In this episode of McKinsey on Building Products, McKinsey partner Rikki Singh speaks with Paul Yacoubian, co-founder and CEO of Copy.ai. They explore how an AI-native company such as Copy.ai leverages product-led growth (PLG), the pivotal role of analytics in the company’s customer outreach strategy, and how start-ups can scale effectively without overexpanding. An edited version of their conversation follows.
Background of Copy.ai
Rikki Singh: Tell me a bit about your background and the journey at Copy.ai.
Paul Yacoubian: I’ve had several roles in my career. I started as a CPA at a public accounting firm and then went into the hedge fund industry, where I invested in SaaS [software-as-a-service] companies over a decade ago. After that, I joined a family business that was a tech start-up and helped run, grow, and sell that company. Then I was in venture capital, which I left in 2020 to start Copy.ai. I have learned a lot.
Rikki Singh: What is the product offering at Copy.ai?
Paul Yacoubian: We’re a go-to-market AI platform. We started the business in 2020 and focused on delivering generative AI use cases to companies around the world. We started with marketing and sales use cases and content generation use cases. Now, we’ve scaled to include workflow-based automation, and we’ve been fortunate enough to work with some of the best companies.
Rikki Singh: Who would you define as your target customer?
Paul Yacoubian: Typically, it is chief revenue officers or chief marketing officers in mostly B2B companies.
Rikki Singh: How have your past experiences shaped the way you think about PLG?
Paul Yacoubian: Throughout my career, I’ve seen more go-to-market motions and strategies and different ways of executing than most people. Through that, I’ve learned to recognize what works in different situations and how to structure teams to grow the fastest. From a PLG standpoint, it’s important to figure out how much friction you can eliminate through the product.
For example, take a company like ServiceNow: it has sold to Global 2000 companies. The things that worked well for them are the land-and-expand motion and SKU-based development, which they’ve built on one platform and continue to add more value to their customers over time.
In the generative AI world, data powers all of these use cases. Most companies have some internal data, but a lot of times, especially for go-to-market use cases, they don’t have good data. For example, 50 different vendors could claim to have the data they need, but buying or procuring that data, sizing up those contracts correctly, and doing all the spend and commits can be a big challenge. We use a bundling approach to take the best parts of each of these different business models to decrease customer friction and add value a lot faster for our customers.
PLG for AI-native products
Rikki Singh: How do the goals of reducing friction and driving sustainable growth shape Copy.ai’s PLG approach?
Paul Yacoubian: When we first started, we didn’t know what the use cases would be. We knew that at the core, large language models could generate words. So we had a simple thesis that asked, “Who buys words today?” Marketers buy words. And what is the format, shape, and structure of the words they buy? People are buying blog posts, social media content, and cold-email content. We took that information and figured out how to replicate the best practices for each of those use cases with generative AI and then give the user a way to input the information they already have.
Rikki Singh: What were the two or three use cases that you homed in on as you were thinking through the PLG journey?
Paul Yacoubian: For most PLG motions, onboarding is huge. How do you get the new user to an activation moment as fast as possible and in a way that delivers longer-term retention? That’s always critical in terms of getting feedback when you have lots of users.
For us, a big component of PLG was running the net promoter score survey. We sent an email follow up and asked, on a scale of one to ten, “How likely are you to recommend our product to a friend?” That gave us a baseline. When you have millions of users, you’re going to get enough of those responses quickly, so you can start to iterate and track the net promoter score in real time.
The second piece of that is qualitative feedback. I highly recommend asking, “What was the most friction that you have experienced with our product?” or “What did you hate the most about the product?” When you ask the question, you make customers feel comfortable enough to give you feedback.
One founder or engineer tendency is to add more features, but when you look at the feedback, nobody is asking for any of that. It’s more important to assess whether your product is easy to use and gives customers what they need from it.
You can also see the different types of feedback that come in. You can bundle all the quality feedback together and solve for certain problems to see if that improves your net promoter score. For example, for one of our first applications of generative AI, we had a massive tailwind because it was a free product, and no one had ever used it before. We got a lot of user growth just through word of mouth.
Rikki Singh: How did you know that you were getting word-of-mouth growth? Was that part what you were tracking?
Paul Yacoubian: Two ways: first, we have a question in our onboarding survey, “How’d you hear about us,” which is helpful. Not everybody takes the survey, but a certain percentage will fill it out, and over time, you can track how people found us across different marketing channels. Second, when we launched the product, we launched it on X [formerly Twitter], and we went from 45 users on the first day we launched to 2,000 users in 48 hours. One week later, we still had 350 people signing up a day, and we weren’t doing any other marketing.
When you have a freemium product, out of 100 people, two might be B2B buyers. Everyone else might be students, freelancers, small business owners, or people who don’t have budgets to pay for a product. Products that are more end-user focused, such as project management tools, start with a basic offering and then iterate to create a product that can be monetized once they accumulate five active users. We had to know when to trigger that.
Rikki Singh: Is the freemium conversion percent that you were tracking a way to know whether you’re trending healthy versus not? Because, to your point, costs have to be covered one way or another.
