The secret playbook for picking AI winners
Everything you’ve been taught about business strategy is a lie.
Everything you’ve been taught about business strategy is a lie.
The winners in AI are doing the exact opposite of it.
No focus. No target market.
But.
It’s working.
I’m calling it the two-wallet strategy.
It relies on two models and three levers.
Model one.
Build a viral consumer product that teaches the model as a loss leader.
Model two.
Sell the enterprise version that prints money and locks you to lock in margin.
The trick then becomes managing the three levers: reach, unit cost and distribution.
OpenAI basically wrote this playbook.
Companies like ElevenLabs, Synthesia, Runway and others are speed-running it.
I’m Blake. I’ve spent the last 10 years in tech advising companies on data and AI strategy.
I’m here to break down for you the most interesting moves the players are making in AI.
Products. Moats. Distribution tactics. Everything is curated so you can stay informed.
By the end of this video you’ll be able to score any AI company in 15 seconds and predict who will win in their market.
Here’s what you’ll get from this breakdown.
The exact playbook all the key players in AI are employing
Why no one is purely B2C and B2B in AI.
Why you are so important to AI companies
And how to spot an AI winner.
To do it well, a company needs to execute flawlessly across three domains.
Reach, unit cost and distribution.
Step one, get as much reach as possible in the consumer world.
Ship a magical unrivalled experience across the generative domain.
Voice, Text, Video, Image, whatever.
Generate a crazy amount of PR and hype.
Ship like crazy
Whatever needs to be done to get as many people as possible to use the platform.
Use that momentum to:
Build trust, habit and familiarity in low stake environments.
Create a crowdsourced feedback loop to improve the product.
Deliver a bug free enterprise grade offering.
And of course, raise more money
Once in motion, move onto step two, locking in scaled pricing discounts on compute so you can:
Further increase margins
Create an enterprise GTM motion with pricing flexibility
And of course, raise more money.
Once this is complete, move onto step three, improving distribution and ecosystem lock in.
This will allow you to:
Increase product usefulness and switching cost
Scale usage across domains and use cases and,
You guessed it, raise more money.
Then, repeat.
OpenAI was the first mover in this space and basically created this playbook.
How?
Picture this, it’s late 2022.
Everyone’s waiting for another mediocre Apple iPhone update.
Then boom, ChatGPT launches.
It rockets to 100 million users within weeks.
Insane reach.
They got to work improving the product, by making you the product.
See, models learn from feedback.
A public chatbot collects thumbs-up, thumbs-down, edits, and edge cases all day.
That shows the model what a “good” answer looks like, and what breaks it.
Everytime you call ChatGPT stupid, that’s a sign.
Get the collective power of the internet together and all weird bugs will be found in hours.
Fix them once, and the enterprise gets a safer, calmer version.
Makes sense why they bought Statsig, the modern product development platform now doesn’t it.
Genius.
Then OpenAI ships the API.
Then ChatGPT Enterprise.
Suddenly they are a B2B company.
This did not look exciting to everyday users. SSO. SLA’s. Snorefest.
What we missed, is that the enterprise public demo led to an absolutely insane amount of inbound demand from the enterprise. .
Why?
People tried ChatGPT at home.
They trauma-dumped.
They created ghibli characters.
They made AI girlfriends.
After they were done with the fun, they took it to work.
So when the boards and IT departments decided on their AI pilot projects, and got to evaluating vendors, they weren’t hearing a cold pitch.
They had already heard about this AI.
They had already used it. A lot.
In fact, inbound requests were so heavy that OpenAI built an AI to triage the leads.
Yes, an AI to sell AI.
But of course there was another hurdle.
They were unproven in the enterprise.
Ahh, they thought of that too.
Most of OpenAI’s early enterprise wins didn’t arrive from door-to-door selling.
They flowed through Microsoft’s Azure channel.
Roughly four out of five ChatGPT Enterprise customers came that way.
Familiar seller. Faster paperwork.
That usage brought money in the door.
That money let them pre-buy chips, run GPUs hot.
