SAV is not a replacement for realized value; it is a companion. Both numbers belong on the KPI tree: realized value answers “what economic benefit did this feature deliver?”; SAV answers “what did it deliver net of its full costs and benefits, including externalities?” For organizations covered by EU CSRD, the SAV calculation produces inputs to mandatory sustainability reporting. For other organizations, SAV is an increasingly visible indicator of governance maturity.
This article teaches the computation, the KPI-tree integration, the reporting discipline that distinguishes credible SAV from greenwashing, and the limits of the metric as it currently stands.
The SAV computation
SAV decomposes into three components.
Core realized value — the counterfactual-adjusted incremental outcome from Articles 18–23. For a contact-center copilot, this might be the $4.2M in reduced handle time. For a fraud model, $9.1M in avoided losses. For a recommender, $2.8M in incremental revenue. This is the number that would appear in a traditional VRR section 4 financial summary.
Externality layer — quantified carbon, water, employment, equity, and privacy externalities from Article 33. Positive externalities add to SAV; negative externalities subtract. A feature with strong net-negative externalities (a large carbon footprint, meaningful employment displacement without redeployment) produces a smaller SAV than its core realized value suggests.
Risk-adjusted probability of sustainability — the likelihood that the externality profile holds across the feature’s continued operation. A feature whose carbon footprint is expected to shrink (grid decarbonization, efficiency gains) has a different risk-adjusted SAV than a feature whose externalities are likely to grow (scale expansion, regulatory tightening).
The formal specification:
SAV = Core realized value − Σ (externality costs) + Σ (externality benefits), all risk-adjusted for probability of sustainability of the externality profile.
In practice, the computation is performed per feature per reporting period and rolled up to portfolio-level SAV. Uncertainty bands are reported at feature and portfolio level, consistent with the uncertainty disclosure discipline from Articles 33 and 7.
Worked example
A GenAI customer-service copilot delivered $4.2M in realized value over twelve months. Its externality profile:
- Carbon: 12 tCO2e for training (amortized $600 at an internal carbon price of $50/t) + 180 tCO2e for inference at current scale ($9,000 annualized) = $9,600 carbon cost.
- Water: 420 m³ attributable to inference; at local water cost and scarcity weighting, $300.
- Employment: No headcount reduction; 8 FTE redeployed to higher-value work (net-positive: +$60,000 estimated redeployment value).
- Equity: Subgroup accuracy variance analysis shows no meaningful distributional impact.
- Privacy: No incremental privacy externality beyond baseline; zero.
SAV core calculation: $4.2M − $9,900 + $60,000 = $4.25M ≈ $4.25M.
Risk adjustment: the externality profile is relatively stable (grid decarbonization expected; scale growth expected but smaller than carbon-intensity decline). Risk adjustment: −5%.
Final SAV: $4.04M.
Against a non-externality-adjusted value of $4.2M, the SAV is 4% lower. The difference is small in this case because the feature’s externality profile is benign. For features with larger negative externalities — large-scale training with carbon-intensive grids, headcount reductions without redeployment, equity impacts with regulatory exposure — the SAV-to-realized-value gap is larger.
Placing SAV on the KPI tree
SAV sits at the top of the KPI tree alongside realized value, one level below the highest-level business outcome. The placement reflects the dual accountability model: organizations are accountable for economic outcomes (realized value) and for broader outcomes (SAV).
Two placement patterns are common.
Pattern A — Parallel top-line metrics. Realized value and SAV sit side-by-side at the top of the tree. Sub-metrics on drivers, activities, and inputs feed both. This pattern is appropriate when both metrics are material for external reporting.
Pattern B — SAV as a top-line with realized value as a constituent. SAV sits alone at the top; realized value is one of its constituent components. Externality metrics are also constituents. This pattern is appropriate when the organization has elected SAV as its primary AI value metric.
The choice between patterns depends on organizational strategy. Organizations whose strategy strongly emphasizes sustainability choose Pattern B; organizations with more traditional economic orientation choose Pattern A.
Reporting SAV without greenwashing
Three discipline markers distinguish credible SAV reporting from greenwashing.
Marker 1 — Method transparency
Every SAV number discloses its computation method: the counterfactual method for realized value, the measurement approach for each externality category, the risk-adjustment factor and its basis. SAV numbers reported as single values without method disclosure are greenwashing.
