Prashant Chandel
Writing
Article·2026-06-08·16 min read

The Measurement Moat: Why Some SaaS Companies Survive 3-7 Years and Others Die in 18 Months

Atlassian grew revenue 21% while seat count fell. Eli Lilly captured half the GLP-1 market despite a 70% price cut against them. Every competitive moat is temporary—but they break on different schedules.

StrategyAIBusiness

Atlassian did something strange in Q1 2026. Revenue grew 21%. Seat count—the number of people paying for Jira and Confluence—fell for the first time in company history. The link between "people using software" and "software value delivered" just snapped.

Meanwhile, in pharmaceuticals, Eli Lilly captured 51% of the weight loss drug market against Novo Nordisk, a competitor with a five-year head start who slashed prices 70%. Novo should have won. They didn't.

And in chip infrastructure, Rambus tripled their operating income in two years selling memory interface controllers with 80% gross margins. But Marvell just shipped a competing design that's 70% more power-efficient, and hyperscalers are switching. Rambus's moat is evaporating—analysts give it 15 months.

TL;DR — Every competitive advantage is temporary. But they don't all break at the same speed. Infrastructure moats last 12-18 months. Measurement moats—when a company can prove their product delivers better outcomes than competitors—last 3-7 years. Distribution advantages can last indefinitely, but only if you also build measurement credibility. The SaaS market is bifurcating: companies in categories where results are measurable (customer support, credit underwriting, weight loss drugs) will survive. Companies in unmeasurable categories (team productivity, collaboration) will commoditize within 18-24 months.

In this piece:

  • Atlassian's revenue grew while headcount shrank because AI automation broke the per-seat pricing model—the value software delivers no longer scales with the number of people using it.
  • Eli Lilly beat a competitor with a five-year lead and 70% lower pricing by proving their drug caused 22.5% weight loss versus 15-17%, showing measurement beats brand, timing, and price.
  • Cognition raised $26 billion at 53x revenue betting on a "routing moat" that expires in 18 months, but they're actually surviving on measurement moat (process embedding and outcome tracking) that lasts 24 months before OpenAI's 40x user advantage catches up.
  • The SaaS winners are companies like Zendesk that can measure customer outcomes precisely enough to charge premium pricing for 3-7 years before competitors figure out how to measure equally well.
  • After measurement commoditizes, the next moat is outcome guarantees—vendors taking contractual liability if customers don't hit promised results—but we don't yet know if that's actually defensible or just another temporary advantage.
The Hierarchy of Moat Durability Infrastructure 12-18 months lifespan Example: Rambus ▸ Memory interface IP ▸ 80% gross margins ▸ 3x operating income in 2 years Why it breaks: Marvell ships 70% more efficient design. HBM supply normalizes. Hyperscalers switch in 12-18mo cycles. SHORTEST MOAT Measurement 3-7 years lifespan Example: Eli Lilly ▸ 22.5% weight loss ▸ vs. 15-17% (Novo) ▸ 51% market share despite 70% price cut Why it breaks: Competitors publish equivalent efficacy data. Generics approved 2028-2030. Regulation forces data sharing. MEDIUM MOAT Distribution ∞* *with measurement parity Example: OpenAI ▸ 3M Codex users ▸ vs. 80K (Cognition) ▸ 40x data collection advantage Why it wins: Measurement parity in 18-24 months (faster data = faster proof). Then distribution beats switching costs. LONGEST MOAT prashant-chandel.org/blog

When the Per-Seat Model Dies

For twenty years, SaaS companies charged per seat. You want Slack for 100 employees? That's 100 seats × $15/month. The math was simple: more people = more revenue. Growth meant adding seats.

That model just broke.

Atlassian's Q1 2026 numbers tell the story. Revenue climbed 21% year-over-year. But for the first time ever, seat count declined. Fewer people were paying for Jira and Confluence, yet the company was making more money.

What changed? Automation. AI tools eliminated the need for some team members while making the remaining users dramatically more productive. A team that used to need 10 people to manage a project now needs 6—but those 6 are shipping more work than the original 10 ever did.

The software is delivering more value (measured in projects completed, bugs resolved, features shipped). But the old metric—seats purchased—no longer captures that value. Companies that stick to per-seat pricing are leaving money on the table or pricing themselves out of markets where automation is shrinking headcount.

