A new Intelligence Impact Quotient (IIQ) quantifies how deeply AI is embedded in work by combining time-decayed token stocks, usage frequency, recency gates and measures of leverage, task complexity and autonomy. Unlike raw usage counts, IIQ aims to separate repetitive low-leverage prompting from higher-consequence, autonomous AI-assisted tasks and offers bounded interpretations of efficiency and financial impact for deployment monitoring.

Main Finding

The paper proposes the Intelligence Impact Quotient (IIQ), a stateful, composite metric to quantify how deeply AI systems are embedded into individual workflows and organizational units. IIQ moves beyond raw seat counts or token volume by combining novelty-weighted, time-decayed token stock with distinct-task frequency, a grace-period recency gate, organizational leverage, rolling task-complexity, and an explicit autonomy multiplier. It is a deployment-focused, interpretable index (normalized 0–1000) intended as a leading indicator of AI integration — not a direct measure of model capability or a causal estimate of realized productivity.

Key Points

• Purpose: distinguish superficial or repeat usage from sustained, semantically diverse, consequential AI integration in workflows.
• High-level formula: IAIp = Tp × Fp × Rp × V × Cp × Ap
  • IAIp: raw Intelligence Adoption Index for period p.
  • Tp: novelty-weighted, time-decayed token stock.
  • Fp: frequency multiplier derived from distinct-task mass (log transform of a decayed accumulator).
  • Rp: recency gate (grace-period based), acts only after inactivity exceeds threshold g.
  • V: organizational leverage (static role-based multiplier).
  • Cp: rolling-window average task complexity.
  • Ap: autonomy multiplier based on agent-initiated turns and active autonomous runtime.
• Anti-gaming / novelty: each interaction i gets a novelty weight νi,p = max(0, 1 − max_j∈H Sim(xi,p, xj)). Effective tokens and distinct-task mass use ν to downweight near-duplicates.
• Temporal dynamics:
  • Token stock evolves: Tp = Tp−1*(1 − αT) + Gp where Gp = Σ νi,p ti,p. Example αT = 0.05/day (≈14-day half-life).
  • Frequency accumulator decays similarly; Fp = 1 + ln(1 + Fraw,p).
  • Recency gate Rp = 1 if inactivity ≤ g; otherwise Rp = exp(−λ(Ip − g)).
• Autonomy: per-interaction autonomy mass ui,p = ωτ·AgentTurns + ωh·ActiveRunHours; period Up = Σ νi,p ui,p; Ap = 1 + γ ln(1 + Up) (log keeps returns sublinear).
• Complexity: Cp = weighted average of complexity tiers over a rolling window W, using ν weights to focus on distinct tasks.
• Normalization & interpretive layers:
  • IIQIndex (0–1000): log10-based scaling of IAI with an upper reference MaxExpected (paper uses 50,000,000).
  • Micro IAI: a point-in-time approximation for a single interaction: Micro IAI = νi·ti·V·ci·ai.
  • Sub-daily updates: convert daily decay αdaily to period decay αperiodic = 1 − (1 − αdaily)^(P/24).
• Efficiency & financial proxies (interpretive):
  • Estimated hours saved Ĥp = min(ρ WorkHoursAvailable, Gp/1000) × k × Cp × Ap
    • k = baseline hours saved per 1,000 effective tokens; ρ caps credible share of work time displaced.
  • Estimated financial impact USDp = Ĥp × (W × V) with W an hourly baseline wage; V amplifies opportunity value by leverage.
• Initialization: neutral base values (T0 = 0, Fraw,0 = 0, R0 = 1, C0 = 1, A0 = 1). Novelty defaults to 1 when no history exists.
• Illustrations: synthetic traces show how identical raw-token volumes diverge under novelty weighting and how the recency gate treats short vs long inactivity differently.
• Deployment intent: IIQ is presented as a practical tracking tool for organizations and researchers, with calibration choices left context-specific.

Data & Methods

• Method type: theoretical framework + mathematical formulation + synthetic illustrative simulations. No empirical deployment dataset is provided in the paper.
• Core mechanisms:
  • Similarity metric Sim(x,y): implementation-agnostic; suggested options include embeddings, normalized edit distance, keyword overlap, or hybrids. This underpins novelty weights ν.
  • Decay parameters: αT, αF (for token and frequency half-lives); example αT = 0.05/day (≈14-day half-life).
  • Frequency uses distinct-task mass Dp = Σ νi,p to avoid rewarding repeated near-identical prompts.
  • Recency gate uses a grace threshold g (periods of allowed inactivity before exponential penalty).
  • Organizational leverage and complexity are captured via tiered discrete mappings (illustrative tables provided: leverage tiers from Individual Contributor to Head of State; complexity tiers from Routine to Agentic).
  • Autonomy logging requires telemetry of agent-initiated turns and active autonomous runtime.
• Computational considerations:
  • Stateful accumulators (Tp, Fraw,p) require periodic updates; conversion formulas given for higher-frequency updates.
  • Micro IAI is offered for point-in-time scoring without historical state.
• Calibration: normalization (MaxExpected), weights (ωτ, ωh, γ, k, ρ, W, V), window lengths, and similarity thresholds are deployment-specific; paper gives illustrative defaults and examples.
• Validation: paper emphasizes IIQ should be validated against realized organizational outcomes (sales, conversion, TFP, hours), but does not present empirical validation itself.

Implications for AI Economics

• Improved measurement of deployment: IIQ offers a more nuanced, stateful measure of AI integration than coarse metrics (seats, raw tokens). It can help distinguish experimentation from persistent, consequential adoption.
• Leading indicator for organizational change: because IIQ encodes recency, novelty, complexity, leverage, and autonomy, it can function as an early signal of where AI is becoming embedded and where reallocation or supervision costs may emerge first (consistent with evidence that integration shows up first in task redesign and oversight).
• Useful for firm- and sector-level monitoring: aggregating IIQ across users or teams could improve firm- or regional-level indicators of AI adoption intensity, informing macro studies of diffusion, productivity potential, or labor reallocation.
• Complement to capability benchmarks: IIQ measures deployment; it can be calibrated against capability metrics (e.g., GDPval) to map what models can do to what organizations actually operationalize.
• Caution about value translation: the paper’s hours/financial conversions are interpretive proxies, not causal estimates. Economists should use IIQ as an input to causal studies (RCTs, difference-in-difference, instrumental variables) rather than a substitute.
• Research & policy use cases:
  • Targeting interventions: identify high-leverage users or teams where modest AI integration can have outsized organizational impact.
  • Monitoring governance and risk: autonomy and agentic tiers can highlight where higher-risk autonomous workflows are emerging.
  • Studying diffusion: IIQ’s stateful nature makes it amenable to analyzing adoption dynamics, churn, and the role of novelty vs repetition in scaling.
• Limitations & risks to watch in economic applications:
  • Sensitivity to similarity metric and parameter calibration — cross-study comparability requires standardized implementations and transparent parameter choices.
  • Potential for manipulation if novelty or autonomy signals are gamed (though novelty weighting mitigates simple replay attacks).
  • Requires adequate telemetry (prompts, embeddings, agent turn logs, run-time measures) which may not be uniformly available across firms or sectors.
  • Mapping IIQ to realized productivity, wages, or GDP effects requires external causal validation; IIQ alone does not prove value realized.
• Overall: IIQ provides a structured, operationalizable bridge between model usage telemetry and economic analysis of AI deployment, but its usefulness depends on careful calibration, robust telemetry, and empirical validation against outcomes.