Product Market Fit Collapse from AI

The Product Market Fit Treadmill

Product Market Fit is a key milestone to reach, but it’s often misinterpreted as being a static moment in time. The reality is that your customer base is always changing and consumer expectations are always growing. Once you get initial product market fit, you not only have to keep it but also expand it.

Casey Winters and Fareed Mosavat created the framework of The Product Market Fit Treadmill in the Reforge Product Strategy course to help visualize this:

• In the earliest days of a product it does not have product market fit. That is represented by the blue line below the brown line.
• As it gets closer to product market fit, company performance increases and crosses the PMF threshold.
• But that PMF threshold is not flat. It continues to increases because over time consumer expectations are constantly growing.

Technology Shifts Accelerate The Product Market Fit Threshold

In technology shifts, the PMF threshold accelerates. We can think about this as a slight slope increase in PMF threshold. In previous technology shifts this acceleration happened over a period of time. Customer expectations didn’t change instantly.

For example, in the mobile era, it took years for smartphones to reach mass adoption and for networks, hardware, and ecosystems to mature. Similarly, during the early days of the internet, it took considerable time before PCs were common in every household and broadband connections were fast enough for richer online experiences.

This “slow” acceleration gave companies breathing room to adapt. Even as they began to lose product-market fit (PMF) due to evolving standards and user expectations, this erosion typically happened over years, allowing product leaders time to plan responses, gather data, and course-correct. But something different is happening with AI.

AI Can Cause The PMF Threshold To “Inflect” Causing Product Market Fit Collapse

With AI when a use case works adoption is happening much faster. Powerful AI tools are readily available at minimal or no cost, and users can incorporate them into their workflows immediately. At a minimum, AI is causing the slope of PMF Threshold to be much steeper than previous technology shifts, and when a use case really hits the PMF threshold “spikes.”

Customer expectations aren’t rising at a predictable, linear pace over longer periods of time—they are spiking exponentially. Suddenly, “good enough” solutions look obsolete when users realize they can receive more efficient, hyper-personalized, and near-instant responses from AI-driven platforms. This creates Product Market Fit Collapse.

This accelerated pace means that once AI proves its value for a given use case, incumbent solutions risk losing their PMF almost “overnight.” There is no lengthy adjustment period to prepare for changing market conditions. There is no time to wait for conclusive data or for long-term strategic planning. The window for adaptation slams shut before they even recognize the severity of the threat. As a result, some businesses that thrived during previous shifts may find themselves struggling—or even folding.

Examples Of Product Market Fit Collapse

Let me first give a few examples of product market fit collapse, and then talk about why I think we will see more happen.

Example: Chegg

The Collapse

In Jan 2024 Chegg was valued at $1.2B. By October 2024 (9 months later) they were valued at$150M. A 90% decline in 9 months losing half a million subscribers in that time. In September 2024, their trailing 12 month revenue was $662.08M so they are valued at 1/4 of TTM which means the market believes they are essentially going to go to 0.

What Happened

Chegg’s primary subscription offers homework help to students. The main value prop was high-quality answers written by curated humans.

If we look at it through the lens of their growth model, their core growth loop is a company generated company distributed content loop. More quality answers, were then distributed via SEO and other channels, which led to more subscribers and engagement, which led to funding more quality answers.

But when OpenAI launched ChatGPT, students could just enter their homework and get an immediate personalized answer. The answer wasn’t always right, but the fact it was instant, cheap/free, and other attributes created both a 10X value prop and immediate distribution.

It broke this loop. As subscribers start to churn, you can fund less company-generated content, which leads to less new subscribers and engagement, which leads to less money, which leads to funding less company-generated content.

The Problem

As Casey Winters points out:

“Interestingly, Chegg has been through this before in the transition of physical media to digital. But that transition happened on a slow enough timeline to pivot away from textbook rentals towards homework help. With AI it happened much faster. So even for a team that knew how to adopt to business model threats, they couldn’t overcome this one.”

Chegg didn’t really have a chance. Customer expectations spiked, they immediately lost product market fit, and users started switching at such a pace that it didn’t leave time for Chegg to respond. Innovating your way out of a problem when you have an established business of that size takes time.

They are now caught in a deadly death spiral of churning subscribers, leading to losses, leading to dwindling cash reserves, leading to layoffs, leading to not having the time and resources to build their way out. Subscription businesses built on information access to facts or things AI models can answer are highly vulnerable.

Example: Stack Overflow

The Collapse

GitHub CoPilot and ChatGPT were launched at the end 2021. Traffic to Stack Overflow started to immediately decrease. The amount has been debated publicly between Stack Overflow and outside developers. No matter the specific amount, it has been significant and impactful.

What Happened

If we look at it through the lens of their growth model, Stack Overflow’s core growth loop was a user-generated, company-distributed content loop. A developer would ask a question → Another developer would answer that question to get social capital → The Company would then distribute that content through SEO and other channels → Which would attract more users and content creators.