This is happening across the entire SaaS industry. Slack raised their price from $20 to $45 per user this year—a 125% increase—calling it the "AI tax." They're trying to capture more revenue from each seat because they know seat counts are going to compress. But here's the problem: Slack can't actually measure whether teams using their product are more productive. Team collaboration and communication quality are fundamentally unmeasurable. You can't prove Slack delivered 30% more value, so you can't defend a 125% price increase.

Contrast that with Zendesk. At their Relate 2026 conference, they announced outcome-based pricing: customers pay per "verified resolution"—a customer support ticket that's confirmed resolved 72 hours later. Not per seat. Not per ticket opened. Per actual problem solved.

That's measurable. Zendesk can prove they resolved your customer's issue. They can benchmark their resolution rate against competitors. If they're better at solving problems, they can charge more—and defend that premium with data.

The Hierarchy of Moat Durability

This week, I tracked five different competitive moats across SaaS, infrastructure, pharmaceuticals, and fintech. All of them are temporary. But they expire on wildly different schedules.

Infrastructure moats last 12-18 months.

Rambus makes memory interface controllers—the chips that connect high-bandwidth memory to processors in AI servers. They had a near-monopoly. Operating income tripled in two years: $91.5 million in 2023 to $260.2 million in 2025. Gross margins hit 80%.

The moat was simple: AI chips needed faster memory, high-bandwidth memory (HBM) was in short supply, and Rambus owned the interface design most chip makers used. You literally couldn't build an AI chip without licensing their technology.

But moats built on scarcity evaporate when supply normalizes. HBM production is ramping up—supply should normalize by end of 2026, not 2028 like earlier forecasts predicted. And Marvell just shipped an alternative interface design that's 70% more power-efficient. Broadcom, Google, and Amazon are already using it in their custom TPU designs.

Rambus's moat window: 15 months from peak margins to competitive parity. Infrastructure advantages compress fast because once a better design exists, hyperscalers with the resources to switch will switch. There's no lock-in beyond the time it takes to redesign your chip—and these companies redesign chips every 12-18 months anyway.

Measurement moats last 3-7 years.

Eli Lilly launched tirzepatide (sold as Mounjaro for diabetes, Zepbound for weight loss) in 2022, five years after Novo Nordisk's semaglutide (Ozempic, Wegovy) hit the market. Novo had every advantage: five-year head start, brand recognition, established supply chains, and a 70% price cut to defend market share.

Lilly captured 51% of the combined GLP-1 market by April 2026.

How? Measurement. Lilly's drug causes 22.5% average weight loss in clinical trials. Novo's causes 15-17%. Lilly didn't beat them on price, brand, or distribution. They beat them on outcomes—and they could prove it with data.

This is a measurement moat. When your product delivers measurably better results, you can charge a premium and customers will pay it because you have proof. Novo dropped prices 70% and still lost market share because efficacy matters more than cost when the difference is 22.5% versus 15-17% weight loss.

But measurement moats expire, too. Lilly's advantage holds until one of three things happens:

  1. A competitor publishes equivalent or better weight loss data
  2. Generic versions of tirzepatide get regulatory approval (likely 2028-2030)
  3. Regulation forces Lilly to license manufacturing or share data to improve access

Lilly's patents technically run until 2035-2036. But regulatory approval pathways for biosimilars and international pricing pressure compress the real moat to 3-5 years of peak margins. After that, they'll still sell the drug—but at lower prices and smaller market share.

Three to seven years is dramatically longer than 12-18 months. That's the difference between infrastructure moats and measurement moats. Infrastructure advantages break when someone builds a better design. Measurement advantages break when someone proves equivalent outcomes—which requires years of clinical data, regulatory approval, and real-world validation.

Distribution advantages last indefinitely—but only with measurement parity.

OpenAI shipped Codex in April 2026—a native GPT-5.5 coding agent that competes directly with Cognition's Devin. Codex hit 3 million weekly active users within two months. Devin has 80,000.

OpenAI's distribution advantage is 40x. They should have crushed Cognition.

They didn't. Devin's user count stayed flat—80,000 to 90,000 from April to June. Developers who had already adopted Devin didn't switch to Codex, even though it was free for OpenAI subscribers and integrated directly into ChatGPT.

