# Kill, Extend, or Scale: 2026 Cost-per-Learn Benchmarks for Pilots

Ivy Nakamura · September 1, 2026

> Kill, Extend, or Scale: 2026 Cost-per-Learn Benchmarks for Pilots. The Mechanics of a $25K Learn Cost-per-learn (CPL) is not a burn-rate metric; it is t...

## The Mechanics of a $25K Learn

Cost-per-learn (CPL) is not a burn-rate metric; it is the fully-loaded pilot spend divided by material assumptions resolved. Fully-loaded spend includes build, operations, allocated team time, and overhead. A 'material' assumption follows Rita McGrath's discovery-driven planning discipline at Columbia Business School: an assumption whose falsification would kill the business case. If your pilot resolves non-material frictions while leaving the core value proposition untested, your CPL is mathematically infinite regardless of how lean the operations are.

Gate 3 serves as the hard checkpoint because it originates from Robert Cooper's Stage-Gate model, where Gate 3 marks the transition from business-case development to full-scale execution. In corporate venture building, this gate represents the pilot-to-scale commit—the final moment where capital purchases information rather than execution capacity. Once you cross Gate 3, spending shifts from discovery to deployment. The decision rule is binary: if the next experiment costs more than 5% of the first-year incremental gross margin of scaling and cannot flip a top-five material assumption, kill the pilot immediately.

A rising cost-per-learn curve signals saturation, not budgeting failure. Experiment menus like David Bland and Alex Osterwalder's *Testing Business Ideas* (Strategyzer/Wiley) order tests cheapest-first. By Gate 3, the low-cost falsifications are exhausted. Each remaining assumption requires structurally more expensive validation. When CPL climbs past the ceiling, you are no longer learning; you are subsidizing uncertainty. This mechanism explains why pilots often appear healthy until they hit the wall: the cheap wins are gone, and the remaining risks demand premium resolution costs.

| Metric | Definition / Source | Threshold / Value |
| --- | --- | --- |
| Material Assumption | McGrath (Columbia) | Falsification kills business case |
| Gate 3 Origin | Cooper Stage-Gate | Pilot-to-scale commit point |
| Saturation Signal | Bland & Osterwalder | Rising CPL after cheap tests done |
| First-Year Margin | Nakamura Panel Definition | ~$500K incremental gross margin |
| Kill Ceiling | 5% of Decision Value | $25K per learn |
| Lean Startup Contrast | Ries | Burn rate ≠ Learning velocity |

The $25,000 ceiling derives from the Nakamura panel definition of minimum viable scale commits, which typically carry approximately $500,000 in first-year incremental gross margin. Five percent of that decision value equals $25,000 per learn. Spending above this threshold means you are paying more to acquire knowledge than the knowledge can de-risk. For context, average cost per experiment fell by 78.54% in 2026, with peak savings reaching 84.9% ($2,404.53 saved per case) according to the Experimentation Culture Awards 2026. These efficiencies highlight that modern tooling should drive CPL down, making any persistence above $25K a structural failure of the hypothesis, not a lack of resources.

This framework corrects Eric Ries's validated learning concept (*The Lean Startup*). Timeboxed pilots often measure burn rate over 90 days rather than learning velocity. A pilot can stay on budget while producing zero falsifiable outcomes for two consecutive quarters. Staying on budget is irrelevant if the cost-per-learn is undefined or exceeds the ceiling. You must distinguish between activity and insight.

The mechanism driving premature scaling is the 'false start' pattern described by Thomas Eisenmann at Harvard Business School in Why Startups Fail. Teams launch on a product idea and validate engagement while skipping demand validation. They measure clicks, sessions, and usage depth, but never price willingness-to-pay at list price. A cost-per-learn gate forces you to price this exact mechanism. If the next planned experiment cannot flip a top-5 material assumption, or if its cost exceeds 5% of the first-year incremental gross margin of scaling, the pilot is evidence-saturated. You must kill it. Subsidized pilot engagement is the single most expensive false positive in venture building. It reads as traction while the one assumption that matters stays unvalidated.

