The Golden Quadrant

The opening scene

Two AI products. Same workload type — automated handling of a customer-service queue. Same underlying model class. Comparable resolution quality, audited by the same vendor-neutral evaluator. Comparable cost structure per inference. Both are sold to mid-market and enterprise buyers. Both have been in market for over a year.

Product A charges $0.10 per action. Each time the agent handles a ticket — whether that means a successful resolution, a partial deflection, or a hand-off — the customer pays ten cents. Across a year of operation, Product A captures roughly 3% of the value the customer measurably derives from the deployment. That value is mostly avoided support cost. Three pence on the pound.

Product B charges 25% of measurable value generated. The contract specifies what counts as resolved, what counts as inactivity, what triggers a billable outcome, and what doesn’t. Across a year of operation, Product B captures roughly 25% of the value the customer derives from the same workload, on the same model class, at the same cost. Twenty-five pence on the pound.

Same workload. Same quality. Same cost structure. Eight-times revenue gap.

The first PM is operating in what Aakash Gupta — in his Feb 25 2026 piece How to Price AI Products: The Complete Guide for PMs — calls the Token Quadrant. The second PM has reached what Aakash calls the Golden Quadrant.

The eight-times gap is structural. It is not a sales-execution gap, a product-quality gap, or a model gap. It is a position gap on the autonomy-and-attribution matrix. And once you understand how that matrix works, you stop being surprised that it exists. You start being surprised when it doesn’t.

This post is about the Golden Quadrant — the structural position at the intersection of high autonomy and high attribution where outcome pricing captures 20–60% of the value the AI generates. It is the apex move of the entire pricing architecture this series has developed across L1-T08, L1-T09, L2-T04, and L2-T09. It is also the structural pivot in how software companies make money — what Phil Fersht of HFS Research and Foundation Capital have started calling Service-as-Software.

If you ship AI products in 2026, the Golden Quadrant is the position you are either deliberately walking toward or accidentally walking away from. There is no neutral.

The central frame

That is the whole post in one sentence. The rest is mechanics, examples, and the hill-climbing playbook for getting there.

The 60-second answer — the Autonomy + Attribution Matrix

Aakash Gupta maps 50+ AI startups across two axes, surfaces six pricing patterns, and provides the decision tree most 2026 AI PMs now use to position their pricing. Internalise the matrix and most pricing arguments resolve themselves.

X-axis: Autonomy. Does the AI act unilaterally on outcomes, or does it require human review per action? At the low end, every action gets reviewed before it ships. At the high end, the agent owns the outcome end-to-end, with humans involved only on exceptions or appeals.

Y-axis: Attribution. Can the value the AI created be cleanly attributed to it, in a way the customer accepts? At the low end, the AI is one of many contributors and nobody can isolate its impact. At the high end, every dollar of avoided cost or generated revenue traces back to a specific AI action through an attribution chain the customer’s finance team has signed off on.

Plot any AI product on those two axes and you land in one of four quadrants:

• Bottom-left (low autonomy, low attribution): Token pricing. You charge per call, per token, per inference. The customer is paying for access. They cannot prove the AI created value, and the AI cannot act independently anyway. Capture: roughly 1–3% of the value generated. This is where API-layer providers live by structural necessity. It is also where many product companies end up by default, because it is the easiest model to sell.
• Bottom-right (low autonomy, high attribution): Hybrid or per-action pricing. The AI is still copilot. Each action is countable. Value is measurable. But because outcomes are owned by humans, pricing has to follow actions rather than outcomes. Capture: roughly 5–15% of the value generated. This is a respectable quadrant. It is not the apex.
• Top-left (high autonomy, low attribution): Per-seat pricing surviving on incumbency. The AI acts unilaterally — but the customer cannot prove it created the outcome they paid for. So pricing reverts to the most familiar SaaS model: a seat licence. Capture: starts moderate, then erodes every quarter as the customer realises the AI is doing the work but the seat count is still tied to the people who used to do it. Renewals compress.
• Top-right (high autonomy, high attribution): The Golden Quadrant. Outcome-based pricing. The AI acts on outcomes. The value is measurable, attributable, and accepted by the customer’s finance team. Pricing maps directly to the outcomes delivered. Capture: 20–60% of the value generated.

Reaching the Golden Quadrant requires both axes. High autonomy alone keeps you in the top-left. High attribution alone keeps you in the bottom-right. Most teams have one and assume they have the other. Most don’t.

Why most teams stay in the bottom-left

The bottom-left quadrant is the path of least friction. The reasons most teams default into it are not strategic. They are operational.

Familiarity. Token pricing is what the underlying model providers charge. If your cost model is per-token and your pricing model is per-token, the maths are easy and the margin is visible. The CFO is comfortable. The sales team has a number to quote. The procurement team recognises the structure.

