Generative AI: Success or Death by 1,000 Papercuts

And how to choose the right strategy

I recently had the honor of taking the stage at the world's leading industry show, Hannover Messe, to talk about AI. One thing that really surprised me was that while Generative AI was all the rage, it was practically invisible on the show floor among the 4,000+ exhibitors. Finding hands-on demos was actually rather hard.

For a lot of industries, there really isn't a single "killer" Generative AI use case that takes center stage. Instead, Generative AI weaves itself into the fabric of daily operations, making small but significant impacts that (ideally) accumulate over time.

This is important to understand because if you approach Generative AI the wrong way, your AI strategy will suffer "Death by 1,000 Papercuts". But if you do it right, it's actually " Success by 1,000 Papercuts".

Let's find out how!

Generative AI use cases are often invisible

"What's the biggest use case for Generative AI?" is a question I'm often asked, but I still don't have a single, stand-out answer. Not because Generative AI isn't useful - in fact, I believe it's transformative - but because its impact is spread across countless smaller applications. Very unlike "traditional AI" systems.

Consider these two examples:

For an energy provider, accurately predicting demand to schedule wind turbine maintenance can easily save more than $100,000 in a single instance. That's a strong argument for investing in a "traditional AI" time series forecasting system.

Similarly, a midsize ecommerce site that uses AI recommendations to upsell just 20% of customers at checkout can see a six-figure increase in revenue. Hiring a data scientist to build this system might be reasonable.

But for Generative AI, we lack these kinds of single-point, high-impact use cases.


The Challenges of Scaling GenAI

Generative AI systems are incredibly hard to run at scale in a fully automated fashion, for two main reasons:

  1. Generative AI can become costly at scale, especially when using GPT-4 class models as a service with high token-based consumption.

  2. Generative AI's inherent creativity and propensity for "hallucinations" (which isn't a bug, but a feature), make it challenging to deploy in highly automated, scaled use cases.

For example, a factory robot is expected to perform the same action 10,000 times per day, with perfect consistency. Similarly, "traditional AI" systems for time-series forecasting or personalized recommendations (Archetype 1 - Supervised Machine Learning) consistently provide the same output for a given input.

To overcome these limitations, the best current approach is to focus on augmented use cases - allowing people to do better work together with AI, rather than having AI replace human tasks entirely.

And this is where it gets tricky.

There isn't a single Generative AI use case - there are thousands

When I run AI Design Sprints with clients, we typically end up with a sheer endless list of potential ways to utilize AI, especially Generative AI. This is great, because with the right prioritization and focus, you can easily build a winning AI roadmap. But without the right expertise, it can also be overwhelming.

Take the example of pharmaceutical company Moderna. As a ChatGPT Enterprise customer, they recently shared that they have 750 custom GPTs (chatbots) in production for everyday tasks like research, data analysis, and legal work. That's a staggering number when you consider that Moderna has about 5,000 employees - an average of one GPT for every 7 people.

Imagine the challenge of spinning up a system like this. As a recent business leader asked me:

“If I have to train and onboard so many people for so many different use cases, how do I make sure my training program is effective and that people actually apply what they've learned?”

Well, I said, you have two options.

Choosing the right strategy for deploying Generative AI

When deploying Generative AI in your organization, you can essentially choose between the following two scenarios (mix and match as needed):

Option 1: Go all-in. If you believe your industry will be heavily impacted by AI (which is the case for many industries like R&D, finance, or legal), your main goal is speed. You want to drive AI adoption as quickly as possible to stay ahead of the competition. For Moderna, this meant rolling out Generative AI within 6 months. To make this work, you need a change management or transformation program that looks at your organization holistically, identifies AI champions, and drives change across your company.

In terms of AI enablement, you go horizontal in this scenario. You’re essentially telling the same (broad) thing over and over again to different people.

Option 2: Adopt Generative AI selectively. If your whole company isn't heavily impacted (or you're not sure yet), it makes more sense to focus on business verticals first that will be impacted, such as Sales, Marketing, or Customer Support. Within these verticals, you can drive domain-specific AI applications rather than a horizontal spread across different departments.

In this scenario, you go vertical - telling things that get more and more specific to (more or less) the same people.

Choose the game you want to play wisely.

If you try to roll out Generative AI across your whole company without embedded change management, you risk "Death by a 1,000 papercuts". There won't be enough learning in your organization to accumulate wins. People will be disappointed and forget what they learned in their prompt courses.

If you go for the vertical approach, it's easier to rally behind a common goal (e.g., make sales more effective), engage with the same people more frequently, and create small AI wins where the accumulated sum is greater than the individual parts. That way, you can compound benefits over time and showcase big wins with Generative AI (e.g., "We increased overall sales efficiency by 25%").

You can ramp up the transformation program later if you need to - but you'll lose time. Whether you can afford that time depends on your industry.


While GenAI may not lead to disruptive breakthroughs in a single instance, its cumulative effect across lots of specific, integrated workflows has the power to transform your business.

But you have to do it the right way. If you get it wrong, you'll either suffer from analysis paralysis (and never get started!), or you'll get bogged down in all the little details of small-scale implementation work.

So choose your game wisely, and then execute the hell out of it by accumulating lots of small wins that compound over time.

It takes strategic patience and an agile experimentation culture to win in the AI age - skills that many companies don't have yet.

If you do - you're a winner.

Have you found your strategy yet? Hit reply and let me know.

Stay augmented to stay ahead.

See you next Friday!


PS: If you found this article useful, please consider leaving a testimonial or forwarding this to a colleague who could benefit from these insights.