Why? Switching costs. Once a team embeds Devin into their development workflow—connects it to their CI/CD pipeline, trains it on their codebase, builds measurement baselines for code quality and review time—switching to a new tool means re-doing all of that work. Even if Codex is marginally better, the switching cost exceeds the benefit.

This looks like a process lock-in moat. But it's not. It's a measurement moat.

Developers stick with Devin because they have data showing it works for their specific use case. They've measured how long code reviews take before and after Devin. They know their bug rates. They've quantified the productivity gain. Switching to Codex means losing that measurement baseline and starting over.

But here's the problem: OpenAI's 3 million users are accumulating measurement data 40x faster than Cognition's 80,000 users. After 18-24 months, OpenAI will have statistical power (enough real-world data) to prove Codex delivers equivalent or better outcomes than Devin across a wide range of use cases. At that point, the measurement moat collapses. New customers will choose Codex because it has better distribution (integrated into ChatGPT) and equivalent proven outcomes.

Cognition's moat compresses from "3-7 years defensible" to "18-24 months before OpenAI catches up." Distribution advantage beats measurement moat when the competitor with distribution also builds measurement credibility faster.

This is the pattern: distribution lasts indefinitely, but you need measurement parity to convert distribution into market share. OpenAI has distribution now. They'll have measurement parity in 18-24 months. After that, Cognition's only defense is locking in customers with long-term contracts or moving to a new moat before OpenAI's measurement catches up.

The Measurable vs. Unmeasurable Divide

The SaaS market is splitting into two groups: companies in categories where outcomes are measurable, and companies in categories where outcomes are not.

Measurable categories survive 3-7 years:

  • Customer support (Zendesk): Resolution rates, time-to-resolution, customer satisfaction scores—all measurable. Zendesk can prove their AI agents resolve tickets better or faster than competitors. That's defensible until competitors measure equally well, which takes 3-5 years of data accumulation and product iteration.

  • Credit underwriting (Cashfree, Indian fintech): Default rates, approval accuracy, processing time—all measurable. Cashfree hit ₹1,000 crore revenue and EBITDA profitability in March 2026 because they can prove their underwriting models have lower default rates than competitors. That measurement moat lasts until competitors collect equivalent loan performance data and tune their models to match.

  • Code quality and review time (Cognition): Bug rates, review cycles, deployment frequency—all measurable. Devin survives against Codex because existing users have measurement baselines proving it works. That lasts 18-24 months until OpenAI's larger user base generates equivalent data.

Unmeasurable categories commoditize in 18-24 months:

  • Team collaboration (Slack): Productivity, communication quality, team cohesion—fundamentally unmeasurable. You can count messages sent or meetings scheduled, but you can't prove Slack made your team 20% more productive. Without measurement, there's no defensible premium. Slack's 125% price increase is unsupported by data. Customers will switch to cheaper alternatives (Teams, which is bundled free with Office 365) unless Slack has lock-in from integrations or workflows.

  • Project management (Atlassian): Project velocity, team efficiency, planning accuracy—hard to measure in isolation. Jira can show you how many tickets closed, but it can't prove your team shipped projects faster because of Jira versus because they hired better engineers or simplified the product roadmap. Without clear attribution, Atlassian's pricing power depends on switching costs (re-creating workflows, re-training teams) not proven outcomes.

This is the bifurcation. Measurable-category SaaS companies can charge premium pricing and defend it with data for 3-7 years. Unmeasurable-category SaaS companies either commoditize (race to the bottom on price) or survive on lock-in (switching costs exceed the benefit of moving to a cheaper alternative).

Aside: There's a third path—hybrid pricing. Snowflake and Datadog charge per usage (consumption-based pricing) instead of per seat or per outcome. They achieved 120%+ net revenue retention (each customer spends 20% more year-over-year) while margins compressed from 80% to 65%. This model works when you can measure consumption (compute used, logs processed) but not outcomes (whether that compute delivered ROI). It shifts revenue predictability risk to customers—you pay more when you use more—but it survives because customers accept the trade-off: they'd rather pay for what they use than overpay for seats they don't need. The catch: consumption pricing only works at scale. Smaller SaaS companies can't absorb the revenue volatility.

What Comes After Measurement?

If infrastructure moats last 12-18 months and measurement moats last 3-7 years, what moat lasts longer than that?

The evidence this week points to outcome guarantees—vendors taking contractual liability if customers don't hit promised results.

