The KPI Hub

Tag: AI COGS

  • SaaS Market Flash — September 27, 2026: GPT-6 Astra Just Reset Every AI COGS Model You Built This Year

    SaaS Market Flash — September 27, 2026: GPT-6 Astra Just Reset Every AI COGS Model You Built This Year

    OpenAI shipped GPT-6 Astra today with the most aggressive framing it has ever attached to a launch: “Welcome to the AGI era.” Marketing hyperbole or not, that’s not the number that should move your board deck. The number that matters is the one sitting underneath it — the companion tier, GPT-6 Sol & Luna, launched on Product Hunt at “half the price” of prior frontier access.

    Read those two facts together and you get the actual market movement of September 27, 2026: frontier AI capability is deflating in price faster than almost any SaaS input cost in the last decade, while the vendor supplying it is running one of the largest negative cash positions in technology history. Per figures surfaced on 20VC via SaaStr, OpenAI’s internal forecast puts total burn through 2030 at $278 billion.

    That combination — collapsing prices, exploding vendor cash needs — is not stable. And every SaaS company with an AI feature line in its gross margin is exposed to how it resolves.

    The Deflation Is Real, and It Is Not a Gift

    If you priced an AI feature in Q4 2025 or Q1 2026, you almost certainly modelled inference cost as a declining but predictable line — the “it gets cheaper every year” assumption that made AI-attached pricing survivable. Today’s launch validates that assumption in the short term and destroys it in the medium term.

    Here’s the mechanic. A 50% price cut on frontier-adjacent access does three things simultaneously:

    • It expands your gross margin this quarter — if you don’t pass the savings through. Most won’t, immediately.
    • It expands your competitor’s addressable feature set — capabilities that were economically unviable at last quarter’s token price just became viable for someone with half your funding.
    • It compresses your pricing power — because buyers read the same headlines. Procurement teams that saw “half the price” will arrive at renewal with that number in hand.

    The third effect is the one that gets underestimated. AI-attached pricing premiums were justified to buyers on the basis that intelligence was scarce and expensive. When the vendor publicly halves the price, the scarcity narrative dies. You are now selling workflow, data gravity, and distribution — not model access. If your AI premium was ever really a model-access markup, it has a repricing event scheduled for your next renewal cycle.

    The $278B Problem Nobody Is Modelling

    Now the other side. A vendor forecasting $278 billion in burn through 2030 does not stay in permanent price-cut mode. That capital has to be recovered, and the recovery paths are limited: higher enterprise pricing, tiered rate limits, usage caps, priority-access fees, or margin extraction on the API layer once competitive pressure eases.

    This is why the single most actionable output of today is a change in contracting posture:

    Price your AI COGS on a 12-month re-negotiation cycle, not a 36-month one. Any multi-year customer contract with AI usage bundled at a fixed price is a short position on model pricing you did not intend to take.

    Practical version: if you sell three-year enterprise deals with unlimited or generously-capped AI usage, you have written an option to your customer for free. Cap it, meter it, or add an explicit input-cost adjustment clause. The CFOs who survive 2027 will be the ones who did this in 2026.

    Cross-Signal: Capital Is Voting Against Pure Model Exposure

    Today’s funding tape tells the same story from the other direction. Two deals landed, and neither rewarded model proximity:

    • Fresha: $80 million from KKR at a $1 billion valuation. A beauty and wellness booking marketplace. Vertical, SMB-dense, payments-adjacent. Not an AI-native story at all.
    • Lio: $30 million Series A led by a16z for AI enterprise procurement — AI applied to a specific, high-friction budget line, not a general-purpose assistant.

    Meanwhile, TechCrunch reported VCs actively publishing what they will no longer fund in AI SaaS. The rotation is unambiguous: away from horizontal AI wrappers, toward vertical workflow ownership and AI applied to a named cost centre. Fresha getting a billion-dollar mark from KKR — a firm that underwrites cash flows, not narratives — on a booking marketplace is the loudest possible signal that durable transaction ownership now outranks model access in the capital stack.

    And the Control-Clause Tail Risk

    The third signal reinforces why model dependency is a governance issue, not just a cost issue. Anthropic was designated a Pentagon supply-chain risk, a $200 million contract was terminated, and a reported $2 trillion IPO slipped to November — the dispute centred on model control terms around autonomous weapons and domestic surveillance use.