Marker 2 — Uncertainty honesty
SAV is less precise than realized value because it compounds multiple uncertain measurements. Point-estimate SAV reported without uncertainty bands is greenwashing. Properly reported SAV carries a range (p10/p50/p90) consistent with the uncertainty discipline from Article 7 (rNPV) and Article 18 (causal-design uncertainty).
Marker 3 — Dual reporting
Both realized value and SAV are reported. Organizations that report only SAV when SAV is favourable and only realized value when SAV is unfavourable are caught out quickly by analysts. The discipline is to report both consistently — in every VRR, on the portfolio scorecard, in the board-grade summary — regardless of which is more favourable in any given period.
Integration with EU CSRD
For EU-covered organizations, the CSRD requires disclosure across multiple material topics that intersect with AI. The European Sustainability Reporting Standards (ESRS), adopted by the European Commission in 2023, specify the disclosure requirements at section level.1 AI-specific disclosure is not separately mandated, but AI is often material under the environmental (E1 climate change), social (S1 own workforce, S2 workers in the value chain), and governance (G1) topics.
The SAV framework produces most of the inputs CSRD disclosure requires: quantified carbon (ESRS E1), quantified water (ESRS E3), quantified workforce impact (ESRS S1 and S2). The AI value practitioner’s job is to ensure the measurements are audit-ready and integrated with the broader sustainability reporting workflow, not to conduct the CSRD compliance work itself.
SEC Climate Disclosure Rules — to the extent they survive ongoing legal proceedings — would create analogous requirements for US-registered public companies. Organizations subject to both regimes find the requirements compatible though not identical; a shared SAV framework reduces duplicate work.
Limits of the SAV framework
Three honest limits.
Limit 1 — Monetization challenges
Putting a dollar value on a tonne of CO2e, a cubic meter of water, or a displaced worker requires valuation choices that are contested. Internal carbon prices vary across organizations from $25 to $200 per tonne. Employment displacement valuations vary with assumptions about redeployment, retraining, and social-safety-net availability. SAV’s apparent precision conceals these valuation choices.
Reporting practice: disclose the valuation parameters (internal carbon price, water scarcity factor, displacement value) and report sensitivity to alternative parameters where parameters are contested.
Limit 2 — Scope choice
What counts as an externality of this feature, versus an externality of the overall organization, versus an externality of the broader economy? A rigid feature-level scoping may miss organization-wide effects (e.g., the AI program’s aggregate effect on employment); a rigid organization-level scoping may dilute feature-level accountability.
Reporting practice: report SAV at feature level with explicit scope notes, and separately report portfolio-level SAV with the scope differences disclosed.
Limit 3 — Time horizon
Environmental and social externalities play out over decades. An SAV computed on a 12-month horizon captures short-run effects; long-run effects (climate change, labor-market evolution) require longer horizons. Most SAV reporting today uses 12-month or multi-year horizons; longer horizons are methodologically immature.
Reporting practice: state the horizon explicitly; where longer horizons are strategically relevant, supplement the primary SAV with scenario analysis at longer horizons.
Cross-reference to Core Stream
EATE-Level-3/M3.5-Art15-Strategic-Value-Realization-Risk-Adjusted-Value-Frameworks.md#externality-adjusted— strategic-level risk-adjusted value framework.
Self-check
- A feature’s core realized value is $6M; externality accounting shows $0.2M in carbon costs, $0.1M in privacy externalities, and $0.4M in positive equity externalities. Compute the approximate pre-risk-adjustment SAV.
- An organization reports SAV for features where it is favourable and realized value where SAV is unfavourable. Which discipline marker is violated?
- An SAV calculation uses an internal carbon price of $50/t; the auditor suggests $100/t should be the benchmark. How does the VRR handle the difference?
- A CSRD-covered organization needs disclosures on E1 (climate), S1 (workforce), and G1 (governance) for an AI feature. Which externality categories in the SAV framework produce the disclosures?
Further reading
- European Sustainability Reporting Standards (ESRS), as adopted by the European Commission (2023).
- OECD AI Policy Observatory, environmental-compute publications.
- Published academic work on the social cost of carbon and alternative-parameter sensitivity.
Footnotes
-
European Commission, Commission Delegated Regulation (EU) 2023/2772 supplementing Directive 2013/34/EU as regards sustainability reporting standards (European Sustainability Reporting Standards, ESRS), 31 July 2023. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A32023R2772 ↩