AI economics through 4 pricing models 🧱🏷️
4 pricing iterations, taking a multi-model approach, designing AI-enabled free trials, and building where foundation models fail — inside Sudowrite's journey.
“We are at our fourth attempt at a pricing model, and it’s still not right.” — That’s Amit Gupta, co-founder of Sudowrite, a widely-loved AI writing tool for fiction writers.
A 4th pricing attempt that is still not right isn’t a flaw in strategy or a confession but an honest take at the intricate craft of pricing anything AI: constantly evolving, and constantly required to evolve.
Amit is a serial founder, who had earlier built and sold Photojojo, after scaling it to $10M/year. He is also a Sci-Fi writer. Amit is now building Sudowrite with fellow entrepreneur and writer James Yu (YC S11 and founder of Parse).
Sudowrite was a very early entrant in AI, coinciding in birth with ChatGPT’s Model 3 back in 2020. Since then, Sudowrite has managed to garner wide-spread love from users, including prolific writers such as:
This edition of PMF /evals will feature notes from our chat with Amit where we peel back the layers on:
👀 the founding insight behind Sudowrite and why Amit chose fiction writing as a problem to take on,
🧱 the thoughtful tech decisions that enable their incredibly high quality of output,
🏷️ the nuanced evolution of their pricing model going from flat to credits spanning 4 different, in-time iterations,
🚴 how they designed an AI-enabled free trial that doesn’t burn a hole in their inference pocket, and more.
Read along and if someone forwarded you this newsletter, don’t forget to subscribe here.
#1 ➤
Building where foundation models fail to deliver
Amit recalls ChatGPT’s model 3 launch in 2020 and the frenzy it birthed, with the world flocking to build wrapper assistants for business use cases like marketing copy. Amit and team saw it but decided to take on a different, much taller challenge: building an AI assistant for fiction writing.
There were a couple of reasons behind this. One, Amit found the challenge of building a copy assistant entirely uninteresting. Two, Amit knew that the simplicity of the problem was going to inevitably attract a lot of competition.
Fiction was different.
It was both personally inspiring for Amit as well as complex enough to warrant building a thoughtful solution. “With fiction, there’s a large gradation of subjective quality. So we felt that there was more interesting work to be done here. You can’t ever get it right.” Amit shares.
But generative writing is a use case that foundation models seem to cover really well. So we had to ask: “Why do we need another tool? Why aren’t the foundation models sufficient?”
As a writer himself, Amit had spent a lot of time tinkering with the foundation models and had often walked away feeling both underwhelmed by its capabilities as well as frustratingly confined by the constraints.
First he ran into out-of-touch censoring. Fiction stories often involve mature themes that model makers consider taboo and censor.
“We quickly realized that this was going to be a huge problem for writers. They are used to writing in MS word, and MS never censors them.” Amit adds.
Then he ran into what could only be described as bland creativity.
Most of the foundation models are good at writing memos and emails, but they’re not good at writing dramatic narratives.”
Writers thrive off of reaching for a personal tempo, the best that foundation models could offer still felt way off compared to what writers wanted.
Sudowrite has a set of deep features designed to tackle this exact swath of problems: it allows writers to go as far as defining tone, setting up characters, choosing models of their choice for different steps, and feeding their own understanding of different writing elements into the tool. Foundation tools don’t offer any of these capabilities.
#2 ➤
A “non-generic” assistant: The tech decisions enabling Sudowrite’s high-quality output
Getting to this level of quality has been far from simple. Today, Sudowrite deploys a two-layered system, comprising a prompting layer and a model layer, with dedicated efforts in improving and innovating on both.
On the model layer, Sudowrite uses a few dozen models, all of which have been thoroughly evaluated for their specific edges/advantages and then brought in.
“Some models are better at extracting salient information from a user’s story bible, which is an area of Sudowrite that gathers the core elements of your story and acts as a source of truth as your work develops. Others are better at what we call the tip of the pen, which is actually writing the first draft prose as the final step. Each step along the pipeline, from a user input to a final version that the user likes, calls on multiple different, specialised models.”
This model-scape is how Sudowrite is able to deliver on quality.
In addition to these pre-selected models, they also allow users to choose models of their own preference for different steps of the process, something that Sudowrite’s users seem to have a strong interest in: “People who are just getting into writing with AI can’t be bothered to learn about different models, but those who’ve been working with this stuff for a year or longer often have favorite models.” Amit adds.
Amit and team are also doing a lot of the work in allowing authors to train models specific to their voice. They recently opened a beta program where authors could upload work to train a model just for their own use and writers have been lapping it up.
These user insights have reaffirmed Sudowrite’s original premise that fiction writers want non-generic assistants, going as far as wanting a model that can be trained to assist them for every step of their unique writing process.
#3 ➤
From flat fees to credits: Evolving through 4 pricing models
“This is the first time in SaaS that you also have real marginal cost to deliver the final product if you use AI. The era of zero marginal cost SaaS might be over.”
Amit opens his notes on pricing by acknowledging the elephant in the room, AI has changed the whole script of pricing software.
Today, Amit and team are on their fourth attempt at a pricing model, and he believes it’s still not right.
They started with a flat, $5/month — which was later changed to $10, then $20 — with no specific stated limit on usage. It soon started showing cracks, with users paying $20 a month while costing them $400 a month.
That’s when they decided to switch to AI words in an effort to find a way to correlate the value users were getting from the product with what they were paying for it. “It was crude because words aren’t really ‘value’, but it was as close as we could get in a meaningful way.” Amit shared.
