# The ai-first.company evidence: where to start

Move by menus: open the part that fits the question (call open with the arguments shown), then call get_evidence with the ids of the rows you pick. Every list shows all its rows, or splits them into a menu when there are too many, so nothing is hidden. Choose by meaning, whatever the language of the question; the evidence is in English. Answer the user in their language.

## By theme: practices, and the cases and studies behind them

Open a theme with `open list=practices theme=…`, `open list=cases theme=…` or `open list=studies theme=…`.

- `process-redesign`: Process redesign, changing how the work is done, mapping tasks, human plus AI (4 practices, 16 cases, 18 studies)
- `people-and-skills`: People and skills, training, skills, jobs, fear and learning (6 practices, 13 cases, 22 studies)
- `data-and-tools`: Data and tools, which tools, data quality, what the AI shows, setting limits (5 practices, 18 cases, 5 studies)
- `strategy`: Strategy, where to start and what to aim for (3 practices, 11 cases, 8 studies)
- `governance-and-risk`: Governance and risk, owners, policies, data rules, checking output, the law, mistakes (8 practices, 27 cases, 9 studies)
- `measurement`: Measurement, baselines, time saved, the cost of checking, readiness scores (4 practices, 14 cases, 20 studies)

## Statistics

Claim checks of widely repeated AI figures, all 7:

- `claim-check:us-small-businesses-using-ai` **“66% of small businesses use AI”** (verdict: Needs correcting)
- `claim-check:mit-95-percent-of-ai-pilots-fail` **“95% of generative AI pilots at companies are failing (MIT)”** (verdict: Needs correcting)
- `claim-check:bcg-ai-maturity-matrix-rates-economies` **“BCG’s AI Maturity Matrix shows how mature companies are”** (verdict: Needs correcting)
- `claim-check:uk-copilot-26-minutes-a-day` **“Copilot saves UK civil servants 26 minutes a day”** (verdict: Needs correcting)
- `claim-check:microsoft-responsible-ai-maturity-model` **“Microsoft has a maturity model for responsible AI”** (verdict: Confirmed)
- `claim-check:mit-sloan-path-to-digital-transformation` **“MIT Sloan shows that the barriers to digital transformation are mainly cultural”** (verdict: No support found)
- `claim-check:cisco-ai-readiness-index-fully-ready` **“Only 13% of companies are ready for AI”** (verdict: Needs correcting)

Other figures, by what they measure:

- `open list=figures line=time-freed`: Time freed (10 calculated by the study’s authors, 7 measured in a study, 5 self-reported): 22
- `open list=figures line=time-put-to-use`: Time put to use (2 calculated by the study’s authors, 9 measured in a study, 7 self-reported, 2 reported, not measured): 20
- `open list=figures line=background`: Background (8 calculated by the study’s authors, 6 reported, not measured, 4 measured in a study): 18
- `open list=figures line=running-cost`: Running cost (6 self-reported, 2 calculated by the study’s authors, 3 vendor list price, 5 reported, not measured): 16
- `open list=figures line=risk-what-it-cost-others`: Risk: what it cost others (2 reported, not measured, 10 court or regulator record): 12

## Company cases (39), by the kind of work AI was used for

- `open list=cases task=customer-service-and-support`: Customer service and support: 8
- `open list=cases task=legal-research-drafting-and-filings`: Legal research, drafting and filings: 7
- `open list=cases task=sales-and-the-online-shop`: Sales and the online shop: 5
- `open list=cases task=documents-reports-and-paperwork`: Documents, reports and paperwork: 4
- `open list=cases task=pricing-quoting-and-forecasting`: Pricing, quoting and forecasting: 4
- `open list=cases task=general-office-assistant`: General office assistant: 3
- `open list=cases task=marketing-advertising-and-content`: Marketing, advertising and content: 3
- `open list=cases task=production-quality-and-logistics`: Production, quality and logistics: 3
- `open list=cases task=product-development`: Product development: 2
- `open list=cases task=hiring-and-hr`: Hiring and HR: 1
- `open list=cases task=security-fraud-and-monitoring`: Security, fraud and monitoring: 1
- `open list=cases task=software-and-it`: Software and IT: 1

By how they turned out:

- `open list=cases outcome=measured`: Measured result (researchers or an official body measured the result): 9
- `open list=cases outcome=company-reported`: Company-reported result (the company published the figures itself): 13
- `open list=cases outcome=went-wrong`: Went wrong (something failed, and there is a public record): 15
- `open list=cases outcome=walked-back`: Partly walked back (the company reversed part of what it did or claimed): 2

Cases can also be opened by `theme`, `industry` (professional-services, agencies, ecommerce-retail, manufacturing, or `other` for none of these) and company size; filters combine.

## Assessments (21): questionnaires, tests and task maps

- `open list=assessments kind=readiness-surveys`: Readiness surveys: 4
- `open list=assessments kind=attitude-and-trust-surveys`: Attitude and trust surveys: 4
- `open list=assessments kind=culture-and-speaking-up-surveys`: Culture and speaking-up surveys: 4
- `open list=assessments kind=ai-knowledge-tests-and-self-ratings`: AI knowledge tests and self-ratings: 3
- `open list=assessments kind=task-and-job-maps`: Task and job maps: 3
- `open list=assessments kind=interview-guides-and-checklists`: Interview guides and checklists: 3

## Values

- Company size: pass `employees`, or `size` = micro-small (1–49 employees), medium (50–249), mid-market (250–999), enterprise (1,000+). 7 of 39 cases have no recorded size; size filters keep them.
- Practice stage: explore, adopt, redesign, run. Priority: do first, do next, do later (editorial judgement, not evidence).
- Evidence rating (practices, assessments): Strong (several rigorous studies agree), Moderate (one rigorous study, or several agreeing studies or large surveys), Emerging (several company cases with numbers, no comparison group), Common practice (widely recommended, untested). Flags: contested (studies disagree), context-dependent (helps in some conditions, harms in others). A rating says how well tested, not how big the gain.
- Claim-check verdicts: Confirmed, Needs correcting, No support found. A figure a company reports about itself is not a measured result.

Rating method: https://www.ai-first.company/evidence/how-we-rate/?utm_source=mcp&utm_medium=01&utm_campaign=promotion

Source: ai-first.company (CC BY 4.0). Ratings explained at https://www.ai-first.company/evidence/how-we-rate/?utm_source=mcp&utm_medium=01&utm_campaign=promotion