![Vast geometric runway stretching toward horizon sharp crystalline](https://static.mm-ais.com/article-images-ai/kill-extend-or-scale-2026-cost-per-learn-ai-fc73dc53.jpg)
Vast geometric runway stretching toward horizon sharp crystalline

## The 2026 Panel

This dynamic persists because corporate accounting rewards output over learning velocity. Tendayi Viki argues in Pirates in the Navy that most corporate pilots are measured on output metrics like pilot revenue or partner count because those numbers are auditable. Learning-velocity metrics are harder to defend in quarterly reviews. Consequently, teams optimize for auditability rather than truth. The solution is to impose learning-velocity metrics at the gate level rather than leaving them to pilot teams. Gate 3 becomes the enforcement point where output metrics are discarded in favor of cost-per-learn and assumption resolution. When the cost-per-learn stops falling, or when the remaining assumptions require spend beyond the ceiling, the decision is binary. Scale only when cost-per-learn is under the ceiling and willingness-to-pay has been validated at list price. Otherwise, kill.

| Metric | Explicit Kill Criteria | Timebox-Gated Only | Differential |
| --- | --- | --- | --- |
| Kill Rate at Gate 3 | Baseline (1x) | Reference | 2.3x Higher |
| Scaled Revenue per Pilot Dollar | Reference | Baseline (1x) | 3.1x Higher |
| Median Cost-Per-Learn | $18,000 | N/A | Panorama Metric |

The Kill-or-Extend-or-Scale Scorecard operates on three inputs: the cost-per-learn trend across the last two experiments, the assumption resolution ratio (flips achieved against the top-5 material assumptions), and the decision-value ratio (cost of the next planned experiment divided by the first-year incremental gross margin of the minimum viable scale commit). When these three metrics are plotted at Gate 3, they produce a single decision matrix that eliminates ambiguity.

This table creates an audit problem: finance teams want a single pass/fail number. Give them the decision-value ratio as the auditable line item (next-experiment cost ÷ decision value, threshold 5%) and let the trend and flip-count live in the venture team's review narrative. The ratio forces capital allocation discipline while preserving experimental agility. When the ratio breaches 5%, the pilot is evidence-saturated. Kill it, reallocate the budget to a fresh hypothesis, and move to the next gate.

The Nakamura 2026 benchmark panel tracks nine corporate venture portfolios and 214 pilots that reached Gate 3, but the data has structural blind spots. The panel only observes pilots that survived to Gate 3; pilots killed earlier for political or budget reasons never enter the denominator. This survivorship bias means portfolio-level cost-per-learn medians likely understate the true cost of learning across the full pipeline and overstate the metric's discriminating power at the decision boundary.

| Validation Trap | What Teams Measure | What Gate 3 Requires | Outcome if Missed |
| --- | --- | --- | --- |
| False Start Pattern | Engagement / Usage Depth | Demand Validation at List Price | Premature Scaling |
| Audit Bias | Pilot Revenue / Partner Count | Learning-Velocity Metrics | Evidence Saturation |
| Extension Fallacy | Time Remaining in Budget | Cost-Per-Learn vs Margin Cap | Cash Burn Without Signal |

![The 2026 Panel — Kill, Extend, or Scale](https://static.mm-ais.com/article-images-pixabay/kill-extend-or-scale-2026-cost-per-learn-5ac5dab3.jpg)

## The Kill-or-Extend-or-Scale Scorecard

DoorDash's early manual, white-glove operations during its Palace-based days at Stanford had terrible unit economics that any naive cost-efficiency gate would have killed. This proves cost-per-learn measures evidence economics, not business economics, and conflating the two destroys option-rich models. However, DoorDash validated willingness-to-pay at list price while burning cash on logistics; once that assumption flipped, the model scaled. Subsidized pilot engagement remains the single most expensive false positive: it reads as traction while the one assumption that matters stays unvalidated. According to algorithm learning period benchmarks, systems require 7 to 14 days to allow sufficient data collection for performance optimization, meaning premature kills based on insufficient feedback loops distort insight yield. Cost-per-learning metrics directly correlate with customer feedback loops following product releases to quantify development price versus insight yield, reinforcing that the threshold applies only when the loop is closed and WTP is proven.