When GitHub CoPilot and other solutions launched, users could:

• Get answers instantly vs waiting for an answer or searching through
• Those answers were directly in their coding environment vs having to go search Stack Overflow.
• Those answers were personalized to the code they were writing.

Yes, there were some issues with AI-generated code. But the value prop combined was strong customer expectations spiked and Stack Overflow lost product market fit. The core growth loop started to break down. People started voting, viewing, etc less which decreased the incentive for contributors to post answers. Just take a look at the metrics below from this post.

The Problem

Stack Overflow has now responded by launching things like OverflowAI and striking licensing deals with foundational models. The challenge is not only is their core growth loop broken, but they don’t own the touch points where developers expect to be able to generate this content with AI (direct in the coding environments). Community-driven content models are particularly vulnerable to AI disruption if they aren’t the first to leverage that content into an AI solution.

Example: Shutterstock + Getty

The (Coming Soon?) Collapse

While the collapse hasn’t happened yet, it seems impending in the stock photography space. Shutterstock and Getty images recently merged as a response to a threat in AI. Those businesses have been under threat as image based models and tools like Midjourney become better and better.

What Is Happening

Both Shutterstock and Getty’s growth models rely on contributors contributing content, which attract more paying customers, which is then shared with the contributors. If the money flow stops, contributors start contributing less, and the flywheel reverses.

As GenAI image models get better and better, the writing is on the wall. You will be able to instantly generate a very personalized image in any style you want within the design tool that you are using. A 10X value prop to using generic stock photography.

The models aren’t there for all use cases and styles and users are still learning how to guide them for what they want. But it seems the collapse is coming.

The Problem

It’s clear that both companies recognize this threat. The businesses a reinvesting in new AI solutions. But can they navigate out of it? It’s unlikely in my opinion. They are dealing with their core growth loop breaking and not owning the new touch point of user habit (design tools) at the same time.

More PMF Collapse Is Coming

All of my examples so far are companies based on company-generated or user-generated content. These have been the most vulnerable due to the content generation ability of early models. Does PMF collapse stop there? I don’t think so.

The capabilities of these AI models are progressing at an exponential rate. Below is how OpenAI’s models have been progressing on a set of benchmarks. The latest release of o3 by OpenAI points towards the technology now being beyond the inflection point of the curve.

But there is a lag between where the latest technology is and new products developed with that technology. It takes time for builders to figure out the best way to leverage a new technology with new capabilities. Most existing products are still built using models like GPT-3 or GPT-4.

New products built on o1 and o3 will come to market new soon. As the benchmark curve shows, their capabilities won’t be incrementally better they will be far better. As they do, we will see customer expectations spike in more categories leading to more examples of PMF collapse. I think this image inspired by Nancy Rademaker visualizes it well.

How To Avoid Product Market Fit Collapse

As product professionals, what can we do about it? This is an extremely hard task. The challenge as Casey Winters says is:

“Netflix and Chegg succeeded at transitioning from physical media (movies, textbooks) to digital. But others like Redbox and Gamefly didn’t. To avoid PMF collapse you have to predict it well ahead of time and then bet the company. ”

There are three areas to get a handle on:

• Understand how your customer expectations are changing.
• Evaluating your level of risk for PMF collapse.
• Allocating your product portfolio of bets accordingly.

Let’s talk a little about each one.

Understanding How Customer Expectations Are Changing

How Customer Expectations Are Changing At The Micro Level

The first step is to have a tight pulse on how customer expectations are changing around your use cases. In a lot of companies, gaining customer understanding is a brutal process:

• A product team has some questions they want to understand.
• They then get in line with a user research team to execute some research.
• That research teams executes the study then synthesizes the results and presents it to the product team (often months later).
• Those insights then collect dust in powerpoint decks and google drives.

In a lot of enterprises, product teams aren’t even allowed to talk to customers directly. This process has three problems:

• Too Slow - The process can take months to get answers.
• High Friction - The process to get customer feedback is so high friction that product teams start to not bother at all.
• Disconnection - The builders are disconnected from the nuance and context of the actual conversations.

We talk about some of these in the different type of Feedback Fragmentation Tax that exists in most organizations. I believe that one of the ways that AI-native product teams will work differently is rethink this process.

This is what we solve with Reforge Insight Analytics. The product aggregates all the sources of customer feedback, uses AI to help you analyze and explore the data, and then puts it directly in the tools the builders are already using in real-time.

Whether you use something like Reforge Insight Analytics or not, customer expectations are changing too fast. If this process isn’t fixed, PMF collapse becomes a higher probability.

How Customer Expectations Are Changing At The Macro Level

It is also helpful to understand how AI is changing customer expectations across technology products. Some examples that are shifting:

“A Place For Me To Create” → “Do The Work For Me”

Many software products we use today are tools that enable us to create different content and experiences. Canva, Notion, Google Docs, Gmail, etc. But what if that work was just done for us? AI is changing customer expectations from “give me a tool where I can create” to “do the work for me.”