    Two lessons for anyone building on a frontier model:

    • Federal and regulated ARR carries control-clause tail risk that never appears in ACV.
      🔗 Recommended Tool

      HubSpot’s CRM gives you the KPI dashboards to track every metric in this article. 30% recurring commission — your first referral pays you every month, forever.

      Start Free with HubSpot →

  • KPI Spotlight: Gross Margin — The Metric AI Quietly Broke in 2026

    KPI Spotlight: Gross Margin — The Metric AI Quietly Broke in 2026

    On September 27, 2026, OpenAI shipped GPT-6 Astra with an explicit AGI framing. On the same day, its companion tier — GPT-6 Sol & Luna — hit Product Hunt at half the price of prior frontier access. And in the background sits a number that every SaaS CFO should have taped to their monitor: OpenAI’s internal forecast of $278 billion in total burn through 2030, as surfaced via 20VC and SaaStr.

    Read those three facts together and you get the defining tension of SaaS unit economics right now. Frontier capability is deflating in price. The vendor’s cash requirement is inflating. Somewhere between those two curves sits your gross margin — and most boards are still modelling it as a static assumption inherited from 2019.

    This is a deep dive on gross margin: what it actually is, what good looks like in 2026, why vertical SaaS gets a structural haircut, where India sits, and the seven levers that move the number.

    Gross Margin, Defined Without the Hand-Waving

    Gross margin is revenue minus cost of revenue, divided by revenue. Simple formula, contested inputs. The entire analytical value of this KPI lives in what you’re honest enough to put in the numerator’s denominator — your COGS line.

    What belongs in SaaS COGS

    • Hosting and infrastructure — cloud compute, storage, egress, CDN.
    • Third-party model and API costs — inference, embeddings, vector storage, fine-tuning runs. This is the line item that did not meaningfully exist five years ago.
    • Customer support — the fully loaded cost of the humans who keep accounts functional.
    • Professional services and implementation — including the “free” onboarding you didn’t invoice.
    • Data and content licensing — enrichment feeds, market data, compliance databases.
    • Payment processing — if you take a cut of GMV, interchange lands here.
    • Amortised capitalised software — the portion of dev cost you capitalised last year and are now expensing.

    What does not belong: sales, marketing, R&D for future features, G&A. The most common reporting sin in Series A and B decks is quietly parking solutions engineers and implementation staff in S&M to inflate gross margin by 600–900 basis points. Diligence finds it. Every time.

    Benchmarks: What Good Looks Like in 2026

    A note on data integrity, because The KPI Hub flags rather than invents: our search layer returned zero indexed results across all fifteen queries today, so the ranges below are drawn from established public-company reporting and long-standing SaaS benchmark conventions, not from a fresh 2026 survey. Treat them as directional, not as gospel.

    • Horizontal, self-serve SaaS: 78–85% gross margin. Best-in-class public comps have historically sat at 80%+ on a non-GAAP basis.
    • Enterprise SaaS with heavy implementation: 70–78%. Services drag is real and often permanent.
    • Vertical SaaS with embedded payments: 45–65% blended. Not a failure — a different business.
    • AI-native products with per-query inference: 40–70%, with enormous variance by workload. This is the widest band in software, and the reason “AI SaaS” is an analytically useless category.

    The pay

    📊 Track These Metrics

    SEMrush gives you the competitive intelligence data behind these benchmarks. See exactly where you stand vs. competitors in real time.

    Try SEMrush Free →

  • India SaaS Brief — September 27, 2026: What GPT-6 Astra and Fresha’s $1B Tell Indian Founders

    India SaaS Brief — September 27, 2026: What GPT-6 Astra and Fresha’s $1B Tell Indian Founders

    Two headlines landed today that, read together, redraw the operating plan for Indian SaaS. OpenAI shipped GPT-6 Astra with explicit AGI framing. KKR put $80 million into Fresha at a $1 billion valuation — a beauty and wellness booking marketplace, not an AI-native anything. One signal says intelligence is getting cheaper. The other says capital is paying premium multiples for boring, vertical, transaction-attached software.

    For a Bengaluru or Chennai founder, that is not a contradiction. It is a playbook.

