2026-10-22-claim-ledger
{
"schema_version": 1,
"research_date": "2026-10-22",
"scope": "Public primary enterprise evidence for the Knauf AI cost optimisation presentation",
"recommended_main_claims": [
"E01",
"E02"
],
"backup_claims": [
"E03",
"E04",
"E05",
"E06"
],
"sources": [
{
"id": "mckinsey-2026",
"publisher": "McKinsey & Company",
"title": "The state of AI in 2026: On the road to ROI",
"publication_date": "2026-08-25",
"retrieved": "2026-10-22",
"url": "https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai",
"stable_pdf_url": "https://www.mckinsey.com/~/media/mckinsey/business%20functions/quantumblack/our%20insights/the%20state%20of%20ai/the-state-of-ai-in-2026-on-the-road-to-roi-aug-2026.pdf",
"primary_source": true,
"evidence_type": "Self-reported online survey",
"methodology": {
"field_start": "2026-05-04",
"field_end": "2026-06-08",
"survey_n": 1719,
"countries": 97,
"weighting": "National contribution to global GDP",
"large_enterprise_share_percent": 36,
"item_specific_denominators_required": true,
"probability_sampling_documented": false,
"unit": "Respondent, not independently verified unique company"
},
"short_excerpt": "About 20 percent of respondents report that AI-related operating costs (including token costs) constrained their AI use.",
"excerpt_word_count": 17,
"inspection": "Read HTML article and official 31-page PDF, including Exhibits 2 and 8 and About the research."
},
{
"id": "datadog-2026",
"publisher": "Datadog",
"title": "State of AI Engineering 2026",
"publication_date": null,
"launch_date": "2026-04-21",
"retrieved": "2026-10-22",
"url": "https://www.datadoghq.com/state-of-ai-engineering/",
"official_launch_url": "https://investors.datadoghq.com/node/16711/pdf",
"primary_source": true,
"evidence_type": "Observed production LLM telemetry from vendor customer base",
"methodology": {
"broad_population": "More than 1,000 Datadog customers sending LLM telemetry; broad methodology says thousands of customer companies",
"fact_5_n": null,
"fact_5_matched_cohort_disclosed": false,
"fact_5_month_endpoints_verified": false,
"unit": "Organisation-level average tokens per LLM span, aggregated across organisations",
"selection": "Datadog customer base; software, infrastructure and cloud adoption bias"
},
"short_excerpt": "the average number of tokens used in customers’ requests more than doubled for median customers",
"excerpt_word_count": 15,
"licence_as_printed": "Report and images: CC BY-ND 4.0",
"inspection": "Read Fact 5, population/methodology, and official dated launch announcement. Current live report may have been updated since launch."
},
{
"id": "finops-2026",
"publisher": "FinOps Foundation / Linux Foundation",
"title": "State of FinOps 2026",
"publication_date": "2026-02-19",
"retrieved": "2026-10-22",
"url": "https://data.finops.org/",
"official_release_url": "https://www.linuxfoundation.org/press/state-of-finops-survey-ai-value-and-skills-top-priorities-as-finops-matures-across-technology-value-98-manage-ai-90-saas-64-licensing-48-data-center-1",
"author_keynote_pdf_url": "https://www.flexera.com/sites/default/files/Flexera-FinOps-Forward-2026-FinOps-Foundation-State-of-FinOps-Keynote.pdf",
"previous_edition_url": "https://data.finops.org/2025-report/",
"primary_source": true,
"evidence_type": "Annual self-selected FinOps community survey",
"methodology": {
"headline_respondents": 1192,
"keynote_cleaned_deduplicated_n": 966,
"ai_management_item_n": 693,
"skills_item_n": 695,
"field_dates_verified": false,
"unit": "Survey practitioner/respondent, not all enterprises",
"longitudinal_panel_verified": false
},
"short_excerpt": "98% of respondents now manage AI spend",
"excerpt_word_count": 7,
"inspection": "Read 2026 report, original Linux Foundation release, Foundation-authored keynote hosted by Flexera and 2025 report. Dedicated AI question distinguished from current-plus-planned scope table."