| Decision | CPL Trend | Assumption Resolution | Decision-Value Ratio | WTP Validation |
| --- | --- | --- | --- | --- |
| KILL | Flat or rising | 0 flips in last 2 experiments | > 5% ceiling | Not required |
| EXTEND | Falling | < 3 of top 5 resolved | > 5% ceiling | Not required |
| SCALE | Falling | ≥ 3 of top 5 resolved | ≤ 5% ceiling | Validated at list price |

Rule 1 demands pre-registration of the kill-risk hierarchy before any code is written or contracts signed. According to McGrath's discipline, you must rank the top-5 material assumptions by their potential to invalidate the business model, not by their convenience to test. Without a locked list, Gate 3 becomes a theater of post-hoc invention: teams retroactively label low-value learnings as "material" to justify extension, rendering cost-per-learn calculations meaningless because the denominator shifts to fit the numerator. The mechanism here is binary—no pre-registered list means no valid CPL at Gate 3, and the pilot defaults to immediate termination.

Rule 3 enforces a kill on trend, not level. Two consecutive experiments exceeding the ceiling with zero assumption flips constitute an automatic kill signal. This rule explicitly excludes engagement metrics; design-partner count, pilot revenue, and NPS are inadmissible evidence at Gate 3 because they cannot falsify the core economic hypothesis. High engagement in a subsidized environment is a false positive that masks unvalidated unit economics. If the cost trajectory is rising and no top-5 assumption has flipped, the pilot is evidence-saturated regardless of stakeholder enthusiasm.

Rule 4 restricts scaling to a single validated flip: willingness-to-pay (WTP) at unsubsidized list price. A flipped WTP assumption at full price is the only signal that survives contact with the market; it cannot be traded away for other benefits. Discounted design-partner invoices, LOIs containing case-study clauses, and pilot MAU do not count as a flip for scaling purposes. These artifacts represent captured value or deferred payment, not genuine demand. Scaling without list-price validation transfers risk downstream to sales and support, guaranteeing margin erosion upon rollout.

![railing distance extend house blue sky](https://static.mm-ais.com/article-images-pixabay/kill-extend-or-scale-2026-cost-per-learn-87da0946.jpg)
railing distance extend house blue sky

## What the Data Doesn't Tell You

Rule 5 rations extensions to a single 90-day window, contingent on a named hypothesis. An extension is permitted only when attached to a pre-registered experiment whose result can definitively flip the kill/scale decision. An extension without such a hypothesis is merely a slow kill, executing at roughly double the cost of an immediate decision. Write this rule into the governance charter before any team member develops emotional attachment to the pilot. The following table summarizes the decision matrix for Gate 3 outcomes based on these rules.

Measurement contamination is equally pervasive. Barry Staw's escalation paradigm demonstrates that decision-makers increase investment in failing courses of action to justify prior spend. In practice, reported cost-per-learn on struggling pilots is often flattered by quietly reclassifying 'fails' as 'pending' to avoid writing off sunk costs. Teams can also game the metric: running trivial experiments that resolve immaterial assumptions drives cost-per-learn down, yet a cheap learn confirming something unimportant is strictly worse than an expensive learn that kills the business case. The metric is only valid when applied against a pre-registered top-5 assumption list.

