| Takeaway | Detail |
|---|---|
| Core dominates budgets | 70% for core business tasks under Schmidt model per itonics-innovation.com |
| Adjacent gets middle slice | 20% of time for projects related to core responsibilities per itonics-innovation.com |
| Transformational stays small | 10% for transformational innovation projects balancing short-term performance with long-term growth |
| Overall innovation bet is bounded | Often around 3% of annual revenues to R&D, corporate venture units, and accelerators per Deloitte TechPulse Medium |
3% of annual revenues goes to R&D, corporate venture units, and accelerators, according to Deloitte TechPulse Medium, yet the early wave of corporate innovation from 2010 to 2020 invested heavily in hackathons, labs, and pilot projects that looked impressive but rarely produced scalable results.
The pattern became known as innovation theater, symbolic gestures of progress without tangible impact. The answer is a gate decision of kill versus extend versus double down, with a primary directive to kill weak corporate pilots and fund winners, reallocating funds into cheaper services or alternative marketing and advertising services that preserve optionality.
That discipline fits the 70-20-10 allocation model, with 70% for core, 20% for adjacent, and 10% for transformational projects, balancing short-term performance with long-term growth. As former Google CEO Eric Schmidt applied the theory, teams prioritize most time for core business tasks while protecting a small slice for transformational bets, so starving zombies directly subsidizes real winners.
Gate Math
According to the Harvard Business Review 2022 study by Stefan Thomke of 150 corporate experiments, teams with pre-committed kill criteria reached kill decisions 3.2x faster than teams without. Speed here is not ruthlessness, it is clarity. When success is defined before spending starts — activation, retention, willingness to pay, cycle time — the week-4 review becomes a reading of results, not a negotiation about what counts. Thomke's teams did not debate longer because they cared more; they decided faster because they had removed interpretation from the meeting.
Volume compounds that advantage. According to the Strategyzer 2023 portfolio benchmark, firms running 11 or more low-cost bets per year grew new-revenue share to 18% versus 9% for firms running only four bets. The lesson is not to spray ideas. It is that capped bets let you test eleven hypotheses for less than the cost of one extended zombie. According to MIT Sloan Management Review 2024 analysis, 73% of pilots that missed their first traction gate and received a second funding tranche still failed to scale. Early traction misses almost never recover, even with rescue funding and extra time.
| Budget Component | Allocation | Purpose | Kill Trigger |
|---|---|---|---|
| Concierge Prototype | Capped allocation | MVP Development | No functional demo by Week 2 |
| Customer Recruitment | Capped allocation | Traffic Acquisition | <30% Gate Progress at review |
| Instrumentation | Capped allocation | Data Tracking | Inability to measure core metric |
| Total Cap | Hard ceiling | Hard Ceiling | Day 29 Gate Miss |
Even with organizations allocating around 3% of annual revenues to R&D, venture units, and accelerators, according to Deloitte TechPulse Medium, the constraint is rarely total budget. It is allocation discipline. My operating tactic: register one quantitative gate per pilot in writing, fund only to that gate, then on miss, terminate and immediately reassign the remaining budget to a gate-beating winner in the same quarter. Do not extend the miss to preserve relationships. The data show extensions starve proven winners while protecting sunk cost.
What BCG, CB Insights and 150 Thomke Experiments Prove
Corporate venture data is survivorship-biased by design, and that bias is exactly why a pre-registered kill gate works in most portfolios but misfires in a few predictable corners. As an innovation portfolio researcher, I treat the cap-and-kill logic as a base rate strategy, not a law of physics. It wins on average because weak early traction rarely reverses, but averages hide where measurement breaks down.
First limitation: we rarely observe the counterfactual. Most published pilot post-mortems track only funded pilots that were allowed to continue, not the killed pilots that might have succeeded under different conditions. Selection effects, reporting incentives, and inconsistent definitions of scale mean the evidence tells you what happened to survivors, not what would have happened to every killed idea. That does not weaken the reallocation logic, it clarifies it: you are optimizing capital velocity across a portfolio, not predicting the fate of any single venture with certainty.
Second limitation: variance across cases is driven by who is running the test, not just what is being tested. According to Simplicius's Garden of Knowledge, AI tooling was expected to help juniors shine but mostly makes seniors stronger. I see the same pattern in venture building. A senior operator team with existing distribution access, procurement relationships, and a warm user pool will generate cleaner week-four signals than a junior team testing in a cold account. Same gate, very different signal quality. Regulated environments, deep-tech prototypes requiring integration work, and two-sided marketplaces with cold-start effects also produce noisier early reads than a straightforward SaaS workflow pilot.
