We Spent 5 Months on a Deal That Was Never Ours
What We Set Out to Win
In early 2026, we entered a competitive evaluation with a mid-market SaaS company looking to replace their manual trial conversion process. Our product was technically stronger. Our demo went well. The champion inside the account was enthusiastic. We had every signal that pointed toward a close, so we ran the full cycle: discovery calls, technical deep-dives, a proof-of-concept, security review, and a pricing negotiation that stretched across three rounds.
Five months. That's how long we worked this deal before the CPO announced they were going with a vendor whose founder had been his college roommate. No RFP scoring. No final comparison. Just a relationship we didn't know existed and couldn't have competed with regardless of what we built.
This is not a story about losing gracefully. It's a post-mortem on the specific mistakes that kept us in a deal we should have disqualified in month one.
What Went Wrong, Step by Step
The warning signs were there. We ignored them because the product fit was real and our champion kept telling us what we wanted to hear.
We were single-threaded. Our entire relationship inside the account ran through one person: a VP of Growth who had no budget authority and limited political capital. We never got a meeting with the CPO. Every request to expand the conversation got deflected with "let me handle the internal alignment." That phrase, in retrospect, was a signal that alignment wasn't happening at all.
The Harvard Business Review's 2006 piece on The Sales Learning Curve makes this point plainly: enterprise sales cycles are shaped by buyer relationships and organizational politics far more than product features, and sales teams that don't map decision-making dynamics beyond technical merit consistently lose to vendors who do. That piece is nearly two decades old. The dynamic it describes has only intensified as buying committees have grown larger and AI vendor evaluations have become more politically charged.
We also misread timeline vagueness as normal enterprise friction. When procurement kept pushing the decision date, we told ourselves it was budget cycles. It wasn't. The decision had already been made informally. The evaluation we were participating in was theater, designed to satisfy a procurement checkbox while the real choice sat in a Slack DM between two people we'd never met.
The honest answer to "why did we stay in?" is that we had already invested so much time that leaving felt like waste. That's the sunk cost trap, and it's particularly dangerous in long sales cycles where each new week of effort makes exit feel more costly than continuation.
What We'd Do Differently: Lessons That Actually Hurt to Learn
The most useful thing I can offer here isn't a framework. It's three specific decisions we'd make earlier.
First: map the political layer in week two, not month four. Before we invest in a proof-of-concept, we now require a meeting with the economic buyer, not just the champion. If that meeting is blocked or indefinitely deferred, we treat it as a disqualification signal, not a scheduling problem. Champions who can't get us in front of decision-makers either lack influence or are managing us away from a conversation that would reveal the deal is already decided.
Second: ask directly about existing vendor relationships. "Is there a vendor you're already working with in this space, or a relationship that would make this a difficult switch?" Most buyers will tell you the truth if you ask plainly. We didn't ask. We assumed the evaluation was open. It wasn't.
Third: build pipeline intelligence into your operations infrastructure, not just your CRM. We now use automated signals to track deal health across the full cycle. When we built the PostHog Trial Conversion Intelligence pipeline, the core insight was that conversion behavior tells you things that self-reported data never will. The same principle applies to deal qualification: behavioral signals (response latency, stakeholder engagement patterns, meeting acceptance rates) reveal deal health more accurately than what your champion tells you on a call. If you're running n8n-based sales operations, the setup guide walks through how we instrument those signals in practice.
This approach has a real limitation worth naming: behavioral signal tracking requires enough deal volume to calibrate what "normal" looks like for your specific market. If you're running fewer than ten enterprise deals simultaneously, the patterns won't be statistically meaningful. In that case, the manual version of this, which is a weekly structured deal review against a fixed set of qualification criteria, is more reliable than any automated scoring system.
One more thing we learned the hard way, unrelated to sales but relevant to anyone building AI-assisted operations: reasoning models don't always behave the way their documentation implies. When we were building and testing our response parsing pipelines, we discovered that an LLM would wrap JSON output in markdown code fences even when the system prompt explicitly said "Return ONLY valid JSON, no markdown." Both test cases in our initial run dead-lettered with "Unexpected token backtick" parse errors. We caught it in testing, but only because we were looking. Now every response parser we ship strips markdown fences before JSON.parse. The model may or may not wrap output depending on prompt length and run variance. You can't predict it, so you always defend against it. The same defensive posture applies to enterprise deals: you can't predict relationship dynamics, so you always qualify against them explicitly.
What We'd Do Differently
We'd set a hard gate at week six. If we don't have a confirmed meeting with the economic buyer by week six of any enterprise cycle, we pause all investment and have a direct conversation with our champion about what's blocking access. Not a gentle nudge. A direct question with a clear consequence: "We need to meet the decision-maker before we proceed to the next stage. Can you make that happen by [date]?" If the answer is no, we treat the deal as low-probability and reallocate time accordingly.
We'd build a "relationship map" artifact in the first discovery call. Not a org chart. A map of who has informal influence over the decision, who has worked with competing vendors before, and who the CPO trusts outside the formal evaluation process. This is an uncomfortable conversation to have early. It's far less uncomfortable than losing five months to a deal that was never open.
We'd track our own disqualification rate as a success metric. The goal isn't to win every deal. It's to spend time only on deals that are genuinely winnable. A higher disqualification rate in months one and two means more capacity for deals where the evaluation is real. That reframe, from "how do we close more?" to "how do we exit bad deals faster?", is the actual lesson from five months of work that ended in a college roommate's Slack message.
If you're building sales operations infrastructure and want to see how we instrument deal-health signals in practice, the 18 months building autonomous agents post covers the broader architecture decisions that inform how we think about signal reliability across long-running pipelines.