At a tiny volume, the serving probably cost about thirty dollars per million tokens.
At extreme volume with long-term capacity locked in, it landed at nearly five dollars.
Almost 80% cost to serve reduction.
Then they started locking in on distribution.
Sora, Voice mode, Atlas, Sky, Apps SDK, Workflows.
Bringing in Jonny Ive to create a device.
Be useful to everyone, by partnering with everyone, across all domains.
Their latest move, Atlas.
So, if you link 1, 2 and 3 together, you get a flywheel:
Consumers try → feedback improves quality → companies adopt → revenue funds more compute → costs drop → increase the territory → repeat.
Ship fast, reduce costs, improve ecosystem, repeat.
That is the whole point of the two-wallet strategy.
OpenAI is absolutely the poster child.
So, can AI companies win B2B-only?
Yes.
One example is Cohere.
They specialize in large language models and AI products for regulated industries.
Right now, they generate roughly $100 million in annual revenue and have strong margins from private deployments.
Tight privacy with on-premise, white-glove enterprise support.
The path they’ve chosen trades speed and mindshare for control and depth.
The catch… fewer organic leads, slower cycles, and less public feedback.
You must be exceptional at the hard, boring stuff in the niche, or bigger brands will out-muscle you.
So now you may ask, what about B2C-only?
Also yes.
Midjourney shows it with about $500 million in revenue and a tiny team.
No API. No enterprise push.
It works because prosumers pay and the art is stunning.
Seriously, go try it.
Buuuuut. There’s a ceiling.
No big contracts, no platform lock-in, and limited leverage on chip deals.
Once rivals get “good enough,” businesses choose the tools with APIs, SLAs, and procurement boxes already ticked.
They’ve won the battle, but I can guarantee it’s going to be hard for them to win the war.
There’s a tricky conundrum.
If a consumer app does everything, why would anyone build on your API?
Or, if you only ship an API, how do you learn what people actually want?
That’s why the companies that do both have a higher chance of winning.
It’s a really big gamble to only focus on one side.
I think Cohere and Midjourney are outliers.
Let’s see what happens in a year.
So who else has run the two wallet strategy?
If you’re still not convinced yet as to why, let’s go to my absolute favourite AI company.
Elevenlabs. The most realistic voice AI platform.
They began with ultra-realistic AI voice text to speech technology. Today, they advance AI audio across speech, sound effects, and music and turn that work into products people use every day.
They started with creators.
They hit a million users in five months.
Domain one, Reach.
Voiceovers everywhere.
That visibility pulled in developers.
Developers pulled in companies.
They now offer teams the ability to deploy voice agents that listen, talk, and act, localize content at scale, tell stories, and improve accessibility.
They offer all the unsexy stuff through the enterprise offering.
“zero-retention” modes, HIPAA add-ons, and private deployment options.
Enterprise data stays out of training by default.
Then scale kicks in.
As usage exploded, Elevenlabs cut generation costs about 50% and passed savings on—down to roughly $15 per million characters at high volume.
Domain two, Margins.
At the same time they learn fast from the consumer side where users opt in.
So they improve weekly without touching a customer’s private calls.
That’s how you turn a viral tool into an enterprise platform without losing money on every clip.
They’re now on the cusp of step 3, distribution.
They’re expanding their ecosystem across the agent experience, working in partnership with many established players in the contact center and voice space.
It’s early, but I’m bullish.
I think every player in AI will eventually adopt this playbook.
If they’re B2C, they will add business plans.
If they’re B2B, they will launch a small consumer app to learn faster and build a brand.
The gravity pulls both ways.
Translation.
There will be no pure B2C or B2B AI Company.
Midjourney will get there.
Cohere will also get there.
AI moves too quickly to not.
To spot an AI winner, find out the following
Where does the reach come from?
Do they have a solid unit cost baseline?
Are they building in distribution?
If one company can answer all three, you’ve probably found a compounder.
I call it the two wallet strategy.
Hopefully this was useful.
I’m Blake.
Thanks for watching.