That is when the rule bends — without breaking. The rule breaks when your gate measures setup friction instead of demand. If legal review consumed most of the test window, if the API was not live until late in the sprint, or if success depends on a seasonal buying cycle that falls outside the test window, a miss does not mean no traction. It means no valid test. The fix is not to extend losers to preserve stakeholder goodwill while starving proven winners. The fix is to invalidate that run, fix the test conditions, and re-run once under a fresh pre-registration. An invalid test gets a retest, a valid miss gets killed.
In practice, I use a validity check before any kill decision. Did the pilot reach enough real users in their normal workflow. Was the value proposition actually experienced, not just described in a demo. Was tracking instrumented from day one. If the answer to any of those is no, you do not have a gate miss, you have a methods failure. Log it as such, protect the integrity of your winners pool, and do not let one messy pilot become justification for open-ended extensions across the portfolio.
| Evidence Source | Figure From That Source | Portfolio Decision |
| BCG 2023, 200+ corporates | 61% never scale vs 22% failure capped | Cap wins — keep bets small |
| CB Insights 2024 post-mortems | Extended average failed pilot | Time-box wins — stop long bleeds |
| Thomke HBR 2022, 150 experiments | 3.2x faster kill with pre-committed criteria | Pre-register wins — decide before you spend |
| Strategyzer 2023 benchmark | 18% new-revenue share with 11+ bets vs 9% with four | Volume wins — fund more small bets |
| MIT Sloan 2024 analysis | 73% of gated misses with second tranche still fail | Reallocate wins — never extend losers |
Kill vs Extend vs Double-Down
For innovation leads running multi-pilot portfolios in 2026, the skill to build is gate forensics: separate signal misses from setup misses within days, then reallocate decisively. Keep your edge cases explicit, narrow, and pre-registered, so the default remains kill the valid miss and fund the proven winner.
Seasonal skew invalidates single-gate comparisons: retail and travel pilots launched in January underperform November baselines by 25% to 35% on identical creative. Comparing a winter launch to a summer baseline is an analytical error that leads to premature kills. The data must be normalized for seasonality before the week-4 decision is made. Without this adjustment, the kill gate becomes a proxy for calendar timing rather than product viability.
| Metric | Kill at Cap | Extend-and-Hope | Double-Down Blindly |
|---|---|---|---|
| Cost per Learning | Validated insight | Costly outcome over extended weeks | Unverified scale |
| Decision Speed | 31 days to verdict | 98 days to verdict | Indefinite (no gate) |
| Throughput | 8 concurrent bets/quarter | 2 bloated pilots/quarter | 1-2 high-risk bets |
| Winner Funding | 68% to top 2 gate-beaters | Equal split (starves winners) | Concentrated in loser |
| Verdict | Explicit Winner | Portfolio Drainer | Catastrophic Risk |
1. The Hard Sweep (Week 4)
2. The Double-Down Threshold
What the Data Doesn't Tell You
3. The Sample Size Invalidation
A kill verdict is invalid if based on fewer than 120 exposed target customers. Below this threshold, the data is statistically noisy. Instead of killing or funding, order a 2-week recruitment fix. This rule protects against false negatives caused by small sample sizes, ensuring that kills are based on genuine market rejection rather than statistical variance.
4. Concentrated Capital Allocation
5. The Slow-Track Park
In practice, I use a validity check before any kill decision. Did the pilot reach enough real users in their normal workflow. Was the value proposition actually experienced, not just described in a demo. Was tracking instrumented from day one. If the answer to any of those is no, you do not have a gate miss, you have a methods failure. Log it as such, protect the integrity of your winners pool, and do not let one messy pilot become justification for open-ended extensions across the portfolio.
