Last week, a friend who runs a B2B AI product team told me a story that should terrify anyone selling AI right now. His AI assistant had been live for a year. Token volume was up 800%. Daily active users were through the roof. The metrics were flawless.
Then came the renewal negotiation. The customer’s CIO looked at the dashboard, nodded, and said: “Usage is great. But next year’s budget is tight. We’re not renewing.”
My friend scrambled to pull up the data to prove their value. The CIO cut him off: “I see the data. But tell me—if we unplug this tool, does my business stop?”
My friend was speechless. No, the business wouldn’t stop. Things would just be a bit slower. A bit more manual. And in the cold, hard math of a CIO’s Excel spreadsheet, “a bit slower” doesn’t justify a line item.
This scene is repeating itself across countless AI product teams. We are building features with beautiful DAU metrics and exploding token consumption, while collectively forgetting the most fundamental rule of B2B software: companies don’t buy your product to “use it.” They buy it because they “can’t live without it.”
If unplugging your AI doesn’t stop a business workflow, you have zero pricing power.
The AI market is currently undergoing a brutal 80/20 culling. In most enterprises, over 80% of internal AI applications are “dead assets.” People use them, sure—but only for edge cases. An HR rep checking attendance rules. A sales rep occasionally generating a cold email script. The core business runs exactly the same way with or without the AI.
Only the top 20% of AI features actually matter. And they matter because they sit directly on the critical path. They are embedded in mandatory contract review nodes. They drive production scheduling. They trigger automated responses to equipment failures. You cannot bypass them without breaking the business.
How do you know if your product is a core asset or a dead asset? Run the “Unplug Test.”
Imagine ripping your AI out of the customer’s system today. What happens?
Scenario A: Employees complain that they can’t auto-generate a summary, sigh, open a blank document, and do it manually. The work gets done. It just takes twice as long.
Scenario B: The contract approval pipeline halts because no one can do the initial risk screen. The predictive maintenance module crashes because no human can translate the raw alarm signals into repair orders fast enough.
In Scenario A, your product is an accelerator. Nice to have, but ultimately disposable. In Scenario B, your product is a business node. If you disappear, the chain breaks.
External tools can be replaced on a whim. Business nodes hold the pricing power.
For the past three years, AI product competition has been entirely concentrated in Scenario A. Everyone is building AI copywriters and AI presentation makers. We are all selling “time-saving.” But time-saving never ranks in the top three priorities of a corporate finance department. What keeps a CIO awake at night isn’t employee inefficiency—it’s broken business processes.
Of course, how deeply you can embed depends entirely on who is sitting across the table from you. You can’t sell to an enterprise elephant the same way you sell to a mid-market cheetah.
When selling to “Elephants” (massive conglomerates), you are dealing with layers of middle management and heavy legacy baggage. They don’t want you to rebuild their SAP systems. They want vines that can grow on top of their existing infrastructure. Your pitch shouldn’t be “disruption.” It should be “compatibility” and “smooth migration.”
When selling to “Cheetahs” (mid-sized companies), you are dealing with a founder who isn’t afraid to make cuts. They are looking at AI as a survival mechanism to replace 30% to 50% of fixed headcount. Your value proposition must be brutally simple: “Use this, and you don’t need to hire half your customer service team.”
When selling to “Gazelles” (small physical businesses like factories or logistics hubs), they don’t care about your LLM architecture. They care if your AI can predict when a machine will break down and save them electricity. They want a high-ROI point solution that deploys in under two weeks.
And when selling to “Whales” (banks, energy, highly regulated industries), efficiency is secondary. Compliance is God. If your AI’s reasoning isn’t traceable and auditable, you won’t even make it into the testing environment.
But navigating the buyer is only half the battle. The real twist in enterprise AI adoption isn’t about the technology—it’s about organizational politics.
Most AI product managers completely ignore the most dangerous stakeholder in the room: the customer’s middle manager.
We assume that if the CEO buys the vision and the frontline workers use the tool, the deal is won. False. Middle managers hold the invisible veto power. And right now, they are terrified.
The direct effect of AI is headcount reduction and layer flattening. These are the exact two things middle managers fear most. If your product makes them feel obsolete, they will ensure your project suffers a “soft death.” They will withhold data. They will refuse to integrate workflows. They will drag their feet until the license expires.
Middle managers hold the invisible veto power. If your AI makes them feel replaceable, they have ten thousand ways to ensure your project quietly fails.
The solution isn’t to hide the reality of AI from them. The solution is to give them a new throne. You must provide a “Human-AI Division of Labor” framework. Tell the middle manager: “AI will handle the scaled execution. Your job is no longer managing people—your job is managing the quality of the AI’s output.”
By giving them the role of AI Supervisor, you turn your biggest obstacle into your biggest renewal ally.
If you are looking at a potential churn risk right now, stop tweaking your model’s accuracy. Do these three things immediately.
First, throw away your feature list and draw the customer’s business swim lane. Map the entire workflow from input to delivery. Find out exactly where your AI touches the process. If your AI’s output still requires a human to copy and paste it into another system, your embedding depth is zero. Push your integration until your AI writes directly back into the core business database.
Second, treat “auto-write-back” as your core feature, not a nice-to-have. If your AI just displays a result on a screen, it’s a browser plugin. If your AI writes the result directly into the Jira ticket or the OA approval flow, it becomes the business. Auto-write-back means the customer cannot unplug you without breaking the pipeline.
Third, actively design the middle manager’s next role. Provide templates for human-in-the-loop intervention. Build audit trails into your AI’s decision-making process. Give the middle manager a dashboard to oversee, correct, and approve the AI’s work. Make them the master of the machine.
The era of treating “Daily Active Users” as the holy grail of AI product success is ending. The PMs who survive this shift will be the ones who walk into the customer’s operational floor, map the swim lanes, and dare to run the Unplug Test.
If your product is just a nice-to-have accessory, don’t blame the customer for cutting the budget. They didn’t do anything wrong. They just finally did the math.
Stop trying to make your AI “easier to use.” Start making it impossible to unplug.
FAQ
Q: If our DAU and token volume are at all-time highs, why would a customer still churn?
A: Because usage does not equal dependency. If your AI is just an accelerator that saves time, it is the first budget line cut during a squeeze. CIOs only renew tools that sit on the critical business path—where unplugging the AI means the workflow completely stops.
Q: How do we practically apply the 'Unplug Test' to our product?
A: Ask yourself: if your AI vanished from the customer's system today, what happens? If employees just sigh and do the work manually, you are a disposable accessory. If a contract approval pipeline halts or a production schedule breaks, you are a core business node with real pricing power.
Q: If AI is meant to increase efficiency, why do middle managers sabotage the rollout?
A: Because AI threatens their survival. AI flattens layers and reduces headcount, which is exactly what middle managers fear. Unless you reposition their role from 'managing people' to 'supervising AI quality,' they will use their invisible veto power to starve your project of data and workflow integration.