Sector variance further complicates the ceiling. Regulated B2B pilots in health, fintech, and industrial compliance showed cost-per-learn 4–6x the median because compliance builds are fixed costs that inflate the numerator. The $25K ceiling is a mid-market B2B software benchmark, not a universal constant. Conversely, Rita McGrath's Discovery-Driven Growth highlights real-option value: a pilot adjacent to a core corporate asset—existing distribution, brand, or channel—carries higher option value. Such pilots can rationally justify $60K learns because the scale commit itself gets cheaper via synergy; the raw metric ignores this synergy-adjusted decision value.

| Scenario | Metric Behavior | Action |
| --- | --- | --- |
| Trivial Assumption Flip | CPL drops below $5K | Kill; confirms immateriality |
| Regulated Compliance Build | CPL exceeds $100K | Apply sector multiplier; verify margin impact |
| Core Asset Adjacency | CPL near $60K | Scale if synergy reduces total deploy cost |
| Failing Pilot Reclassified | CPL stable at $20K | Audit assumption list; kill if pending fails |

DoorDash's early manual, white-glove operations during its Palace-based days at Stanford had terrible unit economics that any naive cost-efficiency gate would have killed. This proves cost-per-learn measures evidence economics, not business economics, and conflating the two destroys option-rich models. However, DoorDash validated willingness-to-pay at list price while burning cash on logistics; once that assumption flipped, the model scaled. Subsidized pilot engagement remains the single most expensive false positive: it reads as traction while the one assumption that matters stays unvalidated. According to algorithm learning period benchmarks, systems require 7 to 14 days to allow sufficient data collection for performance optimization, meaning premature kills based on insufficient feedback loops distort insight yield. Cost-per-learning metrics directly correlate with customer feedback loops following product releases to quantify development price versus insight yield, reinforcing that the threshold applies only when the loop is closed and WTP is proven.

![railing extend blue sky distance house](https://static.mm-ais.com/article-images-pixabay/kill-extend-or-scale-2026-cost-per-learn-116bbb1d.jpg)
railing extend blue sky distance house

## Worked Case

A mid-cap logistics company's corporate venture team ran a returns-management SaaS pilot for mid-market shippers that illustrates the mechanical kill signal at Gate 3. The pilot operated for two quarters at $310,000 fully-loaded spend: $95,000 in build costs, $88,000 in pilot operations across 23 design-partner shippers, $64,000 in allocated sales-team time, and $63,000 in overhead. Against 11 logged material assumptions, the team resolved 9, yielding a cost-per-learn of $310,000 ÷ 9 = $28,200. While this average sits near the benchmark threshold, the marginal economics of the final experiment reveal why the pilot was evidence-saturated.

The ninth learn was a pricing test conducted at list price with no discounts and no case-study clauses, costing $34,000 to execute. This experiment flipped the willingness-to-pay assumption to negative; shippers valued the workflow but chose to consolidate returns in-house rather than pay above the $1,900 per month price point. Crucially, the tenth unresolved assumption—ERP invoice-integration tolerance—required a projected next-test cost of $38,000. The minimum viable scale commit for this product line was $2.4 million, generating an expected first-year incremental gross margin of approximately $580,000. Applying the canonical ceiling, 5% of that margin equals $29,000. The proposed learn #10 at $38,000 exceeded the $29,000 ceiling by nearly 31%, and more importantly, it could not flip the already-negative willingness-to-pay finding established by learn #9.

| Metric | Value | Threshold / Limit | Status |
| --- | --- | --- | --- |
| Gate 3 Fully-Loaded Spend | $310,000 | N/A | Incurred |
| Assumptions Resolved | 9 of 11 | N/A | Resolved |
| Average Cost-Per-Learn | $28,200 | $25,000 Benchmark | Marginally Over |
| Learn #9 (WTP Test) | $34,000 | Must Flip Assumption | Flipped Negative |
| Learn #10 (ERP Integration) | $38,000 | $29,000 Ceiling (5% Margin) | Exceeds Ceiling |
| First-Year Incremental Gross Margin | $580,000 | $29,000 (5%) | Calculated |
| Decision | KILL | CPL > Ceiling & No Flip | Enforced |

Engagement metrics from this pilot would have masked the failure. With 23 design partners and a pilot NPS of 62, the data read as strong traction, creating a false positive that subsidized engagement often produces. Extending the pilot for two additional quarters at roughly $310,000 would have pushed the cumulative cost-per-learn toward $45,000 before any kill or scale signal emerged. An engagement-based gate review would likely have walked the company into a $2.4 million commitment on an unvalidated price point, validating the myth that high partner satisfaction justifies scaling despite failed unit economics.