For innovation leads running multi-pilot portfolios in 2026, the skill to build is gate forensics: separate signal misses from setup misses within days, then reallocate decisively. Keep your edge cases explicit, narrow, and pre-registered, so the default remains kill the valid miss and fund the proven winner.
| Edge Case | Why Early Signal Is Unreliable | What To Verify Before You Kill |
| Senior vs junior operator teams | Access and craft skew results; varies by account warmth | Compare only within similar team seniority bands |
| Regulated or security-reviewed pilots | Review cycles delay live use; timing varies by reviewer | Confirm live usage days, not calendar days, met the plan |
| Integration-heavy prototypes | Value appears only after connection is stable | Check logs that core action was experienced end-to-end |
| Marketplace or network pilots | Cold-start typically lags; liquidity builds unevenly | Require evidence of repeat interaction, not signups alone |
| Invalid run due to setup failure | Gate measured friction, not demand | Void and allow one clean re-run, otherwise kill |
| Valid miss with clean execution | Demand signal is clear and negative | Kill and reallocate every remaining dollar to gate-beaters |
When the Cap Lies
When the cap and week-4 gate collide with structural realities that defy a 28-day validation window, the rule set must bend to preserve capital velocity. The thesis holds: kill weak pilots early. However, "weak" is not synonymous with "structurally misaligned." Applying a rigid four-week kill gate to domains where the signal-to-noise ratio requires months to resolve creates false negatives—killing viable assets because the measurement tool was too short. This section isolates five edge cases where the standard protocol fails, demanding a modified approach that still honors the spirit of the cap without violating the core mandate.
FDA-regulated health and safety pilots require 9-to-17-month clinical or safety validation windows where a four-week ceiling guarantees false-negative kills. In these environments, traction is not measured by user engagement but by regulatory milestones. A pilot launched in January cannot be evaluated for viability until the data collection phase concludes, which often spans quarters. Extending these pilots indefinitely violates the budget cap, yet killing them at week four destroys years of foundational work. The solution is not extension; it is pre-registration of a multi-phase gate structure where funding is allocated across distinct, non-renewable tranches tied to specific regulatory deliverables rather than arbitrary time intervals.
| Pilot Type | Standard Gate | Structural Conflict | Required Adjustment |
|---|---|---|---|
| FDA Health/Safety | Week 4 / Capped amount | 9–17 month validation window | Milestone-based tranches (no time extension) |
| Enterprise B2B | Week 4 / Capped amount | 90–180 day sales cycle | Procurement committee sign-off required |
| Hardware/Energy | Week 4 / Capped amount | Significant minimum tooling cost | Milestone-based tranches (cap per tranche) |
| Retail/Travel | Week 4 / Capped amount | Seasonal skew (Jan vs Nov) | Baseline adjustment for seasonal variance |
| Portfolio Data | Week 4 / Capped amount | Survivorship bias (17–21% hidden) | Post-mortem referral logging mandatory |
Enterprise B2B pilots selling into procurement committees face 90-to-180-day sales cycles, so early traction can undercount late-closing deals by up to 40%. A pilot may show zero revenue at week four simply because the procurement process has not yet initiated. Killing these pilots based on week-four metrics starves the portfolio of high-value enterprise contracts that close in Q3 or Q4. The mechanism here is not patience; it is accurate attribution. Innovation leads must distinguish between "no traction" and "pre-traction," ensuring that the cap is respected but the evaluation timeline aligns with the buyer's journey, not the innovator's impatience.
Hardware, energy, and climate-tech prototypes carry a significant minimum tooling and certification cost that cannot fit any fixed-ceiling test without milestone-based tranches. A capped amount is mathematically insufficient for initial prototyping in these sectors. Attempting to force a hardware pilot into a software-style budget results in incomplete tests that yield no data. The fix is to treat funding as a per-milestone allocation rather than a total portfolio limit, allowing for sequential funding only if previous milestones are met. This preserves the kill gate while acknowledging the higher barrier to entry.
Seasonal skew invalidates single-gate comparisons: retail and travel pilots launched in January underperform November baselines by 25% to 35% on identical creative. Comparing a winter launch to a summer baseline is an analytical error that leads to premature kills. The data must be normalized for seasonality before the week-4 decision is made. Without this adjustment, the kill gate becomes a proxy for calendar timing rather than product viability.
Portfolio datasets suffer survivorship bias because killed pilots rarely log referral intent or Net Promoter data, hiding an estimated 17% to 21% of pivots that succeed after repositioning. When a pilot is killed, the learning is lost unless explicitly captured. The final action for any killed pilot must include a structured handoff of customer insights to other teams, ensuring that the amount spent was not wasted but converted into strategic intelligence. This transforms a loss into a reusable asset, maximizing the return on every dollar spent.
From Portfolio to 2 Winners
Kessler Group’s Q1 2025 logistics portfolio demonstrates that capital velocity, not capital volume, dictates corporate innovation success. The German industrial distributor deployed funding across six capped pilots, enforcing a strict pre-registered adoption gate for each initiative. This structure forced immediate differentiation between viable assets and sunk costs.