    Signal 1: Your AI COGS Just Became a 12-Month Contract, Not a 36-Month One

    The frontier launch came with a companion tier — GPT-6 Sol and Luna, which hit Product Hunt today at what the listing describes as “half the price” of prior frontier access. That is the number that matters more than the AGI marketing.

    Now cross-reference it with the other disclosure circulating today: OpenAI’s internal forecast reportedly puts total burn through 2030 at $278 billion (via 20VC and SaaStr commentary). So capability price is halving while the vendor’s cash requirement is compounding. Those two vectors cannot both hold indefinitely. Somebody eventually re-prices.

    The operational instruction for Indian SaaS companies is specific:

    • Cap AI vendor commitments at 12 months. Any multi-year inference contract signed today is a bet that today’s price is the floor. It is not. It is a promotional rate funded by a $278B burn plan.
    • Instrument gross margin per AI feature, not per product. If you cannot answer “what does this one workflow cost us in tokens per active account,” you cannot defend your gross margin in a diligence call.
    • Build a model-swap layer now. Abstraction between your application and your model provider is no longer architectural hygiene. It is pricing leverage.

    This matters disproportionately here. Indian SaaS has historically won on a capital-efficiency story — best-in-class companies out of India have repeatedly demonstrated that you can reach scale on materially less funding than a comparable US peer. AI inference costs are the single cleanest way to destroy that advantage. A horizontal AI wrapper with 55% gross margins is not an Indian SaaS company. It is a reseller with a Bengaluru address.

    The Anthropic Lesson: Revenue With Control Clauses Is Not Clean ARR

    Also today: Anthropic was designated a Pentagon supply-chain risk, a $200 million contract was terminated, and a reported $2 trillion IPO slipped to November. The break was over model control — specifically autonomous weapons and domestic surveillance applications. DoD moved to alternate vendors.

    Indian SaaS sells heavily into regulated buyers — BFSI, government, healthcare, and increasingly defence-adjacent enterprise. The transferable lesson is that usage-restriction clauses are a revenue risk that never appears in your ACV line. If a single clause can zero out a contract, that contract deserves a discount in your net revenue retention model, not a victory lap in your board deck. And a one-month IPO delay at a $2T ask tells you the public window in late 2026 is being managed, not opened. Plan your own liquidity timeline accordingly.

    Signal 2: Capital Is Rotating Into Verticals — And India Is Structurally Advantaged

    Fresha’s raise is the cleanest read on where money is moving. $80M from KKR at $1B for SMB booking software in beauty and wellness. No AI-native narrative. Just a vertical marketplace embedded in the daily operations of thousands of small businesses, with payments attached.

    Pair that with Lio’s $30M Series A led by a16z for AI enterprise procurement — specific function, specific buyer, specific workflow. Meanwhile TechCrunch reported this week that VCs are publicly publishing what they will no longer fund in AI SaaS. The pattern is unambiguous: horizontal AI wrappers are out, vertical workflow depth is in.

    Here is the India angle, and it is a real one. The Indian SMB market — salons, clinics, gyms, kirana, D2C brands, logistics operators — is one of the largest under-digitised commercial bases on earth, and it sits on top of the world’s best public payments and identity rails. UPI-native billing, Account Aggregator data access, ONDC distribution, GST-linked invoicing: these are infrastructure advantages a Fresha equivalent in Europe or the US simply does not have.

    What that enables:

    • Payments-attached monetisation. A vertical SaaS company with a UPI payment rail can monetise GMV, not just seats. That converts a ₹1,000/month subscription into a meaningfully larger ARPU without a price increase conversation.
    • Lower CAC via embedded distribution. ONDC and marketplace channel partnerships change acquisition economics for SMB SaaS in ways cold outbound never will.
    • Defensible data moats. Transaction data from a vertical you own is training data no frontier model has. That is the only durable AI moat left for an application-layer company.

    The uncomfortable corollary: the Indian SaaS companies most exposed right now are the ones selling thin horizontal AI layers into US mid-market. That is exactly the category capital is rotating out of, and exactly the category GPT-6 Astra’s capability jump commoditises.