Paul Yacoubian: We monitored it, but the North Star for us was total ARR [annual recurring revenue]. We knew that ARR would be driven by enterprise and that we wouldn’t be self-serve in the long term. The initial self-serve business gets you in business, helps you build a team, and helps you develop a product that’s going to drive more value. Pure PLG companies, such as Canva or Grammarly, track their demand generation and new user sign-ups regularly. The conversion rate is an output of the trade-offs that they’re making between usage and getting paid—and getting paid now versus later.
Most mid-sized PLG companies have an entire pod of people who take care of pricing, packaging work, and experimentation. If PLG is the long term driver of the go-to-market strategy, then you can experiment as fast as possible to identify the core metrics that indicate your business health—what drives customer value, what moves the needle on your core product metrics, and which of those metrics drive your ARR.
The evolution to sales-led motion
Rikki Singh: Talk a bit about your monetization approach and packaging strategy. How has it evolved over time?
Paul Yacoubian: With a traditional freemium model, companies traditionally monetize 1 percent to 5 percent of their user base. If the customers are solo users, the percentage of monetization is going to be closer to 1 percent, whereas if the customer is a team and everyone has to upgrade at the same time, it will be closer to 5 percent. Originally, we had no idea how much the usage was going to cost.
In the enterprise world, you have more compound start-up–type products, such as Rippling or Ramp, that start with three use cases. They get customers onboarded, and then they can roll out more use cases and gain more traction, which is a nice way to roll out products. It also helps the customers because they’re not over-buying on the front-end to get any value from the platform.
We like the packaging approach where customers know exactly what they’re getting. To do that sales-led motion, you have to have a very good understanding of where the biggest need and biggest pain are going to be for certain customers and which customers are going to be a good fit for certain products in the longer term.
Rikki Singh: As you transition from the self-serve PLG motion to being sales-led, do you continue to collect feedback from your B2B users and use that feedback to inform your strategy?
Paul Yacoubian: Yes. Product analytics is huge for us, especially around usage. We want to be able to deliver repeatable value that’s usage-driven. That’s going to be the differentiator between platforms and more tactical point-solution tooling. Working with customers hand in hand is the best way to get that feedback.
Rikki Singh: What are the must-have organizational enablers to adopt PLG at a start-up? Which enablers help you scale?
Paul Yacoubian: Start-ups need an experimentation mindset—they need to be willing to be wrong and consciously run hypotheses. To properly do PLG, companies should try to build a quantifiable system, in which they can build a funnel to take in new users and get them to that activation moment without friction. It’s best to eliminate as many steps as possible for the user. A lot of PLG teams with experimentation frames have growth teams that are led by either a growth product manager [PM] or a growth engineer.
At Copy.ai, we have not had a designated growth PM, but we’ve had growth PM hats on at different points, as well as pricing and packaging hats. Right now, it’s all about enterprise for us and up-market consumers. Take a Fortune 500 company, for example: how can we deliver $10 million in value to them? PLG alone may not get us there, because a lot of times, you need to have a strategic conversation to understand which workflows are going to be valuable to them. It’s also important to determine which approach you’ll use for building and deployment because there are trade-offs for both sales-led and PLG, and customers should know how they could get value from different platforms.
Rikki Singh: What have you learned about customer segmentation along this journey?
Paul Yacoubian: Customers often have the same problems. For us, once we know the use case, it’s reusable across companies. What you want to optimize for is the difference between personas. If you can minimize the difference, then you can optimize in different directions. For PLG companies, two personas matter most. First are product managers because they’re communicating cross-functionally, so if you can land a product manager, you’re more likely to land the marketing team, sales team, support team, and all the other teams around them.
Second is the marketing persona, which typically looks for project management tools. For marketers to get any content created or any campaign off the ground, they have to interact with many people and move through many steps, so they need a way to manage all of that. That’s another PLG growth loop that is powerful. Once you identify what the persona needs, you can add more functionality to make the product more valuable.
Scaling businesses in the age of AI
Rikki Singh: What advice do you have for someone who is scaling businesses around PLG?
Paul Yacoubian: Keep an open mind. Also, know yourself. If you want to build with a certain vision in mind, go all out on achieving that. There are two different types of building. One is building something small, iterating on it, seeing who cares, and then going from there. The other is building something specific based on an opinion of how the world should work and what the future should look like. You’ll move between both of those frames of mind, but always feel comfortable having an opinion.
When companies start out, everybody wants to position themselves wide and big. They want to build all the agents for every use case. That’s great, but how do they execute that goal? If they end up building AI capabilities too vertically or horizontally, they’ll never escape the original use case they build.
So think about what’s motivating you to build the product and platform and what your vision is for where you want to take the product. Don’t think about where the ARR is today or about giving your customers products they didn’t ask for. Those things don’t matter right away.
Rikki Singh: There are a few things I took away from our conversation. One is this obsession with the customer, removing friction and getting them to value as quickly as possible. The second is to not go too vertical or too horizontal when you’re building products. Being too general will bury you among the competition and being too vertical will limit your total addressable market. Find the right middle ground to ensure you have the scope to grow and are solving the right customer problems.
Paul Yacoubian: There is an art to it. The founders of Stripe have talked about it well: carve out something narrow enough to get your real market share and then expand in concentric circles around it.