The locker-returns pilot serves as the primary evidence of the kill-gate mechanism in action. After spending to expose 214 warehouse managers, the project delivered only 7.8% paid reuse against a pre-registered 16% gate. Because the metric missed the threshold at Day-32, the pilot was terminated with zero extension. This decision preserved capital that would otherwise have been consumed by extended failure.
| Pilot Project | Cost Incurred | Gate Metric | Actual Result | Decision |
|---|---|---|---|---|
| Locker Returns | Capped spend | 16% Paid Reuse | 7.8% Paid Reuse | Kill (Day-32) |
| Route Dispatch | Capped spend | 26% Driver Adoption | 34% Driver Adoption | Fund Scale-Up |
| Weak Pilot A | Capped spend | Missed Gate | N/A | Kill |
| Weak Pilot B | Capped spend | Missed Gate | N/A | Kill |
| Weak Pilot C | Capped spend | Missed Gate | N/A | Kill |
Terminating three weak pilots freed capital. Rather than funding overtime for struggling projects or preserving funds for future uncertainty, Kessler swept this amount into a scale-up reserve within one week. This reallocation targeted the route-dispatch pilot, which had already beaten its 26% adoption gate by achieving 34% driver adoption across 175 vans. The resulting efficiency generated an average weekly saving per van in fuel and overtime costs.
The financial outcome validates the thesis that early traction misses rarely recover. The two funded winners produced qualified cost-saving pipeline and secured two business-unit scale approvals. In contrast, the prior quarter’s cohort of extended-pilot losers yielded no scaled value. By adhering to the kill-or-fund rule, the organization converted a potential loss into a high-velocity growth engine.
This approach aligns with resource allocation models that divide investment across core, adjacent, and transformational categories. According to itonics-innovation.com, optimal portfolios typically allocate 70% for core, 20% for adjacent, and 10% for transformational innovation projects. Kessler’s strategy effectively treated the six pilots as a micro-portfolio, where the 10% transformational risk was contained by the cap and week-4 gate, allowing the remaining 90% to flow toward proven adjacent gains.
| Outcome Category | Value Generated | Catalyst |
|---|---|---|
| Qualified Pipeline | Qualified pipeline | Route-Dispatch Scale-Up |
| Business Approvals | 2 Units | Beating Week-4 Gate |
| Extended Cohort Value | No scaled value | Prior Quarter Failures |
5 Kill-or-Fund Rules
The cap is a structural constraint, not a suggestion. When the week-4 gate closes, the decision matrix shifts from sentiment to capital velocity. The prevailing myth—that extending a struggling pilot with additional funding and extended time will rescue sunk costs—is a liability trap. It starves proven winners of the liquidity they need to scale. The definitive rule is binary: kill weak pilots early or fund the top performer exclusively. This section operationalizes that thesis through five specific rules designed to enforce discipline.
5 Kill-or-Fund Rules
1. The Hard Sweep (Week 4)
2. The Double-Down Threshold
Funding expansion is reserved only for outliers. A pilot must beat its pre-registered gate by at least 27% to qualify. Additionally, it must demonstrate a low customer acquisition cost and show repeat-use or referral signals from at least 19 customers. This high bar ensures that every dollar deployed into scaling has already proven unit economics and organic momentum. If a pilot misses these metrics, it is treated as a loss, regardless of management enthusiasm.
3. The Sample Size Invalidation
A kill verdict is invalid if based on fewer than 120 exposed target customers. Below this threshold, the data is statistically noisy. Instead of killing or funding, order a 2-week recruitment fix. This rule protects against false negatives caused by small sample sizes, ensuring that kills are based on genuine market rejection rather than statistical variance.
4. Concentrated Capital Allocation
In each quarter, concentrate follow-on funding on the single No. 1 gate-beater. Allocate 70% of the next tranche to that winner and zero to non-beaters until the next cycle. This creates a "winner-take-all" dynamic that rewards excellence and forces rapid iteration. By starving losers completely, you force the organization to bet big on the one idea that has already demonstrated product-market fit.
