Organizations using AI
78%
McKinsey 2024 · 1,491 respondents
Not seeing EBIT impact
>80%
Despite significant technology spend
Productivity gain inside frontier
+40%
MIT Sloan / BCG · 700+ consultants
Performance drop outside frontier
−19pp
When AI deployed beyond capability edge
Revenue uplift per workflow unit
8×
Seat → outcome pricing (PitchBook)
Training data needed with curation
30%
Quality beats volume (Marion et al.)
FOUR-DIMENSION SIGNAL THRESHOLD — displacement occurs when all four are favorable
1 · Cognitive fidelity required
How nuanced is the judgment? How catastrophic is an error? Workflow and error consequence scores.
2 · Compositional depth
How many sequential, globally-coherent steps must execute correctly? Collapse regime from Apple research.
3 · Training distribution coverage
How well-represented is this workflow's constraint structure in model training data? Proprietary data moat.
4 · Organizational readiness
Are workflows redesigned? Is data AI-ready? Is the human-in-the-loop designed for the jagged frontier?
RESEARCH PROVENANCE · EIGHT SOURCES · SIX INSTITUTIONS
PitchBook · SaS thesis · 8× revenue uplift · 10 structural moats
Google DeepMind · AI co-scientist · test-time compute scaling
Apple · Illusion of Thinking · three regimes · collapse boundary
NVIDIA · Hymba · hybrid SLM · 11× cache reduction · on-premise
Marion et al. · 30% curated data beats 100% raw · data = performance
McKinsey · workflow redesign = #1 EBIT driver · only 21% have done it
MIT Sloan · jagged frontier · ±40%/−19pp · three pre-conditions
CaP-X · Berkeley / Stanford / NVIDIA · Physical AI · 4th wave
Four displacement domains mapped against signal threshold and proximity to displacement. Physical AI emerges as a fourth domain below the digital frontier. Click any workflow dot to explore.
Automate now
Build & defend
Sequence next
Revenue intelligence
Protect & invest
Displacement threshold
ROIC multiple vs. investment horizon. Bubble size = confidence level. Priority zone: top-left — high returns, short timelines. Revenue Intelligence shown separately — different equation (revenue amplification, not cost reduction).
Automate now
Build & defend
Revenue intelligence
Protect / Physical AI
Revenue Intelligence ROIC multiples are multiplicative on revenue base. Physical AI horizon reflects compliant-hardware structured tasks only (CaP-X 2026). Strategic M&A excluded — human judgment non-substitutable at current capability levels.
| Workflow | Domain | ROIC type | AI cost/unit | Human cost/unit | Net economics | Payback | Action |
|---|---|---|---|---|---|---|---|
| DIGITAL KNOWLEDGE WORK — cost displacement | |||||||
| Data entry / ETL | Automate now | Cost saving | $0.50 | $8 | 94% reduction | 9 mo | Deploy now |
| Invoice & AP processing | Automate now | Cost saving | $2 | $25 | 92% reduction | 15 mo | Deploy now |
| Tier 1 customer support | Automate now | Cost saving | $3 | $30 | 90% reduction | 18 mo | Deploy now |
| Document generation | Automate now | Cost saving | $1.50 | $20 | 93% reduction | 15 mo | Deploy now |
| Code gen (standard) | Automate now | Cost saving | $4 | $35 | 89% reduction | 21 mo | Deploy now |
| Compliance monitoring | Build & defend | Cost saving | $8 | $35 | 77% reduction | 24 mo | Invest in data moat |
| Contract analysis | Build & defend | Cost saving | $15 | $80 | 81% reduction | 30 mo | Invest in data moat |
| REVENUE INTELLIGENCE — revenue amplification (different equation: uplift × margin / investment) | |||||||
| CRM hygiene & enrichment | Rev enabler | Rev enabler | $0.10/record | $4.00/record | 97% + quality ↑ | 6 mo | Deploy now |
| Lead scoring & routing | Rev amplify | Rev amplify | $0.50/lead | $12.00/lead | +15–40% conv. | 12 mo | Deploy now |
| Pipeline forecasting | Rev amplify | Rev amplify | $2K/mo | $45K FTE | 50–70× ROIC | 3–6 mo | Deploy now |
| Churn prediction + intervention | Rev protect | Rev protect | $5K/mo | Cost of churn | 5–20× retention | 6–12 mo | Deploy now |
| Pricing optimization (known) | Build & defend | Rev amplify | $10K/mo | $200K+ analyst | 2–8% margin exp. | 12–18 mo | Build & deploy |
| Strategic pricing (novel markets) | Build & defend | Rev amplify | $20K/mo | $400K+ team | High upside, risk | 24–36 mo | HITL mandatory |
| MANUFACTURING ROBOTICS — physical displacement | |||||||
| Vision quality inspection (T1) | Pilot now | Cost saving | $0.10 + hw | $2.50/unit | ~91% reduction | 18–24 mo | Pilot now |
| Pick-and-place structured (T1) | Pilot now | Cost saving | $0.001 + hw | $0.80/cycle | ~96%* | 30–36 mo | Compliant hw only |
| Semi-structured assembly (T2) | Build toward | Cost saving | TBD | $1.50/cycle | 2027–2029 | 36–48 mo | Invest task data |
| General flex manufacturing (T3) | Monitor | Future | Unknown | Variable | 2029–2033+ | — | Monitor arch. curve |
* Pick-and-place: compliance-dependent (CaP-X 2026). Cross-embodiment validation required. Revenue Intelligence ROIC is multiplicative on revenue base. All figures indicative.
MIT Sloan six pre-conditions — score before any deployment commitment
Business problem defined
Target workflows identified, subproblems decomposed, AI solution matched to each. Not "AI everywhere" — specific displacement cases with economic rationale.
Data is AI-ready
Datasets complete, governed, and curated. Quality scoring applied (perplexity pruning). Proprietary data identified and protected from unauthorized model training.
Workforce configured for the frontier
Employees know which tasks are inside vs. outside the jagged frontier. Senior judgment governs consequential outputs — not junior AI fluency.
Workflows have been redesigned
Not AI added to existing workflows. The workflow itself rebuilt around AI capabilities. Single largest EBIT driver — McKinsey regression across 25 attributes.
KPIs defined and tracked
Well-defined KPIs including error rate, throughput, cost-per-cycle, and human override rate. McKinsey: most strongly correlated with EBIT impact at larger organizations.
CEO / board governance in place
AI governance at C-suite and board level, not delegated to IT. Delegating implementation to IT is "a recipe for failure" — McKinsey. This is transformation, not technology deployment.
Score: 0 / 6 pre-conditions met
Kahneman two-tier AI strategy — fast experiments feed slow strategy
Fast tier · now → 6 months
Run pilots in "Automate Now" quadrant. Deploy in highest-volume, lowest-signal-threshold workflows. Include Revenue Intelligence (CRM hygiene, lead scoring, pipeline forecasting) — these return fastest. Measure error rates, cost-per-cycle, and throughput. Capture data on what works.
Slow tier · now → 18 months
Build data curation infrastructure. Define proprietary data assets in "Build & Defend" workflows. Establish governance architecture. Design workforce configuration for the jagged frontier. Set ROIC targets and KPI framework. Begin manufacturing robotics Tier 1 pilots where physical workflows qualify.
Capital allocation decision · board level
Sequence investment across four domains using the ROIC waterfall. Prioritize Revenue Intelligence early — different equation, faster returns. Physical AI pilots 2026–2027 for compliant-hardware structured tasks. Tier 2 manufacturing investment in 2027–2028. Reserve architectural regime decisions for 2028 planning cycle.