ServiceNow announced in early 2026 that they're moving toward guaranteed productivity gains. Not "we think our AI agents will reduce ticket volume by 30%"—a contract that says "if ticket volume doesn't drop 30%, we refund you or credit your account."

This is different from measurement-based pricing (pay per verified outcome). It's liability-based pricing (vendor assumes downside risk if outcomes don't materialize).

Why would this be more defensible than measurement alone? Because it forces the vendor to internalize risk. Companies that offer outcome guarantees have to be extremely confident in their ability to deliver—which means they need:

  1. Deep measurement infrastructure to track whether the promised outcome actually happened
  2. Process control to isolate their product's impact from other variables (did ticket volume drop because of our AI agent, or because you hired fewer support reps?)
  3. Financial reserves to absorb refunds when guarantees aren't met

This creates a moat. Competitors can copy your measurement approach in 3-5 years. But they can't copy your ability to guarantee outcomes without first building the same measurement infrastructure, process controls, and financial reserves. That takes 5-10 years.

The open question: does this actually work across multiple business functions simultaneously, or does it collapse the moment one function fails to deliver?

ServiceNow is betting outcome guarantees work because they can isolate variables—they control the IT service management workflow end-to-end. If ticket volume doesn't drop, they can diagnose whether it was their AI agent's fault or an external factor (customer changed their product, hired a bunch of junior engineers who create more tickets, launched in a new geography with different support needs).

But in categories where you don't control the full workflow, outcome guarantees are nearly impossible. Slack can't guarantee your team will be 20% more productive because too many other variables affect productivity. Jira can't guarantee projects will ship faster because project velocity depends on engineering talent, product complexity, and strategic decisions—not just the project management tool.

So outcome guarantees might be the next durable moat—but only in categories where vendors can isolate their impact from external variables. That's a much smaller set of use cases than measurement moats.

Who Survives the Next Three Years

Here's the scorecard based on this week's evidence:

Survivors (3-7 year moats):

  • Zendesk: Measurable outcomes (verified resolutions), outcome-based pricing already live, defensible until 2029-2030 when competitors build equivalent measurement infrastructure.
  • Cashfree and Indian fintech: Measurable outcomes (default rates), regulatory moats (compliance costs equivalent to 32% of annual revenue create barriers to entry), defensible until lending market saturates 2028-2029.
  • Eli Lilly: Measurable outcomes (22.5% weight loss vs. competitors' 15-17%), patent protection until 2035 but real moat compresses to 2028-2030 due to biosimilar approvals and pricing pressure.
  • Cognition: Measurement moat (process embedding + outcome tracking) lasts 18-24 months until OpenAI's 40x distribution advantage translates to measurement parity. Needs to either lock in long-term contracts now or move to outcome guarantees before mid-2028.

Commoditizing (18-24 month windows):

  • Slack: Unmeasurable outcomes, price increase unsupported by data, survives only on switching costs (workflow integrations). Teams bundling advantage wins unless Slack builds lock-in faster than Microsoft subsidizes Teams.
  • Rambus: Infrastructure moat evaporating in 15 months as HBM supply normalizes and Marvell's more efficient design gains adoption. No measurement moat (memory interface performance is measurable, but Marvell already proved theirs is better).

Unknown (too early to call):

  • ServiceNow outcome guarantees: If they work, this is a 5-10 year moat. If they don't—if guarantees fail in production and ServiceNow pays out refunds that exceed the premium pricing they charged—it proves outcome guarantees are not actually defensible at scale.

The real conclusion: SaaS companies in measurable categories should use their 3-7 year window to either (1) move to outcome guarantees before competitors catch up on measurement, or (2) prepare for consolidation—sell to a strategic buyer at 4-5x ARR while the moat still looks defensible.

Companies in unmeasurable categories have 18-24 months to build lock-in (integrations, workflows, data gravity) or accept commodity pricing. The per-seat model is dead. Measurement moats are real. And the next moat after measurement—outcome guarantees—only works if you can isolate your product's impact from everything else affecting the customer's results.

The question isn't whether moats are temporary. They all are. The question is: how long does yours last, and what do you build before it expires?

Sources & Further Reading

SaaS Pricing & Market Dynamics:

AI Coding Agents & Infrastructure:

Chip Infrastructure & Memory:

Pharmaceuticals & GLP-1 Market:

India Fintech & Regulatory Structure: