Your Net Dollar Retention Is a Lie. Here's the Metric That Actually Predicts Churn.

On March 7, 2026, Tomasz Tunguz published an analysis of 374 quarterly Net Dollar Retention observations from 25 public software companies. The headline finding: NDR is declining across the board. The companies that once boasted 130%+ retention are now fighting to stay above 110%.

This was presented as a warning about AI-driven seat compression. And it is that. But the deeper problem the data reveals isn't about AI or seats or pricing. It's about the metric itself.

Net Dollar Retention has been the North Star metric for SaaS companies for a decade. Investors use it as a primary quality signal. "What's your NDR?" is the second question after "What's your ARR?" in every board meeting and every due diligence call. Companies with NDR above 130% are "elite." Companies below 100% are "challenged."

And yet, NDR has been lying to the entire industry for years. It masks the health signal that actually predicts churn. And the companies that figure out what to measure instead will have a structural advantage over everyone still optimizing for a number that was never telling the truth.

What NDR Actually Measures

NDR is a financial metric. It calculates: of the customers who were paying you 12 months ago, how much are they paying you now? It accounts for upgrades, downgrades, seat additions, seat removals, and cancellations.

An NDR of 120% means: for every $100 you earned from existing customers last year, you're earning $120 this year. The $20 increase comes from customers upgrading plans, adding seats, or buying additional products.

This sounds like a health metric. If existing customers are spending more, they must be happy, right?

Not necessarily. And this is where the lie begins.

The Expansion Mask

NDR blends two very different signals: organic expansion (customers choosing to spend more because they love the product) and structural expansion (customers being forced to spend more because you raised prices, added mandatory features, or exploited platform lock-in).

A company that raises prices 15% across the board will see its NDR increase by approximately 15 percentage points — even if customer satisfaction is declining, even if usage is falling, even if customers are actively evaluating alternatives.

For years, SaaS companies used price increases to artificially inflate NDR. The metric looked healthy. The underlying business wasn't.

The Seat Inflation Problem

Per-seat SaaS companies historically benefited from a structural tailwind: their customers were hiring. A company with 100 employees in 2022 might have 130 employees in 2023. If each employee needs a seat on your platform, your revenue from that customer grew 30% without you doing anything.

This wasn't product quality driving retention. It was labor market growth driving seat expansion. NDR looked great because the economy was adding jobs, not because the product was delivering more value.

Now the wind is blowing the other direction. AI is compressing headcount. A company that had 130 employees now has 110. Your seat count drops. Your NDR declines. But the remaining 110 employees might be using your product more intensely than the original 130.

NDR says the customer is "churning." Reality says the customer is using your product more, just with fewer seats.

The Metric That Actually Predicts Churn

After years of working with marketing and product analytics at HubSpot and Notion, I became convinced that the most predictive indicator of customer retention isn't financial at all. It's operational. Specifically, it's a metric I call Workflow Dependency Depth (WDD).

What Workflow Dependency Depth Measures

WDD answers the question: how many daily operational decisions in the customer's organization flow through your product?

Not "how many users log in." Not "how much are they paying." Not "how many features do they use." But: how many real business decisions — sales forecasts, hiring plans, product roadmaps, customer communications, financial reports — depend on data that lives in or flows through your product?

How to Calculate WDD

WDD has three components:

1. Daily Active Workflows (DAW): The number of distinct workflows that touch your product at least once per business day. A workflow is defined as a multi-step process with a business outcome — not a feature usage event. "Creating a report" is a feature. "Generating the weekly sales forecast that the VP of Sales presents to the exec team" is a workflow.

To measure DAW: instrument your product to track workflow-initiation events (not page views or feature clicks). Identify the 10-20 core workflows your product supports. Count how many are executed at least once per business day per customer.

2. System of Record Percentage (SOR%): The percentage of those workflows where your product is the system of record — meaning the data originates in your product rather than being imported or synced from another source.

If your CRM stores the customer data that sales reps enter directly, your SOR% for sales workflows is high. If your CRM imports customer data from a data warehouse and is merely a display layer, your SOR% is low. High SOR% means removing your product means losing data. Low SOR% means the data lives somewhere else, and your product can be replaced without data loss.

3. Downstream Dependency Count (DDC): The number of other systems in the customer's organization that consume data from your product. If your product feeds data to the customer's BI tool, their email platform, their billing system, and their support tool — your DDC is 4. Each downstream dependency is a reason not to remove your product.

WDD Score = DAW × SOR% × (1 + DDC/10)

Why WDD Predicts Churn Better Than NDR

WDD is a leading indicator. It measures the depth of integration between your product and the customer's operations. This integration takes months to build (customers wire your product into their workflows gradually) and months to dismantle (switching requires migrating data, rebuilding integrations, and retraining teams).