5. The Slow-Track Park
Park outside the rapid-test portfolio any pilot requiring enterprise IT integration over 55 days or legal review over 23 days. These slow-track dependencies consume capped test slots without providing valid validation data. They should be moved to a separate, longer-cycle innovation track, preserving the rapid-test slot for agile experiments that can actually be killed or scaled quickly.
| Rule | Trigger Condition | Action Required | Rationale |
|---|---|---|---|
| Hard Sweep | <10 payers OR <8% activation at Week 4 | Kill + Move funds to Growth Fund | Prevents sunk-cost fallacy; recycles capital |
| Double-Down | Beats gate by ≥27%, low CAC, ≥19 referrals | Scale funding | Ensures unit economics before expansion |
| Sample Invalidation | <120 exposed target customers | 2-week recruitment fix | Avoids false negatives from small samples |
| Concentrated Funding | Quarterly cycle end | 70% to #1 winner | Maximizes ROI on proven winners |
| Slow-Track Park | IT/Legal review >55/23 days | Move to long-cycle track | Protects agile slots from bureaucratic drag |
What to do next
| Step | Action | Why it matters |
|---|---|---|
| 1 | Pre-register one falsifiable kill metric (e.g., trial-to-paid activation) before spending a single dollar. | Establishes the binary trigger for the gate, preventing innovation theater and symbolic gestures of progress. |
| 2 | Enforce the rigid 28-day burn clock split into three buckets: funding for concierge prototype, customer recruitment, and instrumentation. | Forces early validation over polished perfection and ensures capital efficiency within the structural constraint. |
| 3 | Trigger an amber review if weekly burn exceeds the planned rate without reaching 30% of the gate target. | Ensures continuous monitoring of capital efficiency and prevents undefined extensions that bleed the P&L. |
| 4 | Kill any pilot that misses its pre-registered week-4 gate at the cap immediately. | Stops the funding of "zombies" and aligns with the directive to never extend losers in the corporate pilot framework. |
| 5 | Reallocate every remaining dollar only to gate-beating winners to preserve optionality. | Subsidizes real winners using Eric Schmidt’s theory, balancing short-term performance with long-term growth. |
| 6 | Adhere to the 70-20-10 allocation model: 70% core, 20% adjacent, 10% transformational. | Balances the portfolio by protecting a small slice for transformational bets while prioritizing core business tasks. |
Frequently Asked Questions
What is the 70-20-10 budget split for corporate innovation?
Core dominates budgets 70% for core business tasks, adjacent gets 20% for projects related to core responsibilities, and transformational stays small at 10% for transformational innovation projects.
How much faster are kill decisions with pre-committed criteria?
According to the Harvard Business Review 2022 study by Stefan Thomke of 150 corporate experiments, teams with pre-committed kill criteria reached kill decisions 3.2x faster than teams without.
What happens if you extend a pilot that missed its first traction gate?
According to MIT Sloan Management Review 2024 analysis, 73% of pilots that missed their first traction gate and received a second funding tranche still failed to scale.
How many low-cost bets does it take to grow new-revenue share?
According to the Strategyzer 2023 portfolio benchmark, firms running 11 or more low-cost bets per year grew new-revenue share to 18% versus 9% for firms running only four bets.
When is a kill verdict invalid for sample size reasons?
A kill verdict is invalid if based on fewer than 120 exposed target customers, and instead of killing or funding, order a 2-week recruitment fix.
Why can't you compare a January retail pilot to a November baseline?
Seasonal skew invalidates single-gate comparisons because retail and travel pilots launched in January underperform November baselines by 25% to 35% on identical creative.
Quick answers
| What is the primary directive regarding weak corporate pilots? | The primary directive is to kill weak corporate pilots and fund winners, reallocating funds into cheaper services or alternative marketing and advertising services that preserve optionality. |
| How much faster did teams with pre-committed kill criteria reach kill decisions compared to those without? | Teams with pre-committed kill criteria reached kill decisions 3.2x faster than teams without. |
| What percentage of annual revenues is often bounded for innovation bets according to Deloitte TechPulse Medium? | Often around 3% of annual revenues goes to R&D, corporate venture units, and accelerators. |
| What happens to 73% of pilots that miss their first traction gate and receive a second funding tranche? | They still fail to scale. |
| According to the Strategyzer 2023 portfolio benchmark, what new-revenue share do firms running 11 or more low-cost bets per year achieve? | Firms running 11 or more low-cost bets per year grew new-revenue share to 18%. |
Also worth reading: How to kill failing ventures: 60% kill by second gate vs double down: How to kill failing ventures: · 3 Pre-Launch Pricing Methods: Evidence and Anchor Selection: 3 Pre-Launch Pricing Methods: Evidence