Sales-cycle ease. Per-seat and per-token contracts close faster. Procurement has approved templates. Legal has standard redlines. The buyer’s PO system has a line item for software seats and one for usage-based services. Outcome contracts require new templates, new approvals, and finance signing off on a definition of value they have never validated before.

Procurement comfort. Most enterprise procurement teams in 2026 are still running playbooks written for SaaS. They want a unit price, a quantity, and a discount schedule. They do not want a percentage of measurable value, a definition of what “resolved” means, and an attribution chain that requires their data warehouse to plug into your platform. The first contract takes six months instead of six weeks. Many sales teams give up.

Low operational infrastructure. Token pricing requires almost nothing operationally — a meter, an invoice, a billing system. Outcome pricing requires the operational definition discipline from L2-T09 (the eight-or-nine-word rule), the attribution chain from L1-T10 (Measurement × Adoption = ROI), the dispute-resolution process, and the customer-success motion that defends the contract through quarterly business reviews. The work is real. The infrastructure is real. Many teams do not staff for it.

So the path of least friction wins, and the bottom-left quadrant fills with products that should be elsewhere. The PMs who reach the Golden Quadrant do so deliberately. They invest in the infrastructure that the quadrant demands, because they understand that the infrastructure is the moat.

The Service-as-Software pivot

Now zoom out. The Golden Quadrant is the apex of an old game. The bigger story is that AI is changing the game itself.

Phil Fersht at HFS Research coined the term Services-as-Software to describe the shift from people-heavy projects and static software toward AI-driven, outcome-focused services delivered like software. The phrasing is deliberate. It inverts the SaaS framing. SaaS was Software-as-a-Service — software delivered the way services used to be: subscription-based, hosted, continuously updated. Services-as-Software is the reverse: services delivered the way software is — productised, instrumented, continuously improving, and sold against the value created.

Foundation Capital has built the strongest argument for the size of this shift. Their 2024 essay (updated through 2026), A System of Agents Brings Service-as-Software to Life, makes the structural case in one number.

Software has historically priced into IT budgets. Globally, that pool is roughly $300–500B. Service-as-Software prices into labour budgets. Globally, that pool is roughly $4.6T. The TAM expansion is roughly 10×. AI-native companies that position into the labour budget capture share of a much larger pool than companies that position into the IT budget.

The buyer is not buying access to a tool. They are hiring a worker who happens to be code. They evaluate it the way they evaluate a worker — on outcomes delivered. They pay it the way they pay a worker — on outcomes delivered. They renew it the way they renew a worker — on outcomes delivered.

The pricing model that lives at this position is outcome-based, because no other model fits the buying motion. You do not pay a worker per keystroke. You do not pay them per minute they are logged in. You pay them for the work they completed.

This is why the Golden Quadrant matters so much in 2026. It is not just a better pricing position inside the old game. It is the only pricing position that is structurally compatible with the new game. The companies that get there are positioning themselves for a 10× TAM expansion. The companies that stay in the bottom-left are competing for share of a budget pool that is, in real terms, declining as workloads shift across the line.

This is what Foundation Capital means when they call Service-as-Software the trillion-dollar reframing of the SaaS thesis. The capture position is structurally larger.

The four 2026 cases

Four products that demonstrate where the Golden Quadrant lives — or where its borders are.

1. Intercom Fin — the canonical Golden Quadrant case

Fin charges $0.99 per resolved outcome, billed only on success-or-no-follow-up (the 24-hour rule). It runs in production at tens of millions of ARR. Customer cohorts report 80%+ support volume handled and 67%+ resolution rates on the workloads where it is deployed. Pricing maps directly to value captured — avoided support cost. Autonomy is high: the agent owns the resolution end-to-end. Attribution is high: the resolution definition is operational, the no-follow-up rule is contractual, and the customer’s finance team can verify the count. Cross-link L2-T04 and L2-T09. Fin sits squarely in the Golden Quadrant.

2. Zendesk Automated Resolutions — adjacent, not apex

Zendesk charges $1.50 (committed) or $2.00 (pay-as-you-go) per Automated Resolution, with the verified 72-hour inactivity rule defining when a resolution counts. Attribution is high — the AR definition is contractual and operational. Autonomy is high but slightly lower than Fin, because Zendesk’s product surface integrates with a wider set of human workflows where the agent more frequently hands off. The result: high attribution, slightly less unilateral autonomy. Strong quadrant. Just not the apex. Cross-link L2-T04.

3. The XPO Logistics framework — internal Golden Quadrant logic

XPO is not selling AI to external customers, but the internal application of the same discipline is instructive. The framework maps AI output directly to income-statement lines: 80% diversion reduction, $29M per efficiency point. The same attribution chain that makes external outcome pricing defensible also makes internal AI investment defensible to a CFO. Cross-link L1-T10 and L3-T09. The lesson: the discipline is the same whether the buyer is external or internal. Without operational definition, attribution chain, and dispute process, no outcome claim survives.

4. Service-as-Software-shaped startups in vertical markets

Fraud detection paid as a percentage of recovered loss. Lead qualification paid per qualified meeting. Pipeline acceleration paid as a percentage of generated revenue. None of these need a single seat licence. None of them sell access. All of them sell outcomes. Each one is positioned into the customer’s labour budget rather than their IT budget. The companies that can reach this pricing position capture value at the scale of what their customers used to pay people to do.

The pattern across all four: operational discipline upstream of pricing. Every Golden Quadrant case has invested heavily in the definition, measurement, and attribution infrastructure before claiming the pricing position. The pricing model is the visible result. The infrastructure is the cause.

The Aakash Gupta matrix as the canonical 2026 PM frame

If there is one published artefact that defines the 2026 AI pricing conversation at the PM level, it is Aakash’s matrix. How to Price AI Products: The Complete Guide for PMs maps 50+ AI startups across the autonomy and attribution axes, identifies six recurring pricing patterns, and provides the decision tree that most 2026 AI PMs are now using to position their pricing.

The matrix has three properties that make it the canonical frame.

It is empirical. It is built from the actual pricing pages of 50+ AI companies, not from a theoretical model. The quadrants describe positions that real companies occupy. The patterns are observed, not invented.

It is decisive. The decision tree gives the PM a single recommendation, not a menu. Given your autonomy and attribution position today, here is the pricing model you should use. If you want to move to a different pricing model, here is the position you need to reach first.

It is hill-climbable. The matrix does not demand a leap from the Token Quadrant to the Golden Quadrant. It accepts that most teams are not at the apex today. It identifies the intermediate positions and the moves between them. Hill-climbing is the operational metaphor — not because pricing is incremental, but because most teams cannot skip stages successfully.

Print the matrix. Pin it above your desk. Walk through it before every pricing review. The PM who carries the matrix into a strategy meeting carries the only frame in the room that is calibrated to 2026 reality.

The hill-climbing approach to reaching the Golden Quadrant

You almost certainly cannot get from where you are today to the Golden Quadrant in one move. The hill-climb has four stages, and the order matters.

Stage 1 — Establish attribution

Before you change the pricing model, build the measurement infrastructure. Without attribution, any outcome-pricing claim becomes a legal dispute. The customer’s finance team will challenge the count, the definition, and the chain from action to outcome. If you cannot survive that challenge in writing, you do not have a pricing position. You have an aspiration. Stage 1 is engineering work — instrumentation, eval suites, the operational definition discipline from L2-T09, the value model from L1-T10. It is unglamorous. It is mandatory.

Stage 2 — Increase autonomy with measurement intact

Move the product from copilot to agent (cross-link L2-T03) without losing the attribution chain you built in Stage 1. This stage is where most teams break. They increase autonomy by removing human review steps — and the attribution chain breaks because the human was the one writing the resolution into the ticket system. The autonomy move and the attribution preservation have to happen together. If you have to choose, attribution wins. Lose attribution and you fall back into the top-left quadrant — high autonomy, no proof.

Stage 3 — Move from per-action to per-outcome

Once both autonomy and attribution are real, the pricing model can shift from per-action to per-outcome. This is the contractual move that most teams associate with the Golden Quadrant. It is actually Stage 3 of four. The contract template changes. The procurement conversation changes. The customer-success motion changes — every QBR is now a defence of the outcome count.

Stage 4 — Move from per-outcome to value-share

The apex of the Golden Quadrant. Pricing as a percentage of measurable value delivered, not as a fixed price per outcome. This is where Foundation Capital’s labour-budget framing becomes literal. You are not charging per resolution; you are charging a percentage of the labour cost the customer would have paid otherwise. This stage demands the highest infrastructure: a P&L attribution chain, finance-team alignment, contractual share definitions, and an audit-friendly process for the percentage. The capture is also the highest. Companies that reach Stage 4 are the ones positioned for the 10× TAM expansion.

The teams that try to skip stages — particularly Stage 1 — fail predictably. The skip is usually motivated by a sales-cycle pressure (“the customer wants outcome pricing, let’s just write it into the SOW”) and the failure is usually six months later when the first quarterly review surfaces a count the customer disputes. By then the contract is open, the relationship is strained, and the next renewal is at risk. Hill-climbing is slower in the first quarter and faster every quarter after.

The McKinsey productivity range and the Golden Quadrant maths

There is one external data point that makes the Golden Quadrant maths structurally large.

McKinsey’s analysis of agentic workflows — the range circulated widely in 2025–2026 industry commentary — finds productivity gains in the 20–60% range when AI agents are integrated into existing workflows, with decision-speed improvements around 30%. This is not the AI doing 20–60% of the work. This is the labour cost reduction in the workflow when AI is added.

Now compose that with the Golden Quadrant pricing capture. The AI generates 20–60% productivity gain on labour cost. The Golden Quadrant pricing model captures 20–60% of that gain. The two compose multiplicatively, and the dollar capture is, in effect, a percentage of a percentage of the customer’s labour budget.

For a workload where the customer’s labour cost is, say, £10M annually, a 40% productivity gain is £4M of value. A 25% capture rate on that value is £1M of revenue per customer per year — on a single workload. That is why Service-as-Software companies can grow into multi-billion-dollar businesses on outcome pricing alone. The maths is structurally large because both percentages are structurally large.

This is also why the bottom-left quadrant is so punishing. A 1–3% capture on the same £4M of value is £40–120K. Same workload. Same value generated. Twenty- to fifty-times revenue gap. The eight-times gap in the opening scene was the conservative version.

Trap / Fix — the four PM mistakes in pursuing the Golden Quadrant

The closing argument — the Golden Quadrant as a structural moat

Every quarter the system runs after a company reaches the Golden Quadrant, the gap to competitors widens.

The measurement infrastructure becomes a moat. Customers integrate their data warehouses with yours; switching costs rise. The attribution discipline becomes a moat. The customer-success team has years of dispute-resolution patterns; new entrants are starting from scratch. The outcome-definition library becomes a moat. Hundreds of edge cases — what counts as resolution in a partial-refund scenario, how a re-opened ticket affects the count, what happens when the customer’s CRM and your platform disagree on the timestamp — are documented, contractualised, and battle-tested. The customer-success motion that has learned to defend outcome-priced contracts through quarterly reviews becomes a moat. None of these are model-layer moats. None are vendor-layer moats. They are operational moats. Operational moats compound.

This is why Foundation Capital frames Service-as-Software as the trillion-dollar reframing of the SaaS thesis. It is not a different SaaS. It is a structurally larger capture position, defended by operational infrastructure that takes years to build and cannot be acquired.

The PMs who get there in 2026–2027 will be talked about as the ones who saw the structural shift and built into it. The PMs who stayed in the bottom-left will be talked about as the ones who shipped good products and watched the value go to someone else. Same model class. Same product quality. Same cost structure. Different position on the matrix.

Remember this

1. Autonomy × Attribution Matrix. Four quadrants. Golden (high autonomy + clean attribution), Operational, Copilot, Tool. The Golden Quadrant is the strategic position; the others have ceilings.
2. Golden Quadrant = outcome pricing at scale + Service-as-Software pivot. Customer-aligned, margin-compounding, competitively superior. The other quadrants top out at per-seat or tiered usage.
3. SaaSpocalypse repricing is real and mandatory in the Golden Quadrant. Per-seat on high-autonomy + clean-attribution use cases is structural revenue-capture failure — 3–10× left on the table.
4. Intercom Fin is the public proof point. $0.99 per resolution → $343M ARR with 393% Q1 growth. Per-seat competitors could not match the trajectory because the pricing model itself is the moat.
5. Pricing-led transformation is the highest-leverage Director-level intervention. Pricing redesign pulls sales, marketing, customer success, and engineering into alignment in a way no roadmap memo can.

In practice

A six-step playbook for moving a portfolio toward the Golden Quadrant. Run it as a quarterly artefact, not a one-off exercise.

Step 1 — Score every product on the Autonomy × Attribution Matrix. Be honest. Most teams place themselves higher than the customer’s finance team would. The quadrant placement determines which strategic moves are even available.

Step 2 — Identify the Golden Quadrant candidates. The products that sit in — or can credibly move to — high autonomy plus clean attribution. Plan the outcome-pricing transition for these specifically; leave the others on tiered usage or per-seat.

Step 3 — Plan the Service-as-Software pivot for the most autonomous use cases. Pricing redesign, sales motion (sell against human-staffed alternatives, not against software), customer success measurement (customer outcomes, not adoption), margin structure (gross margin scales with AI quality).

Step 4 — Run the SaaSpocalypse repricing on Golden Quadrant products. The L2-T09 outcome-based pricing playbook is the operational guide. T07 supplies the strategic frame; L2-T09 supplies the mechanics.

Step 5 — Drive pricing-led organisational alignment. Pricing redesign requires sales, marketing, customer success, and engineering alignment. The PM owns the alignment conversation — not finance, not sales ops. Without organisational alignment, pricing redesign becomes an isolated finance project that ships and stalls.

Step 6 — Translate Golden Quadrant moves for the executive layer. The L2-T07 stakeholder translation discipline. CFO hears margin compounding; CRO hears customer-aligned ARR; CEO hears structural moat; board hears the multi-year capture story. One translation per stakeholder, sourced from the same matrix.