}
],
"claims": [
{
"id": "E01",
"status": "verified_primary_approved_with_scope",
"priority": "main",
"source_id": "mckinsey-2026",
"approved_short_wording": "About 1 in 5 respondents at AI-using organisations say operating costs constrained AI use.",
"value": 20,
"unit": "percent of eligible respondents",
"approximate": true,
"locator": "Exhibit 8; PDF printed page 15; article section While AI investments are increasing, operating costs are constraining AI use for some",
"denominator": "Respondents whose organisations regularly use AI in at least one business function; total item n not separately printed. Overall survey n=1,719 is not the item n.",
"limitations": [
"Self-report, not audited financial records",
"Not a universal constraint",
"Does not measure budget overrun or financial loss",
"Does not establish that agents cause the reported constraint"
],
"suggested_graphic": "Five equal marks, one accent; large 1 in 5; caption names AI-using survey respondents",
"chart_restrictions": [
"No extrapolation",
"No label stating 20% of AI budgets are wasted",
"Do not cite all 1,719 as eligible item denominator"
]
},
{
"id": "E02",
"status": "verified_primary_approved_with_scope",
"priority": "main",
"source_id": "datadog-2026",
"approved_short_wording": "Median customer's average tokens per LLM call: more than 2× year over year.",
"value_lower_bound_exclusive": 2,
"unit": "year-over-year ratio in organisation-average tokens per LLM span at median organisation",
"locator": "Fact 5 and Methodology > Fact 5",
"denominator": "Organisations sending Datadog LLM traces; exact Fact 5 sample n and matched cohort not disclosed",
"supplementary_measure": {
"percentile": 90,
"year_over_year_ratio_as_reported": 4
},
"limitations": [
"Vendor customer selection",
"No exact sample n or verified month endpoints",
"Not per accepted result or end-to-end task",
"Not a cost, spend or ROI measure",
"Changing workload/model/customer mix may contribute"
],
"suggested_graphic": "Single large >2× callout with exact per-call token unit and year-over-year caption; 4× kept in notes",
"chart_restrictions": [
"Do not draw an exact 2.0× bar",
"Do not relabel tokens as euros",
"Do not project the growth rate",
"Do not copy or recolour Datadog artwork"
]
},
{
"id": "E03",
"status": "verified_primary_approved_with_scope",
"priority": "backup",
"source_id": "mckinsey-2026",
"approved_short_wording": "Large-organisation respondents reporting agents at least scaling: 27% in 2025, 40% in 2026.",
"values": [
{
"year": 2025,
"percent": 27
},
{
"year": 2026,
"percent": 40
}
],
"change_percentage_points": 13,
"locator": "Exhibit 2, PDF printed page 6",
"denominator": "Organisations with annual revenue >=USD1bn; 2026 group n=595. Asked of regular AI users and rebased to full revenue-size group.",
"limitations": [
"Scaling in one or more functions, not enterprise-wide rollout",
"Repeated cross-sectional survey, not necessarily same respondents",
"No spend or ROI measurement"
],
"suggested_graphic": "Two bars on 0–100% baseline, direct labels 27% and 40%, note +13 percentage points"
},
{
"id": "E04",
"status": "verified_primary_conditional_on_dedicated_ai_section",
"priority": "backup",
"source_id": "finops-2026",
"approved_short_wording": "Surveyed FinOps practitioners managing some AI spend: 31% in 2024, 63% in 2025, 98% in 2026.",
"values": [
{
"year": 2024,
"percent": 31
},
{
"year": 2025,
"percent": 63
},
{
"year": 2026,
"percent": 98
}
],
"locator": "2026 report dedicated AI section Managing AI Value Becomes the Norm; Foundation keynote printed page 16; original publisher release",
"denominator": "2026 dedicated AI section n=693; prior editions have different respondents/denominators",
"limitations": [
"FinOps community sample, not all enterprises",
"Responsibility, not maturity or success",
"Current-plus-planned technology-scope table has different wording",
"Raw 1,192 and cleaned 966 survey totals should not replace item n"
],
"suggested_graphic": "Three discrete 0–100% bars with direct year labels; no future trajectory"
},
{
"id": "E05",
"status": "verified_primary_approved_with_scope",
"priority": "backup",
"source_id": "finops-2026",
"approved_short_wording": "58% of surveyed practitioners seek AI cost-management skills through people or tooling in the next year.",
"value": 58,
"unit": "percent selecting this desired skillset",
"locator": "Foundation-authored keynote printed page 9; report Team section skillsets chart",
"denominator": "Multi-select n=695",
"limitations": [
"Stated future skill/tool preference",
"Not the percentage with no capability today",
"Not a measured operational failure rate"
],
"suggested_graphic": "One horizontal bar, 0–100%, with people or tooling qualifier"
},
{
"id": "E06",
"status": "verified_primary_approved_with_scope_item_n_unavailable",
"priority": "backup",
"source_id": "mckinsey-2026",
"approved_short_wording": "28% of respondents report spending over 10% of enterprise-wide ICT budget on AI technologies.",
"value": 28,
"unit": "percent of respondents reporting budget share >10%",
"locator": "PDF printed page 16; article While AI investments are increasing section",
"denominator": "Item-specific n not separately printed; overall survey n=1,719",
"limitations": [
"Self-reported budget share",
"Not average budget share",
"All AI technologies, not just model tokens or APIs",
"Does not show waste or savings potential"
],
"suggested_graphic": "Large 28% with explicit budget-share threshold in caption; no pie chart implying 28% budget share"
}
],
"rejected_claims": [
{
"claim": "Executive engagement increases technology-selection influence by 2–4×",
"reason": "FinOps report executive summary comparator percentages conflict with main body and publisher release; also observational correlation, not causal effect."
},
{
"claim": "98% of all enterprises have AI cost controls",
"reason": "FinOps community sample and remit question cannot support all-enterprise adoption or control effectiveness."
},
{
"claim": "AI bills doubled because agents use more tokens",
"reason": "Datadog measures per-call token growth; prices, caching, mix, volume and task outcomes are not held constant, and causality is not identified."
},
{
"claim": "63% of companies get no value from AI",
"reason": "Absence of reported enterprise EBIT attribution is not absence of local value or evidence of loss."
}
]
}