The Revenue You Quoted at Signing Is the Trust You Are Spending

In May 2026, Scott Stevenson, the co-founder of legal AI startup Spellbook, took to X to describe what he called a "huge scam" running through the industry. AI companies, he wrote, were inflating ARR beyond recognition, using constructed metrics to win funding rounds, attract press coverage, and position themselves as category leaders before the category had actually been built. The biggest funds in the world were supporting it. Journalists were being handed numbers that had no operational anchor. The announcement was the product, and the product was the announcement.

Most coverage of this story framed it as a fundraising problem, which is the smallest version of what it actually is. The founders entering partnership conversations carrying those same numbers are not just misrepresenting themselves to investors. They are setting the expectations their strategic partners will carry into every subsequent conversation, every deliverable review, and every revenue reconciliation for the next twelve months. The metric built to close a round does not retire at the close. It travels forward, carried by the partner who accepted it as the basis for their own commitments.

That is the structural problem the ARR inflation moment reveals, and it is the one that never gets named in the post-mortems. Founders lose partnerships over the gap between the number that opened the relationship and the one that materialized inside it, and almost none of them trace the origin of that gap back to the conversation at month zero.

The Metric Was Built for the Pitch

What ARR inflation actually represents is optimization for the close. Founders who have been in enough rooms understand that the first meeting is a performance and the metric is its most important prop. The partner across the table is evaluating leverage, and the founder is presenting the most compelling possible version of their current position. This is not unique to AI companies in 2026. It is the oldest structural problem in deal-making: the incentive at the selection stage is misaligned with the incentive at the execution stage.

The selection stage rewards impression. You need to be selected before you can perform, so the entire evaluation is oriented toward creating an impression that earns selection. The execution stage rewards accuracy. Once you are inside the partnership, the number that matters is the one that appears in the reporting, and the partner is measuring it against the number from the first meeting, whether they say so explicitly or not.

Founders who enter with inflated metrics are not, for the most part, pathologically dishonest. They are doing exactly what the selection stage incentivizes, which is to present the most favorable interpretation of every available data point. The problem is that partnerships are not closed at the selection stage. They are built after it, and the building happens in the space between what was implied and what is true.

Builder.ai raised $450 million from Microsoft, SoftBank, and others. It promised to democratize software development through AI and positioned itself as a company with the technical architecture to deliver it at scale. By mid-2025 it had filed for bankruptcy, laid off over a thousand people, and left hundreds of clients and partners demanding refunds. The postmortem pointed to fraudulent accounting, misrepresented capabilities, and a business model that could not survive contact with scrutiny. Its partners were not blindsided by a product that underperformed. They were blindsided by the distance between the capability that had been described and the one that existed, and that distance had been present in the room at signing.

When the Number Unravels, So Does the Agreement

The cost of entering a partnership with fabricated leverage is that the relationship begins in debt. Every week that passes before the number corrects itself is a week your partner is building execution plans, making commitments to their own stakeholders, and calibrating their investment of time and resources against an expectation you set deliberately. When the correction arrives, it does not arrive as a single conversation. It arrives as a recalculation that runs backward through every prior interaction and reprices each one.

This is why partnerships that collapse over "revenue disappointment" are almost never actually about revenue. The revenue is the final visible symptom of a trust architecture that was faulty from the beginning. The founders involved are rarely able to reconstruct the original conversation accurately, because both parties have been revising their internal version of it for months, each revision bringing the remembered agreement closer to the outcome they were already experiencing. By the time the number is officially wrong, the partnership is already over. The formal conversation is just documentation.

The founders who consistently build partnerships that compound share a specific and observable quality: they enter conversations with an accurate account of their current position, including its limitations. This is not modesty, and it is not a negotiating disadvantage. It is the only approach that creates a partner who can actually plan around the real business rather than the presented one. A partner who knows your real number at month zero has no distance to travel when month twelve arrives. A partner who was given the optimized version of that number has twelve months of accumulated miscalibration to work through, and they will do that work at the precise moment you are most dependent on them to stay the course.

The partnership selection process rewards founders who perform confidence. The partnership execution process rewards founders who established accuracy. Most founders never resolve that tension, which is why the most common post-mortem in a broken deal is not about the deal structure or the market conditions. It is about a number that was given in the first conversation that turned out to be the most expensive thing anyone in the room said all day.

Platforms like onSpark are built around the idea that partnership quality is a function of match quality, and match quality depends entirely on what both parties actually bring to the table rather than what they chose to present at the close. The founders who use it to find partners with the most durable results are the ones who put the real number in the room from the start, because they understand that the partner who accepts accurate information and still wants to move forward is the only partner worth having.

The founders who are best at closing partnerships are not always the ones who built the most durable ones. The founders who built durable partnerships almost always turned out to be the ones who were least impressive in the first meeting, because they were describing something real rather than performing something aspirational, and the partner sitting across from them could tell the difference.