“One Size, I Customize” → “Custom Made For Me”

Many B2B are made in a way that requires the customer to do heavy customization to their process, workflow, and data. Take the CRM category as an example. Setting up a CRM requires so much work that there is a multi-billion dollar per year services industry around systems integrators configuring Salesforce and other CRMs to meet your need. But what if that wasn’t required?

Manual tasks → Automated Tasks

Many products require you to perform a lot of manual tasks in order for the team to get value out of it. As a product team the one that probably hits close to home is JIRA. The time required to create tasks, keep them up to date, groom them over time. Or the CRM category that requires sales reps to spend endless hours inputting data about their conversations and deals. But what if these were automated?

Pay For Seats or Month → Pay For Work Completed

In the past 25 years, a lot of products consumer expect to pay per seat or per month. The customer does some rough calculation in their head if their use of the product is approximately of greater value than what they are paying for. The actual value is a step removed from the price. But what if you could pay for the actual value?

Understanding Your Level Of PMF Collapse Risk

The second thing is to assess the risk level of product market fit collapse. I go much deeper on this in AI Strategy with Ravi Mehta but here are some of the high-levels.

1. How directly do you own the customer relationship?

There are products that have tight ownership over the customer relationship (i.e. Github). There are others (i.e. Stack Overflow) where Google or some other channel tend to be the primary way users enter and return to the product. Product market fit will be easier to maintain and defend the more you own the customer relationship. A way to assess this is to measure the percentage of your users that come directly to your product vs through an intermediary.

2. What is the frequency of your use case?

In Retention + Engagement we focus on defining your use cases and identifying your natural frequency of usage. There are low frequency products used 1 - 2X per year (i.e. Travel) and high frequency products used daily (i.e. Slack).

Low-frequency products are at higher risk in my opinion. The low-frequency nature provides easier opportunities for a user to switch the next time the need comes around. The habit is not as strong. High-frequency products have an established habit with users and habits can be hard to break even if there is a better alternative on the market. Learn how to determine your natural frequency here.

3. Do you own the creation workflow?

AI’s “killer use case” often emerges exactly where the user creates something—in the coding environment (e.g., GitHub Copilot), the writing surface (e.g., Notion AI), or the design canvas (e.g., Figma). If your product sits “downstream” or outside of these creation surfaces rather than being the place where users do the core work, you’re more easily replaced. AI can directly integrate into the workflow and be a more seamless alternative vs leaving their primary work environment (like leaving a dev environment to Stack Overflow).

4. Do you have proprietary data?

Data is the new oil (for real this time) in an AI world. Specifically proprietary data that the foundational LLM’s and models don’t have access to. If data (or content) is available and ingestible by the large language models then that is not defensible. The more proprietary data you have, the less risk of product market fit collapse.

5. What would break your core growth loop?

You need to understand your growth model deeply. Don’t just map your growth loops, but understand why a user takes each step in their growth loop. If that “why” breaks then loops start to spin in a negative vs positive direction. For example, growth models that rely on user-generated content can unravel quickly if the incentives for contributors vanish. This is what we are seeing in the Stack Overflow case.

6. How tech forward are your customers?

Fareed Mosavat had a good point:

“We’re seeing the real disruption today at the tip of the adoption curve (code, design, tech, students). Businesses that cater to less savvy customers are likely less susceptible.”

The earlier your audience tends to be on the adoption curve of new products, the quicker product market fit can break. They are willing to try new things, break old habits, and have low loyalty if a clearly better alternative emerges.

Allocating Your Product Strategy Portfolio

In the Reforge Product Strategy course, Fareed and Casey talk about the 5 types of product work:

• Product Market Fit - Going from zero to one on initial product market fit
• Feature Work - Creating and capturing value by extending a product’s functionality and market into incremental and adjacent areas.
• Growth Work - Creating and capturing value by accelerating adoption and usage by the existing market.
• Scaling Work - Focusing on bottlenecks to ensure the team can continue to move forward and take on new levels of feature, growth, and product market fit expansion work.
• Product Market Fit Expansion - Increasing the ceiling on product market fit in a non-incremental way by expanding into an adjacent market, adjacent product, or both.

A core part of product strategy is understanding how to allocate your product bets among those different buckets. Depending on your assessment and level of risk of PMF collapse, you’d likely want to allocate a larger amount of resources towards PMF Expansion or true Product Market Fit work even though the existing product usage data isn’t telling you too.

This Is Just The Beginning

We are still at the beginning of the AI technology shift.

• AI capabilities are progressing at an exponential rate.
• Humans are just getting started creating new experiences with it.
• We have the devices in the world’s hands to run AI which will decrease friction of adoption compared to previous tech shifts.

All these things add up to the Product Market Fit Treadmill speeding up at a minimum, and more “inflection” points at a maximum causing more instances of product market fit collapse.

A few recommendations:

• Check out Reforge Insight Analytics
• Reforge’s Product Strategy and AI Strategy courses to help think through this intense strategic environment.
• Reforge’s Mastering Customer Feedback course to nail the discovery of your changing customer expectations.