    Signal 3: The Retention Math Nobody Models

    Jason Lemkin published a piece today on “The Keepers” — the employees you never let go of, even if you have to invent a role. His test was disarmingly simple: ask your manager what one thing you can take off her plate.

    This is not a soft topic in the Indian context. Senior AI and platform engineering talent in Bengaluru, Hyderabad, and Pune is being actively arbitraged by global capability centres and US remote offers. Replacement cost for a strong staff engineer runs multiples of an aggressive retention adjustment. Yet most Indian SaaS boards will approve a seven-figure sales hiring plan and litigate a 25% raise for the one person who holds

    📧 Track This Yourself

    The KPI Hub’s free KPI Monitor tracks the exact signals covered in this article — no spreadsheets, no guesswork.

    Start Free →

  • Benchmark Update — September 27, 2026: The Four KPIs That Just Got Repriced

    Benchmark Update — September 27, 2026: The Four KPIs That Just Got Repriced

    Benchmarks don’t move because someone publishes a new survey. They move because the underlying unit economics shift and the survey catches up nine months later. Today is one of those days.

    Three things landed on the same morning: OpenAI shipped GPT-6 Astra with explicit AGI framing while its companion tier launched at “half the price” of prior frontier access. Anthropic lost a $200M Pentagon contract over model-control clauses and pushed a $2T IPO to November. And Fresha, a beauty and wellness booking marketplace with zero AI-native positioning, raised $80M from KKR at a $1B valuation.

    Each one hits a different line of your KPI stack. Here’s the read.

    Data Integrity First

    The KPI Hub’s search layer returned zero indexed results across all 15 queries today. Everything below is built on primary-source headlines plus previously published benchmark data that we cite by source and vintage. Where we extrapolate, we say so. If a number isn’t attributed, treat it as directional — not decision-grade. That distinction is the entire product.

    CAC and CAC Payback: The Denominator Is Moving Faster Than the Numerator

    The standard reference points most SaaS operators still quote come from the 2024–2025 benchmark cycle: median CAC payback for private B2B SaaS clustering in the 18–24 month range, with best-in-class sub-12 and anything past 30 months flagged as structurally unhealthy (SaaS Capital and ICONIQ Growth survey cohorts, last published vintage). Those numbers were computed in a world where inference cost was a rounding error or a fixed line item.

    GPT-6 Sol & Luna launching at half the price of prior frontier access breaks that assumption in both directions.

    The good news

    If AI features are a material share of your gross margin, a 50% cut to frontier access is a direct gross-margin expansion event. CAC payback is calculated on gross-margin-adjusted revenue — so a two-point gross margin improvement shortens payback without you acquiring a single additional customer. If your AI COGS ran 8–12% of revenue and half of that is frontier-model spend, you just recovered roughly 200–300 basis points of gross margin. On a 22-month payback, that’s roughly a one-to-two month improvement. Inference, not a surveyed figure — run it against your own COGS split.

    The bad news

    OpenAI’s internal forecast puts total burn through 2030 at $278 billion (via 20VC x SaaStr). A vendor burning that much capital does not hold prices flat forever. The correct planning assumption is a 12-month AI COGS renegotiation cycle, not 36 months. If you’ve modelled three-year gross margin expansion off today’s inference pricing, your LTV is overstated.

    Practical revision: run two CAC payback scenarios — one at current inference pricing, one at a 40% cost increase on the same volume. If the second scenario pushes payback past 30 months, your AI feature set is a marketing expense wearing a product costume.

    Churn and NRR: Control Clauses Are the New Concentration Risk

    The Anthropic-Pentagon break is the most underrated KPI story of the quarter. A $200M contract died — not over price, performance, or procurement politics, but over model control terms: autonomous weapons and domestic surveillance use. DoD moved to alternate vendors.

    Net revenue retention benchmarks have been grinding down for three years. The last widely cited private-company medians sat around 102% for B2B SaaS overall, with enterprise-focused cohorts holding 105–110% and SMB cohorts frequently below 100% (SaaS Capital 2025 survey vintage). Public comparables tell the same story — Freshworks has reported net dollar retention in the 104–107% band across recent quarters, a long way from the 115%+ that defined 2021-era enterprise SaaS.