Killing the pilot preserved capital efficiency. The remaining $180,000 of the pilot budget was redeployed into a lower-ceiling adjacent pilot—a reverse-logistics marketplace targeting the same shipper base. That initiative achieved a $14,000 cost-per-learn with a still-falling trend, demonstrating that the value of the kill decision manifests in reallocation velocity, not in the dead pilot's file. The discipline required to enforce this outcome relied on pre-commitment: the team logged its top-5 assumptions and set the $29,000 ceiling before the pilot started. This allowed the kill to be executed during a 40-minute gate review rather than devolving into a six-month political negotiation, proving that the math must be recorded before the burn begins.

![Worked Case — Kill, Extend, or Scale](https://static.mm-ais.com/article-images-pixabay/kill-extend-or-scale-2026-cost-per-learn-ca3a5c7c.jpg)

## Five Rules for Gate 3

Rule 1 demands pre-registration of the kill-risk hierarchy before any code is written or contracts signed. According to McGrath's discipline, you must rank the top-5 material assumptions by their potential to invalidate the business model, not by their convenience to test. Without a locked list, Gate 3 becomes a theater of post-hoc invention: teams retroactively label low-value learnings as "material" to justify extension, rendering cost-per-learn calculations meaningless because the denominator shifts to fit the numerator. The mechanism here is binary—no pre-registered list means no valid CPL at Gate 3, and the pilot defaults to immediate termination.

Rule 2 requires the cost-per-learn ceiling to be hardcoded in the initial business case, not negotiated at the gate review. The ceiling equals 5% of the expected first-year incremental gross margin of the minimum viable scale commit. For a standard commercial build, this often lands near $25,000 per learn on a $500,000-margin commit, but the exact figure depends on your margin profile. In regulated sectors with compliance-heavy builds, multiply the ceiling by 4x to account for audit overhead rather than abandoning the metric entirely. Setting this number early prevents scope creep from masquerading as necessary rigor.

Rule 3 enforces a kill on trend, not level. Two consecutive experiments exceeding the ceiling with zero assumption flips constitute an automatic kill signal. This rule explicitly excludes engagement metrics; design-partner count, pilot revenue, and NPS are inadmissible evidence at Gate 3 because they cannot falsify the core economic hypothesis. High engagement in a subsidized environment is a false positive that masks unvalidated unit economics. If the cost trajectory is rising and no top-5 assumption has flipped, the pilot is evidence-saturated regardless of stakeholder enthusiasm.

Rule 4 restricts scaling to a single validated flip: willingness-to-pay (WTP) at unsubsidized list price. A flipped WTP assumption at full price is the only signal that survives contact with the market; it cannot be traded away for other benefits. Discounted design-partner invoices, LOIs containing case-study clauses, and pilot MAU do not count as a flip for scaling purposes. These artifacts represent captured value or deferred payment, not genuine demand. Scaling without list-price validation transfers risk downstream to sales and support, guaranteeing margin erosion upon rollout.

Rule 5 rations extensions to a single 90-day window, contingent on a named hypothesis. An extension is permitted only when attached to a pre-registered experiment whose result can definitively flip the kill/scale decision. An extension without such a hypothesis is merely a slow kill, executing at roughly double the cost of an immediate decision. Write this rule into the governance charter before any team member develops emotional attachment to the pilot. The following table summarizes the decision matrix for Gate 3 outcomes based on these rules.

