This is part 3 of a series of blog posts (here’s part 1 & part 2) about what I’m learning as I hunt for a new startup idea to work on).

PE Rollups Create Opportunity

Private equity is rolling up every industry you can imagine ¹ . Plumbers, electricians, dental offices, primary care offices, machine shops, lawn businesses, RV parks, and software.

PE is brilliant at extracting returns from an unoptimized business. It’s bad at focusing on the customer and building a great product. As industries roll up, prices increase, the quality of service decreases, and opportunities for a new disruptor at the bottom of the market emerge.

Why does this happen?

When PE acquires a business, the original founders lose motivation. This is the rule, not the exception. If founders were still interested in the business, and not just acquisition money, they wouldn’t sell. They’d either run it themselves or find other capital partners. The passion, excitement, and irrational focus on the company that makes it great is hard for any non-founder to have.

This makes PE-owned companies great competitors. They are slow, unmotivated, and focused on numbers and not users.

Pray for a startup idea where your competitors are all owned by PE.

Buyer’s Incentive Matter

At larger organizations, the buyer’s incentives aren’t always rational.

For instance, having a larger team is perceived as better in most large organizations. Employees are considered successful if they "grew the team from 5 headcounts to 20." This is viewed as a good thing since leadership must have given them more resourcing and responsibility because of their work.

If you come in with a solution that decreases the number of employees needed and increases the work the leader needs to do, it makes their life worse, not better. This is an obvious statement– but what isn’t obvious is how you increase your customer’s work! Negotiating with you (a new external vendor), managing a technology implementation, and taking on the work of a human exception handler when your shiny new product goes down is all work they didn’t have to do before.

Organizations closer to the business’s core product act more rationally. Leaders and managers are scrutinized more closely and are less able to shield leaders higher in the management chain from the details of their work. Organizations like finance, HR, etc, are generally seen as cost centers in an organization and not areas to innovate. As long as the cost to run those organizations is reasonable, it’s not worth it for management to rock the boat and create more work for themselves to optimize that part of the organization. If it ain’t broke, don’t fix it. Any second spend cutting costs isn’t spent increasing revenue.

Organizations like these are challenging to sell into because they don’t want more efficiency, better tools, etc even if they say they do. They want to grow their team and avoid improvement because innovation requires change and doing things differently. That takes more work and risk, which is against the culture of these functional areas of an organization.

Minimum Viable Employees

As I started experimenting with LLMs and imagining solutions that could optimize or eliminate existing business workflows, I thought businesses would be excited about eliminating the boring and repetitive work their employees currently execute.

This isn’t universally true.

If a function in an organization is critical to that part of the organization (i.e. invoice processing within a finance team) and requires relatively few people to manage, it’s high risk to replace even part of the team with an AI solution.

For instance, suppose you had three people managing accounts payable in your company, and most of their time was spent transcribing invoices and managing data entry. You’d expect the CFO to be interested in automating that work with AI. However, there are risks associated with replacing part of the team with an AI system:

• If you decrease the team to two and one person quits, you are in a tough spot. Now that other employees can’t go on vacation, get sick, etc. More importantly, you are now the backup on top of your existing duties.
• If the software breaks and you decrease the team size, you put more work on your smaller team to fix the issue and risk burning them out.

In any of these scenarios, you are increasing the workload and stress level of leader who would consider purchasing your solution. They would need to jump in and fill the gap if something went wrong. Not good.

Functions inside an organization have a "minimum viable employee" headcount that they are hesitant to go below, even if it saves money.

Selling AI tools as a cost-saving solution doesn’t work for high-impact, low-headcount work. Enabling your existing team to "do more" works, but there’s also…

Point Solution Fatigue

Non-technical users have been burned by software that promises too much and delivers too little. They’ve seen startups come and go and have experienced the pain of hiring expensive consultants to implement new software only to get a half-baked solution after a year’s worth of pain.

Many users (especially non-technical users in non-tech companies) are default skeptics when considering new software. They don’t want to consider buying it because, most likely, it’ll end up just like the last one. It’s like someone who has been on too many first dates and just doesn’t want to keep trying.

This is partly why the multi-product company thesis is becoming more popular.

Incumbents Can Deploy AI

Last year, I worked on a text-to-SQL app. We did not proceed with it largely because we believed that existing data tools (Helix, PowerBI, Snowflake, etc.) were much better positioned than us to deploy AI copilots for SQL generation.

