Hitting the AI Adoption Sweet Spot

Hey there,

Coming back from vacation, I recently shared a story on LinkedIn about a road trip. The point was, that this 3,000 km trip wouldn’t have been possible without technology (a reliable car). But more advanced tech (a fancy sports car) may not improve the experience.

Like choosing the perfect car for a road trip, finding the right level of AI to enhance analytics is about hitting the "sweet spot". More advanced tech doesn't always mean a better solution.

So, in this newsletter, we'll explore how to find that sweet spot and avoid both over- and underutilizing AI.

Let’s go!

Add more AI, you might actually get less value

According to a recent McKinsey study, the number of organizations implementing AI has grown by 2.5x over the past couple years.

At the same time, growth has hit a plateau.

While these numbers are taken from last year’s report, the 2023 edition (which unfortunately did not continue the fine graphic above) confirmed that AI adoption has stalled - even Generative AI didn’t change this.

A key reason for this is that many organizations have realized that deploying more AI technology does not necessarily lead to better results.

In fact, without the necessary understanding and expertise, blindly applying AI can actually lead to negative returns.

As AI technology quickly advances and becomes more accessible, it's often adopted by individuals or organizations that may not yet have the capability to fully leverage its potential.

For example, if you deploy a complex AI system on a critical core business process without the necessary readiness in terms of people and data, this will not end well.

That's why you should carefully select the "dose" of AI in your business, much like choosing a car for a road trip.

When I talk about “dose”, there are essentially 3 levers you can pull:

• the level of AI service integration,

• the degree of AI service automation,

• the level of complexity of the AI service.

Let's walk through three common scenarios that organizations face when adopting AI and discuss how to adjust the levers.

Scenario 1: Overreliance on AI

This scenario happens when organizations start completely delegating decisions to AI, without understanding the underlying system and even when the recommendations might seem counterintuitive.

You might instantaneously think about classical corporates here, but the prime example actually comes from the startup space:

A couple of years ago, Zillow, an online real estate company, lost $304 million because their AI overestimated house prices. This - among other factors - led to big losses when the market slowed, forcing Zillow to sell 7,000 homes and cut 25% of staff. This shows the risks of overrelying on AI without oversight, even for large tech players.

Here are some key signs of overreliance on AI:

• No human intervention in decision-making

• Inability to explain how AI works or its limitations

• Never questioning AI even when counterintuitive

• No or poor monitoring systems to detect model drift

• Total dependence on vendors to manage AI systems

If you're hitting too many of these points, chances are you’re using more AI than you should.

To get out of this situation you should:

• Keep the level of integration, but don’t push it anymore

• Remove the level of automation, e.g. put a human in the loop for critical scenarios

• If you can’t find the root causes of what's going wrong, consider removing complexity of your AI system. Try a simpler model or a non-AI baseline instead and add back complexity piece by piece in a controlled manner.

Scenario 2: Underutilization of AI

On the flip side, some organizations underuse AI due to fears of risks or a lack of resources and knowledge. However, this leads to missed opportunities.

Some industries are particularly susceptible to this. For example in traditional sectors like finance and manufacturing, there’s a plethora of business processes that could be used to inform data-driven decision making and ultimately power AI systems. However, old legacy systems and low analytical maturity, paired with a non-digital company culture, often hinder AI adoption.

Some signals you may be underutilizing AI:

• You haven't thought about AI yet and what it could do for your business

• You have large amounts of relevant, high-quality data, but rely on manual reporting

• Your business is asking for custom reports, but you lack resources to deliver

• You’ve never launched even a single AI prototype in your organization

If any of these points resonate with you, it may be time to reconsider your approach to AI.

Here’s how you should pull your levers:

• Don’t integrate AI yet - build a prototype on the side to validate value and collect (internal) customer feedback

• Don’t automate yet - aim for augmented solutions that support an existing manual workflow and solve a concrete pain point

• Don't be afraid to use an off-the-shelf solution for your prototype. Even if this means solving “simple” problems with seemingly “complex” solutions like Auto ML, it’s probably still faster and easier for you than building and deploying your own decision tree model in sklearn. Bear in mind: it's essential to have a basic understanding of the models these systems produce.

Scenario 3: Hitting the Sweet Spot for AI

The key is to balance human judgment and AI. Keep people in the loop to combine expertise with AI-generated insights.

Here are some examples of companies that have found the "sweet spot" for applying AI to enhance their business:

• Netflix uses AI for predictive content recommendations, but human curators still create genres and collections to deliver the most relevant titles to each viewer. The combination drives their recommendation engine, increasing overall time spent on the platform.

• Spotify employs AI to customize playlists and surface new music recommendations, but human experts are still needed to categorize tracks, and understand cultural nuances. The mix allows personalization at scale and discovery of new artists.

• Airbnb developed an AI-based pricing tool to recommend competitive rates for hosts, but still allows hosts to set prices themselves based on demand. The data aids human judgment and drives more revenue for the platform.

Good signs you've found the sweet spot:

• AI feels indispensable to your workflow

• Conceptual grasp of how the AI works

• Recognize AI limitations and when to verify results

• Able to easily explain AI benefits and trust recommendations

• Started small and scaled successfully

If that’s you - congrats! I’d love to hear about your use case!

Conclusion

In the end, think of AI as that ideal car for a long road trip.

It's not about the flashiest tech or the loudest engine, but about reliability, understanding, and partnership.

Especially in analytics and BI, the goal for AI is generally to support better decisions, not to replace people.

And just like with any good road trip, the journey matters as much as the destination.

Keep that in mind, and you'll always find your AI sweet spot.

I hope you enjoyed today’s newsletter.

See you next Friday!

Tobias

Want to learn more?

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

2. 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).

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