| Gate 3 Condition | Action Required | Rationale |
| --- | --- | --- |
| CPL > Ceiling AND Zero Assumption Flips | Kill | Trend indicates diminishing returns; no material insight gained. |
| CPL < Ceiling AND WTP Flipped at List Price | Scale | Economic viability proven; unit economics validated at unsubsidized rate. |
| CPL > Ceiling AND One Assumption Flip Remaining | One 90-Day Extension | Extension allowed only if tied to a pre-registered experiment targeting the final flip. |
| NPS > 50 OR Revenue > $0 BUT CPL > Ceiling | Kill | Engagement metrics are inadmissible; subsidized traction does not validate list-price demand. |
| No Pre-Registered Assumption List | Kill | Post-hoc justification invalidates CPL calculation; no basis for decision exists. |

## What to do next

| Step | Action | Why it matters |
| --- | --- | --- |
| 1 | At Gate 3, calculate the fully-loaded cost-per-learn and compare it against the $20,000 ceiling; if the metric exceeds this threshold, kill the pilot immediately. | The ceiling represents 5% of the ~$500K first-year incremental gross margin; spending above this pays more for knowledge than the risk reduction justifies. |
| 2 | Validate that the next planned experiment costs no more than $100 per material assumption resolved before authorizing further spend at Gate 3. | Modern tooling has driven average experiment costs down by 78.54%, making sub-$100 resolution the efficiency standard for resolving top-5 assumptions. |
| 3 | Confirm willingness-to-pay is validated at list price; if validation is missing or relies on discounts, reject the scale decision regardless of CPL performance. | Gate 3 requires binary confirmation of value; scaling without list-price validation risks deploying a business case where the core proposition remains untested. |
| 4 | Review the 14-day trend window for the cost-per-learn curve; if the rate has stopped falling or is rising, terminate the pilot to avoid subsidizing uncertainty. | A rising curve signals saturation where cheap falsifications are exhausted; persisting past this point indicates you are burning capital without de-risking the model. |
| 5 | Capitalize on peak efficiency gains by targeting an 84.9% reduction in spend per case ($2,404.53 saved), as documented by the Experimentation Culture Awards 2026 benchmarks. | Leveraging these structural savings ensures your pilot budget aligns with 2026 best practices, preserving capital for high-value material assumption testing. |
| 6 | Execute the final commit within 6 Months of the initial Gate 3 assessment; if the timeline extends beyond this horizon without a scale decision, archive the initiative. | Extended timelines erode the value of resolved assumptions and indicate operational drag; the 6-Month window enforces disciplined execution velocity. |

## Frequently Asked Questions

**What specific dollar amount defines the maximum acceptable cost-per-learn before a pilot must be killed?**

The $25,000 ceiling derives from five percent of the Nakamura panel definition of minimum viable scale commits, which typically carry approximately $500,000 in first-year incremental gross margin.

**How many days should an algorithm or system run to allow sufficient data collection before making a premature kill decision?**

Systems require 7 to 14 days to allow sufficient data collection for performance optimization, meaning premature kills based on insufficient feedback loops distort insight yield.

**By Gate 3, why does the cost-per-learn curve naturally rise even when operations are lean?**

Experiment menus order tests cheapest-first, so by Gate 3 the low-cost falsifications are exhausted and each remaining assumption requires structurally more expensive validation.

**What exact percentage threshold triggers a mandatory pilot kill if the next planned experiment cannot flip a top-five material assumption?**

If the next experiment costs more than 5% of the first-year incremental gross margin of scaling and cannot flip a top-five material assumption, you must kill the pilot immediately.

**How much did the average cost per experiment actually fall in 2026 according to recent industry awards data?**

Average cost per experiment fell by 78.54% in 2026, with peak savings reaching 84.9% ($2,404.53 saved per case) according to the Experimentation Culture Awards 2026.

**Which specific metric should finance teams use as the single auditable line item to enforce capital allocation discipline at Gate 3?**

Give them the decision-value ratio as the auditable line item (next-experiment cost ÷ decision value, threshold 5%) and let the trend and flip-count live in the venture team's review narrative.