    What Anthropic adds is a new churn vector that doesn’t appear in any existing benchmark: values-based termination. Federal and large-enterprise ARR now carries a control-clause

    📊 Track These Metrics

    SEMrush gives you the competitive intelligence data behind these benchmarks. See exactly where you stand vs. competitors in real time.

    Try SEMrush Free →

  • KPI Spotlight: Gross Margin — The One SaaS Metric That Broke on September 23, 2026

    KPI Spotlight: Gross Margin — The One SaaS Metric That Broke on September 23, 2026

    Three frontier model releases landed inside 48 hours. OpenAI shipped GPT-6 “Astra” with AGI-onset claims, Anthropic followed with Claude Opus 5.5, and Meta dropped Muse Spark 1.3. Product teams are busy re-writing roadmaps. Finance teams should be doing something less glamorous and far more urgent: re-modelling gross margin.

    Because here’s the uncomfortable truth about 2026 SaaS. Revenue growth is still the headline metric investors quote. But gross margin is the metric that decides whether that growth is worth anything — and it is the metric most exposed to a frontier model launch. Every dollar of ARR you booked on GPT-5-era inference assumptions has a cost structure that just moved underneath it.

    This spotlight breaks down what gross margin actually means in an AI-heavy SaaS stack, where the 2026 benchmarks sit, and the seven levers that reliably move it.

    Definition: What Gross Margin Means When Your Product Thinks

    The formula hasn’t changed:

    Gross Margin % = (Revenue − Cost of Revenue) ÷ Revenue × 100

    What changed is the contents of Cost of Revenue. Classic SaaS COGS was a short, boring list: hosting, third-party licences, customer support, professional services delivery, payment processing, and the amortised cost of capitalised software.

    The 2026 COGS stack is longer and far more volatile:

    • Inference spend — per-token or per-request model API costs, the single most variable line item in modern SaaS
    • Retrieval and vector infrastructure — embedding generation, vector DB storage and query costs
    • Fine-tuning and evaluation — ongoing, not one-off, if you ship model updates
    • Guardrail and moderation layers — safety classifiers, PII redaction, audit logging
    • Agentic compute — sandboxed environments, browser instances, tool-call overhead
    • Human-in-the-loop review — the quiet margin killer in “autonomous” products

    If any of these sit in R&D or G&A on your P&L, your reported gross margin is fiction. The single most common accounting distortion we see in Indian and global SaaS decks right now is inference spend booked as R&D because it was “experimental” in 2024 and nobody reclassified it when it started serving production traffic.

    The two variants that actually matter

    Blended gross margin is what you report. Marginal gross margin — the margin on the next unit of usage — is what you manage. A company at 78% blended and 41% marginal is a company whose margin profile is deteriorating with every new logo, and no board deck will show it unless you build the view.

    Track a third: contribution margin per account, net of support and inference. That’s where you discover that your top-decile power users are your worst customers.

    2026 Benchmarks: Where the Lines Sit Now

    Benchmarks have bifurcated hard. Treat these as working bands, not gospel:

    • Traditional horizontal SaaS (seat-based, low AI dependency): 75–85% gross margin. Public comps hold here — Freshworks has reported GAAP gross margin in the low-to-mid 80s, and Zoho’s self-hosted infrastructure strategy keeps it in the same neighbourhood at far larger scale.
    • AI-native application layer: 50–70%. The best operators are clawing back toward 70% through routing and caching. The thin wrappers are sitting at 35–55% and calling it “growth investment.”
    • Vertical SaaS marketplaces: highly variable — 55–80% depending on payment-processing pass-through. Take-rate businesses carry structurally lower headline margin but far better retention.
    • Infrastructure and data-heavy SaaS: 60–75%.

    The investor bar has moved with it. TechCrunch reporting this week confirms VCs are now explicitly screening against thin-wrapper, no-

    📊 Track These Metrics

    SEMrush gives you the competitive intelligence data behind these benchmarks. See exactly where you stand vs. competitors in real time.

    Try SEMrush Free →

  • The 80% Gross Margin Era Is Officially Over: Inside ICONIQ’s September 19, 2026 Pacesetter Reset

    The 80% Gross Margin Era Is Officially Over: Inside ICONIQ’s September 19, 2026 Pacesetter Reset

    For roughly fifteen years, one number anchored every SaaS board deck, every Series B model, and every public comp sheet: 80% gross margin. It was the entry fee. Below it, you weren’t software — you were services with a login screen.

