AI has the same problem, at ten times the clock speed.

By the end of this read, you will have a working financial model that maps your full token cost stack, isolates hidden consumption, and tells you whether your pricing survives a provider rate shock before the market does.

The free deep dive told you what to model. This one hands you the model.

If you have not read “The New Math of AI,” start there. It covers the four paths, the token iceberg, and why the fastest-growing AI startups are running at 25% gross margins while the healthier cohort operates closer to 60%. This post assumes you understand the problem. It is here to solve it.

What follows is a walk-through of the Token Waterfall Model: a six-sheet working financial model built specifically for AI operators who need to price a product before launch, not discover their cost structure after. Every number in this article maps directly to a cell in the model. Open both. Work through them together.

What This Model Does That Generic SaaS Models Do Not

Every financial model template on the market was built for SaaS. Revenue minus COGS, COGS being hosting and support, gross margin being 80%. That math does not apply here.

The Token Waterfall Model is built around four realities that are specific to AI products:

Token costs are not fixed. They vary by user behavior, task complexity, model version, and provider pricing decisions. A model that assumes a flat cost-per-user will be wrong by 2x to 10x depending on your product.

Hidden token consumption is the primary cost driver. Internal reasoning chains, agent loops, retry logic, and tool calls often represent 50-90% of total token spend. Almost no standard model accounts for this.

Provider rate changes are a strategic risk, not a line item. Pricing tier restructuring by major AI providers has in some documented cases resulted in materially higher infrastructure costs for enterprise clients within a single contract period. Your model needs a shock test built in before you need it.

Power users destroy average-based pricing. The math that works for your median user can be catastrophic when applied to your top 5%. You need to know which segments are profitable and which are subsidized before you set a single price.

The Token Waterfall Model addresses all four. Here is how each sheet works.

Sheet 1: The Token Flow Calculator Where every pricing model should start. Almost none do.

Before you can price anything, you need to know what a single transaction actually costs in tokens. Not what the user prompt costs. Not what the response costs. Everything.

The Token Flow Calculator maps your product’s full token consumption across two layers.

Read more

I have spent nearly two decades building commercial systems where pricing and cost structure were not finance problems, they were survival problems. At Aurora, I ran strategic growth and partnerships in a market where mismodeling the cost of a financing product by even a few basis points could destroy a deal. That same pattern is playing out right now in AI, at a much faster clock speed.

By the end of this read, you will understand what your AI product actually costs to deliver, why that number is probably two to five times what you think, and how to build the financial model that closes the gap.

Every company building with AI right now is on one of four paths.

Which one you are on determines your cost structure, your margin profile, and your strategic exposure. Most companies are on path one. Most have no idea what path one actually costs them.

The fastest-growing AI startups are running at 25% gross margins. Sometimes negative. The healthier cohort operates closer to 60%. Traditional SaaS sits at 80-90%. Bessemer Venture Partners documented this spread in their State of AI 2025 report [1]. The cause is physics: every inference call has a price. Every generated sentence incurs real compute cost that traditional SaaS never faced.

Here is what each path actually means.

1. Path one: use an existing API. Fast, flexible, and someone else’s infrastructure decision. Also someone else’s pricing decision, their rate hike, and increasingly their competing product built on the same stack. One important note: path one on frontier models carries rising cost exposure. GPT-4o and Claude 3.5 class models cost more per token than their predecessors in many use cases, particularly for long-context and reasoning-heavy tasks. The inference cost curve is falling for older models. It is not falling uniformly across the stack.

2. Path two: run open source models on your own infrastructure. Rent the compute, own the weights. Lower margin risk than API dependence, lower capital than a proprietary build. The fastest-growing path among sophisticated builders right now, for exactly the margin reasons described in this piece.

3. Path three: customize a foundation model on your own data. More control, more cost, and a longer runway to production.

4. Path four: build proprietary from scratch. A serious 30-billion parameter model starts at $5-10 million in compute alone. A frontier-class model costs north of $100 million to train. Billion-dollar runs are coming before the end of 2026.

Most companies are on path one. Most have not yet modeled what path one actually costs them per transaction, fully loaded, including every token their product consumes that a user never sees.

That is where the margin goes.

The Token Is the New Unit of Production

In traditional SaaS, the marginal cost of serving one more user approaches zero. In AI, it does not.

A token is roughly 0.75 words or 4 characters. A typical user interaction costs 200-300 tokens on the surface. The real number is often ten times that once you account for the full system. This is the Token Iceberg, and most operators only price for the visible tip.

Before you set a single price, answer one question: what is your actual token flow? That means every system prompt, every reasoning step, every agent loop and retry your product runs in the background. In agentic AI products, internal token consumption is 50-90% of total usage, the cost operators most often fail to model.

The breakdown matters more than the total. A product with 800 tokens of visible consumption and 7,200 tokens of internal consumption does not have a usage problem. It has a pricing problem, because the cost structure is almost certainly invisible to the team setting the price.

Make sure the Math is Math-ing.

Work through your own numbers with the Token Flow Calculator at kylekelly.co/tools/token-calculator.

What Surviving Companies Did Differently

Three pricing models have emerged. Each represents a different trade-off between vendor cost predictability and customer value alignment.

As you move from consumption to outcome-based pricing, you accept more cost risk in exchange for tighter alignment with customer value. Most early-stage AI companies default to consumption because it is the easiest to implement. It is also the easiest for customers to abandon when spend becomes unpredictable.

The most defensible commercial position is outcome-based pricing, but only if you have fully loaded your cost model first. Intercom charges $0.99 per resolved support ticket. That model works because they built the cost structure before setting the price, not after. Build the model wrong and outcome pricing becomes a liability, not an advantage.

Subscribe now

The Credit Is an Abstraction Layer, Not a Cop-Out

Credits exist because the true unit of AI production (the token) is too granular for customers to reason about. A credit is a pricing wrapper that converts technical consumption into commercial clarity. But credits solve a communication problem, not a financial one. Your internal cost structure still runs on tokens.

Here is how three representative products have structured it:

The Manus Standard tier reveals the gap most clearly. At 0.5 cents per credit, with a cost-to-serve closer to 0.85 cents, they are selling below cost at current pricing. That is a deliberate land-and-expand play. It only works if two things happen before the runway runs out: engagement deepens enough to justify a price increase, and volume grows enough to negotiate better infrastructure rates. Most companies running this playbook will not hit both.

Cursor’s margin profile is contested. Multiple credible sources reported negative gross margins in mid-2025, with significant API costs against subscription revenue. Their trajectory is a useful signal regardless of the exact number: even products with strong PMF and rapid growth can face structural margin pressure when token consumption scales faster than pricing catches up.

The Four Numbers That Determine Whether You Make It

A rigorous operator does not set price from intuition. They build a model that converts token costs into commercial decisions. There are four layers to that model, and most operators only think about one.

Layer 1: Revenue Architecture

The revenue line is more complex than it looks. Subscription revenue anchors too low in almost every product I have analyzed, because operators underestimate how much credit consumption scales with engagement. The operators getting this right are building four revenue streams, not one: subscription, overage, tier upsells, and enterprise contracts. Each has a different margin profile. Each needs to be modeled separately before you set a blended price.

Layer 2: Cost of Goods Sold

Three line items most operators miss entirely.

Inference costs include every token type your product consumes, not just the visible input and output. Internal reasoning steps, agent loops, and retry logic are all billed at the same rate as user-facing tokens. If you have not instrumented this, you do not know your COGS.

Infrastructure overhead runs 10-15% of inference cost annually across a typical AI product stack: hosting, monitoring, logging, security. This is not a rounding error at scale.

Third-party API costs are the line item teams most often forget. Every search call, vector database query, or external tool call in your agentic workflow has a cost. Even $0.002 per transaction compounds significantly at 200,000 monthly transactions.

Your gross margin is the gap between what you charge per credit and what every token in that credit actually costs you to produce. That gap is almost always smaller than operators expect.

Layer 3: The Gross Margin Reality

The 60% gross margin threshold matters because it is the level at which LTV:CAC math becomes defensible for most institutional investors. It is a target, not a gate. The Bessemer Supernova cohort runs at 25% and still attracts capital because growth rate can temporarily substitute for margin in early stages. [2] The question is not whether you are at 60% today. It is whether you have a credible path to it before your runway forces the conversation.

If you are selling credits at $0.005 and your cost-to-serve is $0.0085, your gross margin is negative before you pay for a single sales call or engineer. Model the full waterfall before you set a price.

Better pricing buys you runway. It does not fix a broken cost structure at the model layer. That requires volume, routing, or your own model. But most operators are not close to that problem yet. They are still losing margin to costs they have not measured.

Layer 4: The Questions That Expose Your Real Risk

Most pricing conversations focus on what to charge. These four questions focus on what you are not accounting for. Operators who answer them before launch live in a different risk category than those who discover them after.

What is the fully loaded cost of your most expensive task type, including abnormal reasoning chains? Most cost models are built on average usage. Power users and edge cases destroy average-based pricing.

Do you currently lose money on your heaviest users, and do you know who they are? If you do not have this number, you do not have a cost model. You have a guess.

What happens to your gross margin if your model provider raises API costs 20%? This is not a hypothetical. Anthropic introduced Priority Service Tiers in 2025 that doubled infrastructure costs for some enterprise clients overnight. If your pricing does not survive a 20% cost shock, you have a dependency, not a strategy. This is also where open source infrastructure changes the math permanently.

What external APIs or agents are embedded in your standard workflow that you are not yet billing against? Every third-party tool call in your product stack has a cost. Most operators only discover this at scale.

Where This Is Going

The companies sustaining 60% gross margins share one behavior: they modeled the full cost stack before pricing, not after. The operators who survive will not do so by building better models. They will do so by building better pricing systems.

The token is the ingredient. The outcome is the product. Price accordingly.

References

1. State of AI, Bessemer Venture Partners, Sep 2025

2. The Gross Margin Debate in AI, Tanay Jaipuria, Sep 2025

3. Cursor Pricing, 2025

4. Manus Pricing, 2025

5. Perplexity Pricing, 2025

6. GPT-4 Training Cost, Sam Altman / Stanford AI Index, 2024

7. Frontier Model Cost Projections, Dario Amodei / Fortune, Apr 2024

8. Inference and Compute Costs, Scale AI Foundation Model Report, Aug 2025

9. Custom LLM Build Costs 2025, Follow.life / Epoch AI Analysis, 2025

Read more

Read more

I spent over a decade doing partnerships of every kind. Product partnerships are my favorite. They’re technical, challenging, very strategic, and when executed right, they drive customer acquisition, brand elevation, r…

Read more

Meta is becoming a power trader. Microsoft is restarting nuclear reactors. Amazon is building its own energy infrastructure. These aren’t energy side projects. They’re GTM plays.

Read more

Read more

Two weeks ago, Shopify reported Q3 results that deserve a closer look. When a company re-accelerates growth to 32% while posting 18% free cash flow margins, expands its addressable market 18-fold through product innovat…

Read more

Read more

Apple’s partnership with Google’s Gemini isn’t a headline about AI. It’s a masterclass in sequencing strategy.

When a company this dominant changes its build-buy-partner calculus, it signals a platform shift big enough t…

Read more

Cohort analysis is supposed to be descriptive, not predictive.
But if you’re the kind of operator who looks for signal in the noise, hear me out.

When you segment by behavior instead of time, patterns emerge early.
You sta…

Read more

You all know I’m obsessed with how partnerships actually create value.
Not the logos, co-marketing, or handshakes, but the systems that turn collaboration into compounding revenue.

I’ve led teams that built partner ecosys…

Read more

AI protocols are still evolving, but the early signs are clear: standardization is becoming strategy. Whether these layers converge or fragment, they’re already changing how software products distribute and compete.

Read more

This week, I analyzed how ChatGPT, Midjourney, and Loom reached massive scale without traditional marketing budgets. While most companies still treat product and distribution as separate functions, I’ve been thinking about what…

Read more

This week, I analyzed Intuit’s recent Investor Day presentation. While companies announce new AI releases every week, I’ve been thinking about what separates announcements from execution. Everyone talks about copilots and feat…

Read more

This week, I’ve been thinking about distribution, not as a marketing function but as the ultimate power layer of AI. Every company talks about models, but the real game is reach, control, and compounding access. I’ve watched this same…

Read more

Subscribe now

Leadership frameworks aren’t invented in boardrooms. I learned mine at 5 AM on the Schuylkill River.

While my classmates slept, I was learning how rhythm, feedback, and shared struggle compound into performance. Rowing taught me what every operator eventually realizes: what moves a boat forward under pressure is the same discipline that drives organizations through growth and chaos.

Here are the four rules every operator should know.

My Leadership Manifesto:

1. Operating Rhythm > Coordination

2. Feedback in the Moment Scales Faster

3. Shared Struggle Builds Culture

4. Adaptability Is the Real Advantage

1. Operating Rhythm > Coordination

There’s a difference between eight rowers moving at the same time and eight rowers moving as one. The first time it clicked, the shell lifted out of the water. Effort turned effortless.