This has proven true, and I expect it to continue to be the case: a lot of wins in AI will accrue to existing incumbents. They have a data moat, free distribution, and unlike other platform shifts (desktop => mobile, on-prem => cloud) all they need to do is use some nice shiny APIs in their existing web application. It’s trivial to start using AI via the well-designed OpenAI APIs. You can often have a proof of concept up in days, not weeks.

Understanding and predicting your competitors’ velocity and ambition is going to be key. If you are building a platform that didn’t exist before (like Retell), you have a much higher chance of success than building a copilot for a tool that already has a great SaaS tool built for it (like our text-to-SQL failure).

Barbell Strategy for Early Adopters

Although the cost-optimization AI startup has some structural friction, there is an interesting opportunity at the very top where there wasn’t before (early adopters are generally found down the market). A barbell strategy for acquiring early adopters.

Let’s take the invoice processing case I mentioned earlier.

A medium-sized organization may have a handful of people handling this function. A massive organization may have 100s. The cost savings for them—and for the internal leader looking for a promotion—is big enough to be worth pursuing. However, it’s really hard to get these deals.

One interesting way to execute this sort of deal is to find an investment firm that can help broker the relationship with these massive organizations and acquire them as a strategic investor so that incentives are aligned.

On the opposite side of the spectrum, SMBs that do not have any employees and are hesitant to hire any will be very eager to try new AI tools that will eliminate the need to add more folks to their payroll. These are terrible customers but could help bootstrap the product and prove it out so that more risk adverse folks in the middle feel more comfortable giving your product a try.

AI for High Frequency, Low-Value Tasks

Where AI is most disruptive is tasks, which you do often, but if you get them wrong, there’s not much impact.

For instance:

• Tier 1 & 2 customer support
• Bookkeeping transaction categorization
• Marketing copy for long-tail product listings
• Ad experimentation
• Cold emails
• Social media images, tweets, etc
• Common administrative tasks (scheduling, purchasing)
• Inbound phone calls

Ideas where incumbents are slow (PE-owned is a great place to look) and aren’t going to deploy well-designed AI tools across their existing tools (which have a great data and distribution moat) are a great place to look for ideas.

There’s Always Room at the Top

There’s usually room for a premium product in the market. With a deep understanding of the user, you can craft a premium experience and create a new segment of the market.

Some examples:

• Linear
• SuperHuman who would have thought another consumer email client company could survive?
• Notion
• Resend how many email companies have there been?
• TrackSmith

Most folks wouldn’t say issue management is broken ("GitHub issues are just fine!"), but if you are highly opinionated and work with Jira every day, you’ll know it’s terrible and could imagine a better solution.

The trick here is to be a power user of the existing solutions. Allow yourself to get frustrated and slowly build a very opinionated set of beliefs about exactly what is wrong with the current set of solutions and what the future should look like.

Imagine a 10x better solution. A premium product for the super users. If you build that in a large enough market, you’ll pull the rest of the crowd along with you and create a nice business.

There’s Room at the Bottom, Too

I personally think this is less exciting, mostly because it means rebuilding something that has already been done well enough.

But, if you can build something 10x or 100x cheaper (like ONCE or an ETL solution that promises lower costs) you’ll carve out a new segment in the market.

AI expands the area of opportunity here. Building a new business from the ground up with an AI-first mindset enables you to decrease company overhead to deliver an awesome product. Even with a lower cost and smaller TAM, you could still end up with a nice business.

AI Enables New Pricing Models

It’s well-known that you want to price on value, not inputs. This is true in consulting but also in product companies. The easiest proxy we have is the per-seat pricing model.

More people using your product means it’s creating more value.

But, it’s not perfect. Think about Slack. Ideally, you’d want to price on something like "knowledge transferred" or "speed and clarity of resolving a communication gap". Some users send more messages than others, so per message doesn’t make sense, and some users don’t use Slack too much, so per-seat isn’t perfect.

However, pricing models like "$X per successful knowledge transfer" require analyzing an application event and making an intelligent determination about what happened.

This is now possible with AI. My friend Matt has written about this and they are executing on this idea with conversion-based pricing in their startup. They’ll be a set of startups that rethink how to price software in a way that incumbents can’t because of the innovator’s dilemma.

Where am I wrong?

I’ve learned a lot on this journey, and I know there is still a lot to learn! I’m curious to hear your thoughts if you disagree with any aspect of this post. Where am I wrong? What am I missing? Drop a comment below or email me at mike@mikebian.co.

---

¹
We have some private data that we can’t share that formed a lot of this belief, but there is a lot of public data available to support this. From a 2020 report,
private equity "directly employed 11.7 million workers in 2020, an increase from the 8.8 million in 2018". Even in a down year (2023) PE raised $1.2 trillion in fresh capital.