    That number died today.

    ICONIQ Growth’s updated Pacesetter Index — the benchmark set that tracks top-decile B2B software companies — now pegs the leading cohort of B2B+AI companies at 115% growth at $100M+ ARR, 55% gross margins, and $655,000 in revenue per employee. Read that middle number again. Fifty-five percent. That is 20 to 25 percentage points below the rule of thumb that has governed SaaS valuation logic since the Salesforce-era playbook was written.

    This is not a rounding error or a cohort quirk. It is a structural repricing of what software costs to deliver, and it lands in the same 48-hour window that OpenAI shipped GPT-6 Astra with an explicit AGI framing and Microsoft’s MAI-Transcribe-2 claimed simultaneous leadership on speed, cost, and accuracy in speech recognition. Frontier capability and commodity pricing arrived together. Your margin structure is caught in the middle.

    What Actually Changed: Inference Is Permanently in COGS

    The traditional SaaS gross margin math was beautiful because it was boring. You built the software once, hosting cost scaled sublinearly with usage, and every incremental customer dropped 80-85 cents of every revenue dollar straight to gross profit. Support and infrastructure were the only real drags.

    AI-native products broke that. Every query to a frontier model is a metered transaction against someone else’s GPU cluster. When a customer uses your product more, your cost of goods sold goes up, roughly linearly. That is the economics of a utility reseller, not a software licensor.

    The Pacesetter data confirms what operators have been whispering in Slack channels for eighteen months: the fastest-growing AI software companies are deliberately buying growth with margin. They are not failing to optimize. They are choosing 115% growth at 55% GM over 40% growth at 78% GM — and the capital markets, for now, are rewarding it.

    Three implications land immediately:

    • Your LTV/CAC model is broken if it uses a static gross margin assumption. Contribution margin per customer now varies with usage intensity. A power user can be a negative-margin user.
    • Rule of 40 needs a variant. Growth plus FCF margin was designed for a world where gross margin was a constant. When GM itself is a strategic variable, you need to track growth + GM + FCF as a triangle, not a sum.
    • Pricing must be usage-linked or you are short volatility. Pure seat-based pricing on an inference-heavy product is a structural short position against your own customers’ engagement.

    The Metric VCs Are Actually Underwriting: $655K Revenue Per Employee

    Buried in the Pacesetter triangle is the number that should reshape your 2027 hiring plan. $655,000 in revenue per employee at the top decile is a dramatic move from the $200K-$250K RPE that characterized “efficient” SaaS in the 2019-2021 cohort.

    This is the compensating mechanism. If gross margin compresses 25 points, the only way to protect operating leverage is to radically shrink headcount per dollar of revenue. AI companies are doing exactly that — using the same models that inflate their COGS to deflate their opex.

    Practically, RPE is now a term-sheet gate. If you are running a $20M ARR company with 150 employees ($133K RPE), you are not in the conversation regardless of your growth rate. The efficient frontier has moved, and it moved fast.

    The Second Signal: Retention Replaced Growth as the Diligence Anchor

    Today’s VC sentiment wire carried a parallel story: what investors have stopped funding. The kill list is specific — thin wrappers over frontier models, companies with no proprietary data asset, and anything that cannot produce hard net revenue retention evidence across at least four quarters.

    MAI-Transcribe-2 is the perfect illustration of why. A transcription-dependent product that built its pricing on OpenAI Whisper economics now faces a competitor claiming better accuracy at lower cost. Expect 20-40% price compression pressure flowing through any transcription-heavy COGS line this quarter. If your product’s differentiation was “we call the ASR API and add a nice UI,” your moat evaporated on a Tuesday.

    The counter-example also shipped today. Fresha, the beauty and wellness booking marketplace, raised $80 million from KKR’s Next Generation Technology fund at a $1 billion valuation. Fresha is not an AI story. It is a take-rate story — a vertical marketplace embedded in the payment flow of salons and spas, where the data asset is transaction history that no frontier model can replicate. Lio, the AI procurement startup, raised a $30M Series A led by a16z on similar logic: agentic workflow automation sitting on top of proprietary spend data.

