the user, as indicated by the blue box around the digit 9 below.

✅ Unobtrusive

This is where Apple really nails it. The augmentation just works. The user does not have to click any extra buttons, request some AI ✨ magic, or interrupt their workflows in any other way. The timing of the augmentation comes in such a manner that it does not block the writing flow and, at the same time, does not have too much wait time. Finding this sweet spot is critical and demands a lot of both UI experience and technical capability to deliver it.

✅ Specific

This isn't shown in the video, but if we look at the original WWDC demo, we can see that the user can select a previously defined variable, change it, and update all dependent outputs (results, graphs) accordingly. This gives them ultra-specific feedback in real-time and highlights the dependencies of that variable to the user.

Conclusion

"I can do all that without that fancy AI," you might say. I agree. Not all of us need an augmented Math Notes feature like this. But that's not the point. The point is that there will be a ton of users who will. Meeting users where they are and providing them with TRICUS augmentation is the basis for effective augmentation – be it with AI or whichever technology you use for this.

Apple didn't need a single new "AI" button in their app to deliver this experience. As Steve Jobs famously said: "It just works."

Embrace this mindset the next time you're working on your own workflow augmentation.

Don't add AI to your existing clutter, but redefine workflows so AI will actually allow you to remove clutter and provide timely, relevant, insightful, credible, unobtrusive, and specific insights when needed.

This is how you improve your workflows with AI.

See you next Friday!

Tobias