Soon enough, as Sudowrite went multi-model, they had to switch again, this time from words to credits.
“With the level of flexibility that came with offering different language models, and the fact that we let users develop their own custom tools in Sudowrite, using whatever prompts and LLMs they want, it meant that someone could use the most expensive models at each step, inhaling multiple novels as input, and only output a single word, one word that cost us $30 to make. That was not sustainable. We couldn’t charge by words as an output metric.”
This usage-credits based model was their attempt to further align their costs with what they were charging customers. It also allowed them to offer greater flexibility and customization, as well as lower costs for the vast majority of their users. But it came with its own challenge: it was confusing.
The way Sudowrite is built, they charge different amounts of credits based on different models you use at different steps along the way, which meant that it was really hard to predict exactly how much pressing a button would cost you.
To help with the confusion, they decided to offer a one-week easy cancel policy. In addition, they also run coaching and education efforts to make sure that users are able to understand their credits and their implications as well as possible.
One useful parallel that Amit goes to is the apparel industry. Where, again, choosing the right size (plan) of t-shirt solely from photos is quite difficult.
“Ultimately, the solution that worked is that you can buy the t-shirt and you can exchange it if it’s not the right size. I think that’s the kind of solution where we’ve landed too. You can return or exchange any plan in a week if you buy it and it doesn’t fit.” Amit added.
Note: Check out this LinkedIn post by Chargebee’s CS leader Madhuri detailing the strategies that Fin (by Intercom) uses to ease difficulties around their outcome-based model priced on AI resolutions.
#4 ➤
Designing AI-enabled free trials that don’t burn a hole in your inference pocket
“All of our plans include free trials. We have had a similar journey of iterations with our free trials as we did with our overarching pricing models.”
Amit and team started with a flat 2-week trial, and then switched to a AI-words-first trial as they ran into the familiar set of challenges: of finding a balance between value for the customer and cost to the business.
Eventually they switched to a credits-based free trial where users get a limited number of credits to use in their trial.
Amit’s philosophy behind this was clear: They wanted to give people a taste of the product and time to see if it’s right for them but they also had to be mindful of the cost of servicing these trials. Amit shared that at one point a huge percentage of their overall inference costs was going towards servicing trials.
“When you’re writing a novel, each inference call might use 10s of thousands or even 100s of thousands of tokens of input for context. These are very real numbers.” Amit adds.
So they decided to cap the trials with a credits limit and make them shorter.
Interestingly enough, they saw that shorter free trials converted better. Amit’s hunch was that it created a sense of urgency and their ideal customer, professional authors, were serious about making a buying decision.
Amit and team have also made a deliberate effort to invest into education to ensure that users are enabled in their exploration and the business sees better returns from their free trial.
“We’ve taught classes for a few years now; we teach at least one class every day and we have different topics throughout the week. It’s a good way for people who are still learning the program to get up to speed without having a long trial potentially.”
They have also started experimenting with one-on-one onboarding where users can schedule a time with them and actually learn how to use the program with an expert over 20 or 30 minutes and then get more out of the trial.
Amit believes that creative tools can get complex, and educating people on how to use them and how to get the most out of them will become increasingly important.
#5 ➤
“Not for everyone”: The further-ordained position of pricing in finding PMF
Amit thinks of pricing as a fundamental lever of their product-market fit — as much about defining the ideal customer as it is about staying approachable.
Last year, they did a deep dive to understand who gets the most value from Sudowrite over time. They discovered that their core users were professional authors, people who’ve written more than one book and essentially earn their living through writing.
For these writers, Sudowrite was an easy purchase.
“It could probably be three times more expensive and still be valuable because it helps them write much faster and produce better books.” Amit found.
On the flip side, they saw that the cohort of aspiring writers and students considered $10-20 a month prohibitively expensive. This gap meant that they needed to pick a lane, there was no way to have a price point that could serve everyone.
It also gave them more clarity about where their PMF was going to come from, which then they doubled down on: “We’re focusing on being the highest quality solution in our segment, which means we can’t serve everyone.”
#6 ➤
Why “everyone will vibe-code their own software” might be an exaggeration
There has been much too much chatter about how vibe-coding could eat away SaaS or that people might start building their own software. If you’ve been on X at all, you know.
Amit has a more tempered take on this. He believes that as software becomes easier to build with AI, people will make more software for themselves. But the growing notion of personalized, bespoke software might not be as big as people think. He draws an analogy to help make sense of why:
“There was a period where personal publishing was really big and everyone published their own blog. Then it became easier to tweet and post on Facebook, and because it was easier, that’s what people did. Most people didn’t want to do the work of setting up a blog and trying to get readers even though it’s trivially easy.”
He thinks the same will be true of software for most people…
“If a solution exists and it’s good enough, they’ll prefer to use that than to create their own software and make the hundreds of decisions that go into making a piece of software.”
That said, Amit shares that there will be tremendous potential in allowing people to customize their software.
Someone could very well be like: “Oh, I hate how my video conferencing software works. I wish it had feature X. I’m not going to build a whole new Google Meet myself just to get X. But if I could prompt it and add that feature easily, I might customize it that way.” Amit adds.
P.S. A big thanks to Amit for taking the time to talk to us. And all the best to the Sudowrite team for what’s ahead.
PMF /evals ◎ is just getting started. Tell us about how you’re approaching AI-native building in the wild! Or what you’d like us to cover. Hit reply.