    The pattern is unmissable. Capital is flowing toward owned transaction data and take-rate economics, and away from inference-arbitrage businesses. Fresha at $1B and Lio at $30M Series A are the same trade expressed at different stages.

    The India Angle: Why 55% GM Hits Different Here

    Today’s wire carried no India-native headlines — a genuine coverage gap worth flagging. But the Pacesetter reset has sharper consequences for Indian SaaS than for Bay Area peers, for three reasons.

    First, the India SaaS thesis was built on capital efficiency. Zoho, Freshworks, and Zoho-adjacent bootstrapped companies sold investors on high gross margins plus low-cost engineering — an 80% GM business with Chennai or Bengaluru salary structures produced superior unit economics to anything in SOMA. If GM compresses to 55% globally, the labor arbitrage covers a smaller share of the gap. The moat narrows.

    Second, RPE is where Indian SaaS is structurally disadvantaged on the reported number. A company billing in USD while pa

    🔗 Recommended Tool

    HubSpot’s CRM gives you the KPI dashboards to track every metric in this article. 30% recurring commission — your first referral pays you every month, forever.

    Start Free with HubSpot →

  • KPI Spotlight: Gross Margin — The Metric That Decides Who Survives the AI Price War (September 15, 2026)

    KPI Spotlight: Gross Margin — The Metric That Decides Who Survives the AI Price War (September 15, 2026)

    Three things happened today that all point at the same line item on your P&L. OpenAI shipped GPT-6 Astra and declared the AGI era open. Microsoft AI released MAI-Transcribe-2, undercutting OpenAI, Google, and ElevenLabs on both price and latency. And VCs told TechCrunch, on the record, that they will no longer fund thin AI SaaS layers.

    None of those are growth stories. They’re all gross margin stories.

    For fifteen years, gross margin was the boring KPI in SaaS — the one you assumed was 80% and stopped thinking about. Software had near-zero marginal cost, so the interesting questions were all above the line: CAC, NDR, magic number, payback. That assumption is now dead. When your product calls a model on every user action, you have variable cost of goods sold for the first time since the on-prem era. You are, structurally, closer to Twilio than to Atlassian.

    This is today’s KPI Spotlight: what gross margin actually measures, where the 2026 benchmarks sit, and the seven levers that move it.

    Definition: What Gross Margin Actually Measures

    Gross margin is the percentage of revenue left after the direct cost of delivering the product:

    Gross Margin % = (Revenue − Cost of Goods Sold) ÷ Revenue

    The formula is trivial. The fight is over what goes into COGS. Most SaaS companies understate COGS — sometimes accidentally, often not — and report a gross margin that flatters the business by 10 to 20 points.

    What belongs in SaaS COGS

    • Hosting and infrastructure — cloud compute, storage, egress, CDN
    • Model inference and third-party AI APIs — the line that didn’t exist in 2019 and now dominates for AI-native products
    • Third-party data and licensing — enrichment providers, market data, map APIs
    • Customer support and success — the portion tied to service delivery, not upsell
    • Professional services and implementation — including solutions engineers doing deployment work
    • DevOps/SRE — the humans keeping the service up
    • Payment processing — critical if you touch transactions

    What does not

    • R&D and product engineering
    • Sales and marketing
    • G&A
    • Pre-sales demo engineering (debatable, but most classify it in S&M)

    If your reported gross margin is 82% and you cannot produce a COGS breakdown that includes inference spend, support headcount allocation, and payment fees, you don’t have a gross margin. You have a slide.

    Benchmarks: Where the Bar Sits in 2026

    Gross margin benchmarks now fragment by architecture, not by stage. Grouping “SaaS” into one band stopped being useful the moment inference entered COGS.

    Tier 1 — Classic seat-based SaaS: 78–85%

    Pure-play subscription software with predictable infrastructure load. Public comparables cluster in the high-70s to low-80s — Salesforce and Zoom have both operated in the mid-to-high 70s at scale. If you sell seats and you’re below 75%, you have a support-cost or hosting-efficiency problem, not a market problem.

    Tier 2 — Consumption and data

    📊 Track These Metrics

    SEMrush gives you the competitive intelligence data behind these benchmarks. See exactly where you stand vs. competitors in real time.

    Try SEMrush Free →