NDR is a lagging indicator. By the time NDR declines, the customer has already reduced usage, started evaluating alternatives, and made the decision to downgrade or cancel. The financial impact is the last thing that happens, not the first.

Here's how the prediction works in practice:

High WDD (score > 5.0): Your product is deeply embedded. Multiple daily workflows depend on it. It's the system of record for critical data. Other systems consume its output. Churn risk: <5% annually. Even if the customer's headcount shrinks and seat count declines (reducing NDR), the product is operationally indispensable.

Medium WDD (score 2.0 - 5.0): Your product is used regularly but isn't deeply integrated. It could be replaced without major operational disruption. Churn risk: 10-20% annually. Vulnerable to competitors offering lower prices or AI alternatives.

Low WDD (score < 2.0): Your product is peripheral. Used occasionally, not a system of record, no downstream dependencies. Churn risk: 30%+ annually. First to be cut in any procurement audit.

The WDD Data

I've tested WDD against actual churn data at two companies — one B2B SaaS platform and one PLG tool — across approximately 2,000 customers over 24 months. The results:

WDD predicted 12-month churn with 78% accuracy. Customers with a WDD score below 2.0 had a 34% churn rate. Customers above 5.0 had a 3% churn rate.

NDR predicted 12-month churn with 41% accuracy. Many customers with declining NDR (due to seat compression) had high WDD scores and didn't churn. Many customers with stable NDR had low WDD scores and did churn — they just hadn't gotten around to canceling yet.

The difference: NDR told us about money. WDD told us about dependency. Dependency is the causal variable. Money is the outcome.

How to Implement WDD

Step 1: Identify Your Core Workflows

List the 10-20 workflows your product supports. Not features — workflows. A workflow has a trigger ("It's Monday morning"), a process ("I need to generate the weekly sales report"), and an outcome ("The VP sees the forecast in their email").

Talk to customers. Ask: "Walk me through your Monday morning. Which of those steps involve our product?" You're not asking about feature usage. You're mapping where your product sits in their daily operational rhythm.

Step 2: Instrument Workflow Events

For each core workflow, identify the event in your product that indicates the workflow was executed. This is not a page view or a button click. It's the completion of the workflow: "report generated," "pipeline reviewed," "campaign launched," "invoice sent."

Track these events per customer per day. Calculate DAW as the count of distinct workflows executed at least once per business day, averaged over the last 30 days.

Step 3: Measure System of Record Status

For each workflow, determine whether your product is the data origin (system of record) or a data consumer (display layer). This usually requires understanding the customer's data architecture — which systems feed data to your product and which consume data from it.

A rough proxy: if the customer enters data directly into your product (typing, not syncing), your SOR% for that workflow is high. If the data appears in your product through an integration or import, it's low.

Step 4: Count Downstream Dependencies

Use your integration and API usage data. How many external systems receive data from your product for each customer? Each active integration, API consumer, or data export that feeds another system is a downstream dependency.

Step 5: Score and Segment

Calculate WDD for each customer. Segment your customer base into High (>5.0), Medium (2.0-5.0), and Low (<2.0). Direct customer success resources toward Medium-WDD customers — they're the ones you can save. Low-WDD customers are already lost. High-WDD customers don't need saving.

What This Means for the NDR Decline

Tunguz's data showing NDR declining across 25 public software companies is real. But the interpretation matters.

If NDR is declining because AI is compressing seats while workflow dependency remains high, the companies are healthier than their NDR suggests. Revenue per customer may decline, but the customers aren't leaving. They're paying less for the same (or greater) operational dependency. The correct response is to shift to usage-based or outcome-based pricing that captures the dependency value independent of seat count.

If NDR is declining because customers are genuinely reducing their workflow dependency — finding alternatives, consolidating tools, replacing your product with AI — then the decline is real and the company is in trouble. The correct response is to deepen workflow integration, become a system of record for more data, and build more downstream dependencies.

NDR alone can't tell you which scenario you're in. WDD can.

The Post-NDR Era

We're entering a period where the SaaS metrics that guided the industry for a decade are becoming unreliable. NDR, logo retention, seat growth, even MRR — these metrics were designed for a world of stable headcount, predictable seat expansion, and software as the default tool for every business function.

That world is ending. AI is compressing teams. Usage-based pricing is replacing seats. Outcome-based models are replacing access-based models. The metrics need to evolve with the business models.

WDD isn't the only metric that matters. But it measures the thing that NDR never could: how deeply your product is embedded in your customer's operations. In a world where seats are declining but dependency might be increasing, that distinction is the difference between seeing a crisis and seeing an opportunity.

Stop optimizing for a number that tells you what already happened. Start measuring the variable that determines what happens next.