Our coach called it collective rhythm. Synchronization is just hitting deadlines together. Rhythm is a shared flow where output compounds beyond any one person.

Operator lesson: The best teams don’t just align on goals; they move in rhythm. At Zappos, a seven-person team in flow outperformed groups of twenty-five operating in coordination. In deals or product launches, rhythm is when finance, product, and sales stop defending silos and start pulling together.

2. Feedback in the Moment Scales Faster

“Hey 4-seat, watch the blade sky! Hands level at the catch.”

I was that 4-seat that day. One quick cue from our coxswain and within a few strokes the boat was flying again.

Our U.S. National Team and Collegiate Hall of Fame coach had one rule: he called an issue once, just once. It wasn’t harshness. It was preparation for race conditions where real-time correction determines whether you medal or miss the podium. It was about fixing it while you can impact the race.

In business: quarterly or annual feedback cycles are malpractice. Teams need micro-corrections that compound. In partnerships, in M&A diligence, in product roadmaps, delayed feedback means compounding errors.

When Zoom’s usage exploded 30x during COVID, engineering didn’t stick to quarterly plans. They implemented daily capacity reviews, hourly performance monitoring, and real-time resource reallocation. Speed of learning beat perfection of planning.

Operator move: build feedback loops that move faster than your competition’s decision cycles. If your industry adjusts quarterly, you need monthly cycles. If they adjust monthly, you need weekly recalibration. Fix things while you’re in the arena.

“The frequency of feedback is directly proportional to the rate of skill acquisition. Research shows that immediate feedback improves performance up to 40% faster than delayed feedback.” — Dr. K. Anders Ericsson

3. Shared Struggle Builds Culture

A 2,000-meter race is seven minutes of anguish. Frozen hands, burning lungs, legs on fire. Yet those are the memories rowers talk about decades later. Struggle forged bonds no offsite ever could.

Operator lesson: shared adversity creates trust faster than comfort ever will. In the boardroom, the champagne toast on close day doesn’t build culture. The grind of integration does. The hardest projects, the messiest restructurings, and the pivots under pressure are what forge real teams.

When I’ve deliberately assigned “stretch projects” that felt uncomfortable, teams didn’t just deliver, they came out bonded and more resilient.

4. Adaptability Is the Real Advantage

Lineups changed constantly. One week I was starboard, the next port. Bow pair could move to the engine room overnight. Constant reconfiguration forced adaptability, not as a value but as survival.

Operator lesson: change is no longer cyclical; it’s constant. Deals collapse, markets shift, priorities reset overnight. The leaders who win aren’t the ones with the most rigid plan—they’re the ones who treat reconfiguration as normal.

In M&A: the best integrations succeed because teams expect to reconfigure midstream. They adapt, recalibrate, and keep the boat moving.

McKinsey found that adaptable teams outperform others by 24% during major change. PwC found that adaptable organizations are three times more likely to become industry leaders.

“Research across 150 organizations found that teams with high adaptability scores outperformed their less adaptable counterparts by 24% during periods of significant change.” — McKinsey, Organizational Resilience Study (2023)

The Compound Effect

These four rules operate like the mechanics of a racing shell.

• Synchronization without feedback leads to coordinated failure.

• Shared struggle without pressure doesn’t build capability.

• Feedback without synchronization creates chaos.

When all four combine, something remarkable happens. Teams move faster than their individual capabilities suggest. They anticipate problems before they appear. They know when to push and when to adjust.

Rowing taught me that leadership isn’t about shouting from the stern. It’s about building systems where excellence emerges from the crew.

Final Pull

This is part of LineofSight.io, where I turn AI noise into operator strategy for non-technical GTM leaders.

Subscribe for weekly frameworks on building leverage, structuring deals, and scaling teams that compound.

Subscribe now

TLDR: Your best acquisition target may already be in your partner ecosystem. The integrations, joint customers, and roadmap overlap you see every day are not just partnership signals. They are M&A intelligence hiding in plain sight.

The Blind Spot in Corporate Development

Your next acquisition target is not buried in a banker’s pitch. It is already inside your ecosystem.

Snowflake underscored this at its 2025 Summit. The company highlighted how partner integrations now drive the bulk of usage across its Data Cloud. Partners are not just feature extensions — they are revenue engines. And when Snowflake acquired Streamlit (data apps) and Neeva (AI search), it wasn’t a surprise. Both had already been battle-tested inside the ecosystem.

