Overcoming AI fatigue: Practical advice for product managers

As a product manager, do you find the mere mention of the word "AI"exhausting? Many of our experiences look like this: Forces at the top of the chain at your company start mandating the use of AI across your workstreams. You start having conversations with engineers and other product managers where phrases like "everyone wants to use AI for everything" become commonplace.

You feel frustrated by the company's constantly shifting priorities around AI. If this sounds like your situation, you are not alone. For example, the 2024 Digital Work Trends Report by Slingshot found that 77% of workers feel confused about how to use AI in their jobs.¹

Many product managers silently navigate their mixed feelings about AI's increasing presence, but our community needs to have an open dialogue about it. How can we feel just a little more sane when faced with the daunting task of using AI for everything? I’m going to share a few tips that will help you avoid going down rabbit holes.

With layoff fear running high at the moment because of recent redundancies, it’s tempting to want to be a low profile and be highly agreeable. Between 2022 and 2024, for example, nearly 400,000 people working in the tech sector in the US lost their jobs (Crunchbase News). This caution is understandable, but I would argue that staying silent about AI, including your hesitance about its use, will only hurt you.

Now, more than ever, making our voices heard is crucial to avoid becoming collateral damage in a restructuring. The alternative? Watching from the sidelines as engineers who can rapidly adapt to AI technologies take over traditional product manager responsibilities. The path forward isn't avoiding AI discussions but rather shaping them with knowledge and confidence. Here's how to start.

Step 1: Develop a strong voice as a product manager

The best way to develop a strong voice on AI topics is to become a user of the technology. In 2018, while at UT Austin, I founded BizBot, a chatbot service helping entrepreneurs navigate the legal and regulatory challenges of starting businesses.

It was developed using IBM Watson, a chat interface that delivered abysmal results. My negative experience with chatbots had initially made me skeptical about OpenAI's ChatGPT. However, once I began exploring it hands-on, I quickly changed my mind.

As a product manager, I started envisioning numerous potential applications in finance. This positive shift in perspective gave me the confidence to volunteer for prompt engineering assignments at my company.

Getting hands-on with ChatGPT is a great way to gain intuition on how to use Large Language Models (LLM) models. Beyond generative AI, there are also predictive models which make projections based on historical data.

While predictive models like neural networks might seem intimidating, you don't need coding skills to develop intuition for their capabilities. Interactive tools like Machine Learning Playground let you experiment with various algorithms without writing a single line of code.

You can even upload your own datasets to test real business scenarios. For example, the visualization below shows a k-nearest algorithm classifying data points into orange or purple categories based on their characteristics.

A crucial insight into machine learning is that all data must be converted into numerical values for algorithms to process them. Text, images, categories, names, and almost anything else can be transformed into numbers.

Now that you've built your understanding of AI technology through firsthand experience, it's time to leverage that knowledge to lead your organization away from misguided AI initiatives through strategic veto powers.

Step 2: Know your AI veto powers

"We can’t pursue this product or feature because we don’t have the data to back it up." Data is the ultimate trump card when deciding which AI product features are practical to pursue.

Consider Spirit Airlines, which is known for its limited customer support. Their management might be tempted by Agentic AI promises to solve customer service challenges. However, developing an effective chat-based model would be impractical without substantial customer interaction data.

Yet, companies pursue dead ends like this all the time. As a product manager, it's our role to call out problems with data availability that could prevent us from being able to produce a successful product. The traditional advice is to start with a problem and then find a solution. When it comes to AI, I suggest trying the reverse. Ask yourself what data you have and if there are any valuable problems it can solve.

By using data availability and quality as your guide, you can more easily redirect unrealistic product visions toward achievable outcomes. "We can’t pivot because we’ll lose our competitive edge."

Chasing trends often positions firms as followers rather than leaders. Every significant pivot carries the risk of market share loss. When these pivots involve switching technologies, the costs multiply. Teams need time to ramp up, systems require re-architecture, and new specialized talent must be hired.

For startups especially, strategy shifts typically extend the path to profitability while increasing burn rate. To bring sanity back, focus management on how pivoting will impact your market share. Consider IBM Watson Health's cautionary tale.

In 2011, IBM's Watson pivoted aggressively into healthcare AI, investing billions to revolutionize cancer treatment. They promised Watson could analyze medical literature and patient data to recommend superior treatment plans.

However, by 2022, IBM Watson Health was considered a failure and sold for a fraction of its initial investment value². Meanwhile, less ambitious but more persistent players in the field, such as Flatiron Health continue as a business today and were acquired for 1.9B³.

Step 3: Remember to fail fast

So, you've shared a warning that a specific AI strategy might fail, but your organization still wants to pursue it.

Pivot your approach toward failing fast. You might even discover that the chosen AI path is worthwhile after all.

LLMs turn out to be an easy way to validate whether a problem could be solved with a machine learning approach.

You only need to secure an account to begin using a model like Llama or OpenAI’s ChatGPT. Let’s say I want to predict what months are the most popular for eating ice cream. I could upload a CSV containing a month-by-month tally of how many ice cream cones are consumed daily. If the LLM can respond reasonably to this question, then I can infer this may be a good problem for machine learning.

The next step for me as a product manager would be to take the same data set and results to our development team and ask them to run further experiments using a design approach they deem appropriate. Other ways to fail fast (or not) include conducting a focus group to determine if there is willingness-to-pay for a product or by deploying an Minimum Viable Product (MVP).

Finding your voice and avoiding AI rabbit holes

The next time you need a leg to stand on in an AI product or feature debate, ask the following questions: "Is there sufficient (annotated) data available to solve this problem?", "How much will it cost to pivot our AI strategy, and how will this impact our competitive edge?" and, "Is there a surprisingly quick way to validate this new approach?".

The answers to these questions should help your team avoid entering an AI spiral. The AI revolution is here, and as irritating as it may be to have this technology thrust upon us, adaptation is essential.

Develop a strong voice on AI product strategy to prevent engineering or management from dominating decisions. Know your veto powers as a product manager and redirect your team toward higher-ROI opportunities. Start with quick LLM tests to validate concepts before deeper investment. And remember – all transformative technologies create challenging transition periods, so patience with yourself and your organization is key.