Compliance & Assurance

The evidence behind lever ⑦ — the shift from occasional, sampled manual audits to algorithmic auditing of every digitally-verifiable transaction. Unlike the other levers this is a benchmark picture, not a per-country table: it sizes what assurance costs today and what verifiable evidence changes.

Canonical published output — distilled from research/compliance-assurance/.

Tax administration — assurance is sampled and labour-heavy

Indicator	Figure	Source
Cost of collection (operating cost ÷ net revenue)	≈ 0.65% (2023; 0.88% in 2014)	OECD / ISORA
Staff on audit / investigation / verification (large-taxpayer offices)	60.4%	OECD / ISORA
Audit adjustment rate	~60%	OECD / ISORA
Additional revenue raised through audits	3–4.2% of total collections	OECD / ISORA

Audits are sampled yet consume most verification staff and recover only a few percent of revenue — exactly the model algorithmic auditing of every verifiable invoice replaces.

The undetected pool — coverage, yield & deterrence

Sampled audit reaches a tiny share of transactions, so most non-compliance is never seen. Verifiable documents flip coverage toward ~100% at ≈ zero marginal cost and add a deterrence effect audit cannot.

Indicator	Figure	Source
Tax audit coverage	individuals 0.44%, corporations 0.74%	IRS Data Book FY2023
Random-audit non-compliance (the unbiased baseline)	~15% of tax (voluntary compliance ~85%)	IRS NRP / Tax Gap TY2022
Customs operating model	~96% of trade clears untouched	WCO PCA Guidelines
Customs random undervaluation gap	declared values 37% below reference; only 3.2% flagged	World Bank RCT, Madagascar (2020)
EU VAT compliance gap	€128bn (9.5%)	EU VAT Gap 2025
US gross tax gap	$696bn (TY2022)	IRS Pub. 5869
E-invoicing realised uplift	Mexico +14% declared rev; Italy €2–4bn/yr; Peru +8% VAT liabilities	IDB–CIAT / govt / IMF

The honest reading (two guardrails): (1) targeted ≠ random — risk-targeted audits show high hit rates because they self-select; only random programmes size the population pool. (2) An error-line rate is not a revenue rate — e.g. Australia’s random sample finds errors on ~32% of lines, but dollar-weighted leakage is far smaller (under 5%), and ~$0.25bn/yr is recovered via voluntary disclosure. The defensible large-pool case rests on lower-compliance economies (much bigger gaps) and the deterrence channel universal verification adds on top of detection — which is also what enlarges the incentive pool customs and tax can share with adopters.

Evidence-based assessment works (VAT-strong, CIT-thin)

Indicator	Figure	Source
Peru e-invoicing → reported value-added (4-year)	+21%	IMF WP/19/231
Peru e-invoicing → reported VAT liabilities (year 1)	+8% (chiefly deterrence)	IMF WP/19/231
First-year effect, concentrated in smaller firms	+5–7% sales/purchases	IMF
Mexico CFDI → declared revenue (3-year)	+14%	IDB–CIAT 2018
Italy SdI → additional VAT collected	€2–4bn/yr	Italian govt projections

Strong proof for VAT; the effect is chiefly deterrence (firms self-correct under universal visibility), not just detection. Corporate-tax evidence-based assessment remains the frontier (modelled as a qualitative upside, option A, not yet a number).

(Note: an earlier “+9.3% net tax revenue over 4 years” figure for Peru is not supported by the IMF primary source — net VAT payments dipped short-term as firms drew down VAT-credit stocks — and has been removed. The robust claims are the reported value-added / liabilities effects above.)

Customs & banks — the cost of assurance

Customs (WCO): post-clearance audit is explicitly risk-based / sampled and field audit carries “high administrative cost.” A worked case recovered ~US$26.6M on ~US$2.6M of audit cost — but only over the audited sample.
Banks: KYC review $1,500–3,500 per commercial client; AML/sanctions screening 95–98% false-positive rate; trade-finance loss-given-default just 0.10% / 0.02% (import/export LC) — so assurance cost, not default risk, is the binding constraint on lending, especially to SMEs (Fenergo; Wolfsberg/ICC; ICC Trade Register).

The fraction of transactions/applications where verifiable evidence is sufficient to auto-assess (excluding sanctioned/blacklisted and novel-risk cases) has no authoritative figure — it is a defended assumption, anchored on the 95–98% AML false-positive rate and trusted-trader coverage, used with a sensitivity range.

How this feeds the business cases

Lever ⑦ books the assurance cost saved (audit / due-diligence / KYC labour) plus risk reduction; the revenue and finance that better assurance enables are counted under ②③④. See the Assessment Methodology.

Sources

OECD Tax Administration 2025 / ISORA (cost of collection, audit staffing); WCO post-clearance audit; Wolfsberg/ICC/BAFT trade-finance due diligence; IMF (Peru e-invoicing); Fenergo / LexisNexis (KYC cost); ICC Trade Register. Full extraction in research/compliance-assurance/.