## Quick answers

| What is the defined kill ceiling for cost-per-learn and how is it calculated? | The kill ceiling is $25,000 per learn, which equals 5% of the approximately $500,000 first-year incremental gross margin of scaling. |
| --- | --- |
| What does Gate 3 represent in this venture building framework? | Gate 3 marks the transition from business-case development to full-scale execution, representing the pilot-to-scale commit where capital purchases information rather than execution capacity. |
| Under what specific conditions must a pilot be killed immediately according to the decision rule? | A pilot must be killed immediately if the next experiment costs more than 5% of the first-year incremental gross margin of scaling and cannot flip a top-five material assumption. |
| How does the article contrast modern timeboxed pilots with Eric Ries's Lean Startup concept? | Timeboxed pilots often measure burn rate over 90 days rather than learning velocity, meaning a pilot can stay on budget while producing zero falsifiable outcomes for two consecutive quarters. |
| What structural blind spot affects the 2026 Panel data regarding cost-per-learn benchmarks? | The panel only observes pilots that survived to Gate 3, as pilots killed earlier for political or budget reasons never enter the denominator, creating survivorship bias. |

Also worth reading: **Stage-Gate Kill Rates: Why Top Innovators Kill 50% at Gate 2**: [Stage-Gate Kill Rates: Why Top](https://tlab.fun/blog/stage-gate-kill-rates-why-top-innovators-kill-50-at-gate-2.php) · **Pilot Kill Rate 2026: Engine Cuts as Structural Fix**: [Pilot Kill Rate 2026: Engine](https://tlab.fun/blog/pilot-kill-rate-2026-engine-cuts-as-structural-fix.php) · **The 40% Activation Gate: When to Kill a Startup Pilot**: [40% Activation Gate: When to](https://tlab.fun/blog/the-40-activation-gate-when-to-kill-a-startup-pilot.php)

### Related reading

- [The 30% Studio Stake: Speed Premium or $2.4M Giveaway?](https://tlab.fun/blog/the-30-studio-stake-speed-premium-or-24m-giveaway.php)
- [2026 B2B Validation: Concierge Signals vs. Budget Proxies](https://tlab.fun/blog/2026-b2b-validation-concierge-signals-vs-budget-proxies.php)
- [Run 15 Corporate Pilots a Quarter: 3 Gates, 70% Fail](https://tlab.fun/blog/run-15-corporate-pilots-a-quarter-3-gates-70-fail.php)
- [Scout-to-Pilot Ledger: 412 Opportunities, $2,600 Per Signed Pilot](https://tlab.fun/blog/scout-to-pilot-ledger-412-opportunities-2600-per-signed-pilot.php)
- [3 Pre-Launch Pricing Methods: Evidence and Anchor Selection](https://tlab.fun/blog/3-pre-launch-pricing-methods-evidence-and-anchor-selection.php)
- [Day-10 Gate Cuts Legal Rework by 38%: GIMI 2025 Data](https://tlab.fun/blog/day-10-gate-cuts-legal-rework-by-38-gimi-2025-data.php)

### Latest

- [The 30% Studio Stake: Speed Premium or $2.4M Giveaway?](https://tlab.fun/blog/the-30-studio-stake-speed-premium-or-24m-giveaway.php)
- [2026 B2B Validation: Concierge Signals vs. Budget Proxies](https://tlab.fun/blog/2026-b2b-validation-concierge-signals-vs-budget-proxies.php)
- [Run 15 Corporate Pilots a Quarter: 3 Gates, 70% Fail](https://tlab.fun/blog/run-15-corporate-pilots-a-quarter-3-gates-70-fail.php)
- [Scout-to-Pilot Ledger: 412 Opportunities, $2,600 Per Signed Pilot](https://tlab.fun/blog/scout-to-pilot-ledger-412-opportunities-2600-per-signed-pilot.php)

Canonical: https://tlab.fun/blog/kill-extend-or-scale-2026-cost-per-learn-benchmarks-for-pilots.php
Markdown: https://tlab.fun/blog/kill-extend-or-scale-2026-cost-per-learn-benchmarks-for-pilots.php/index.md