Most corporate development teams miss this. They spend months scanning banker decks, attending demo days, and waiting for processes to start. Meanwhile, the highest-signal targets are already embedded in their product stack, serving joint customers, and quietly proving value every quarter.

The blind spot is not a lack of capital or vision. It is treating partnerships only as distribution, instead of what they truly are: live diligence processes.

Partnerships as Diligence

Partnerships have evolved. They are not just about co-marketing or channel reach. They are the ultimate diligence process.

Through a strong partner integration, operators can see:

• Real usage data

• Joint customer outcomes

• Cultural alignment in working teams

• Roadmap fit against strategic priorities

That is intelligence you will never find in a CIM.

Partnerships as Due Diligence

The strategic shift is simple. Partnerships are not just about distribution anymore. They are the ultimate diligence process.

Through a strong partner integration you can see:

• Real usage data

• Joint customer outcomes

• Cultural alignment in working teams

• Roadmap fit against your strategic priorities

This is intelligence that no banker deck will provide.

Case Studies: When Partnerships Became Acquisitions

• Salesforce + MuleSoft ($6.5B, 2018)
  Three years of partnership powering 1,000+ integrations. Premium of 36 percent, but de-risked by usage data.

• Dropbox + DocSend ($165M, 2021)
  Two years of API integration with 65 percent customer overlap. Integration complexity solved before the deal closed.

• Square + Afterpay ($29B, 2021)
  Eighteen months of payments integration with thousands of merchants. The partnership created a moat competitors could not cross.

These weren’t speculative bets. They were validated partnerships upgraded into acquisitions.

The Operator’s Take

Think of partnerships as extended job interviews. You would never hire a VP without observing performance. Why would you acquire a company without testing how it integrates with your systems, your customers, and your culture?

Traditional M&A relies on projections and promises. Partnership-first M&A relies on performance data. The difference is dramatic.

The Partnership-to-Acquisition Signal Framework

Green Light (Consider Acquisition):

• Partner drives more than 20% of new customer pipeline

• Joint solution is mission critical for shared customers

• Roadmaps align with your 3-year strategy

• Integration costs declining while value creation rises

• Retention rates improve for joint solutions

Yellow Light (Deepen Partnership):

• Strong technical fit but limited market traction

• Cultural alignment but unclear strategic value

• High customer satisfaction but low revenue attribution

• Integration works but monetization not clear

Red Light (Keep at Arm’s Length):

• Strategic priorities diverge

• Integration complexity rising

• Partner’s wins do not translate to your success

• Cultural friction shows up in collaboration

What Strategic Operators Can Track

You do not need to understand APIs or financial modeling to spot potential acquisitions. Track three business metrics:

• Customer retention: Joint solution retention vs standalone products

• Revenue attribution: Growth from partner-sourced deals and upsells

• Support volume: Tickets on integrated features (declining = good)

If these trend positively, you are looking at an acquisition candidate.

Every Partnership Is M&A Intelligence

Traditional process:

• Identify targets through banker outreach

• Run 60–90 days of diligence

• Decide on projections

• Hope integration works

Partnership-first process:

• Partner first to test fit

• Gather 12–24 months of data

• Decide based on proven performance

• Integrate along an already tested path

The Operator’s Checklist

Ask these five questions of every strategic partner:

1. How much of their revenue flows through our ecosystem?

2. Are we solving the same customer problems?

3. Are customers clearly more successful using us together?

4. Are our leadership teams strategically engaged?

5. If a competitor acquired them tomorrow, how badly would it hurt?

These separate “nice to have” partners from acquisition-ready opportunities.

Your Competitive Advantage

The edge is not having more partnerships. It is knowing when one should become an acquisition.

The playbook is straightforward:

• Partner first to test strategic and cultural fit

• Measure relentlessly to build performance data

• Act decisively when the signals align

Every integration is due diligence. Every joint customer is validation. Every quarterly business review is M&A intelligence.

Operators who adopt this mindset consistently find better targets, at better prices, with lower risk. If you are not evaluating your partner ecosystem with an M&A lens, someone else already is.

This is how you find your next acquisition hiding in plain sight.

Subscribe now

This is what I write about at Line of Sight: operator frameworks and playbooks that turn hidden signals into strategic advantage. Subscribe to get them directly.