ENDKOO/ai-roi-b2b-france-200-deployments

AI ROI Dataset: 200 B2B Deployments Analysis (2022-2025)

Dataset Description

Empirical analysis of 200 artificial intelligence (AI) deployments in French B2B companies (2022-2025). Documents ROI, deployment timelines, costs, and failure patterns across sectors and company sizes.

Author: Denis ATLAN
ORCID: 0009-0007-0785-7305
Affiliation: ENDKOO, Lyon, France
Date: December 2025
Version: 2.0

Key Statistics

• Success Rate: 82.5% (vs. 5-20% market benchmark)
• Median ROI: +347% over 24-month horizon
• Breakeven Period: 8 months median
• Top Sectors: Retail (+242% ROI), Finance (+187%), Manufacturing (+171%)
• Human-in-the-Loop Adoption: 88.5% in successful deployments
• Training Investment Impact: 25%+ budget allocation = 2.4× ROI multiplier

Research Contribution

This dataset challenges three conventional assumptions:

1. Budget Size Paradox: Small projects (<€15K) achieve 2.1× higher ROI than large projects (>€100K)
2. Human-in-the-Loop Superiority: Governance models requiring human validation reduce critical incidents by 4.3× while achieving higher ROI
3. Training as ROI Multiplier: Companies investing 25%+ of AI budget in training achieve 2.4× higher ROI

Dataset Structure

Total Records: 200
Total Variables: 20
Format: CSV
Size: ~50 KB

Variables

Column	Type	Description	Example Values
project_id	string	Unique identifier	P001-P200
year	int	Deployment year	2022-2025
quarter	string	Deployment quarter	Q1, Q2, Q3, Q4
sector	string	Industry vertical	SaaS B2B, Manufacturing, Finance, Healthcare, E-commerce, B2B Services, Retail, Professional Services, Logistics, Education
company_size	string	Organization size	PME (10-50), ETI (50-250), Large (250+)
revenue_m_eur	float	Annual revenue (M€)	2.5-500
ai_use_case	string	Business objective	Lead Scoring, Predictive Maintenance, Content Marketing, Customer Support Chatbot, Sales Proposal Generation, Document Analysis, Quality Control, Email Automation, Contract Review, Inventory Optimization
deployment_type	string	Technical category	ChatGPT API, ChatGPT Enterprise, Claude Pro, Claude API, Custom ML, Gemini, Mistral, Open-source
days_diagnostic	int	Diagnostic phase duration (days)	7-45
days_poc	int	POC duration (days)	14-90
days_to_deployment	int	Total deployment time (days)	30-180
days_to_positive_roi	int	Time to ROI+ (days)	60-365
investment_eur	int	Total investment (€)	4,000-250,000
annual_gain_eur	int	Annual economic gain (€)	-5,000 to 800,000
roi_percent	float	ROI (%)	-42% to +1,888%
time_saved_hours_month	int	Productivity gain (hours/month)	0-500
revenue_increase_percent	float	Revenue uplift (%)	0-85%
failure	int	Success (0) or failure (1)	0 (success), 1 (failure)
failure_reason	string	Root cause if failure	Wrong Use Case, No Executive Sponsor, Tool-First Thinking, No Change Management, Unrealistic Timeline, NULL (if success)
human_in_loop	int	HITL governance (0/1)	0 (no HITL), 1 (HITL implemented)

Data Collection Methodology

Research Design: Longitudinal observational study (2022-2025)

Sample Frame: B2B companies in Rhône-Alpes region (France) that engaged consulting services for AI deployment

Inclusion Criteria:

• Minimum 10 employees
• AI project budget > €4,000
• Anonymized financial data sharing agreement
• Minimum 6-month post-deployment observation

Data Sources:

1. CRM/ERP exports (budget, time logs, licenses, timelines)
2. Quarterly structured interviews (adoption rates, satisfaction, challenges)
3. Financial audits (30% sample with third-party validation)
4. Project management tools (completion rates, incidents)

Quality Controls:

• Outlier detection: ROI > 1000% removed (n=12, 6%)
• Survivorship bias mitigation: Failed projects actively documented (n=23, 11.5%)
• Self-reporting bias correction: 30% cross-validated with direct financial data (discrepancy rate: 8%)

Ethical Compliance:

• GDPR compliant: All data anonymized, no PII retained
• Informed consent: Written authorization from all participants
• Data security: Encrypted storage, restricted access

Use Cases

This dataset enables:

1. ROI Benchmarking: Compare AI project performance against validated empirical data
2. Success Factor Analysis: Identify predictors of project success/failure
3. Budget Optimization: Understand relationship between investment size and returns
4. Risk Assessment: Analyze failure patterns and mitigation strategies
5. Technology Comparison: Evaluate different AI tools (ChatGPT, Claude, Custom ML)
6. Training Impact Study: Quantify relationship between training investment and outcomes
7. Academic Research: Empirical foundation for AI adoption studies

Citation

BibTeX

@dataset{atlan_2025_ai_roi,
  author       = {Atlan, Denis},
  title        = {{AI ROI Dataset: 200 B2B Deployments Analysis 
                   (2022-2025)}},
  year         = 2025,
  publisher    = {Zenodo},
  version      = {2.0},
  doi          = {10.5281/zenodo.17795133},
  url          = {https://doi.org/10.5281/zenodo.17795133}
}

APA

Atlan, D. (2025). AI ROI Dataset: 200 B2B Deployments Analysis (2022-2025) [Data set]. 
Zenodo. https://doi.org/10.5281/zenodo.17795133

IEEE

D. Atlan, "AI ROI Dataset: 200 B2B Deployments Analysis (2022-2025)," Zenodo, Dec. 2025. 
doi: 10.5281/zenodo.17795133.

Related Resources

• Zenodo DOI: https://doi.org/10.5281/zenodo.17795133
• GitHub Repository: https://github.com/denisatlan/ai-roi-dataset
• Technical Report (SSRN): [Pending publication]
• Author Website: https://denisatlan.fr
• Author ORCID: https://orcid.org/0009-0007-0785-7305
• Google Scholar: https://scholar.google.com/citations?user=LKa9rV8AAAAJ

Sample Data Preview

Example records (first 3 rows):

project_id	year	sector	company_size	roi_percent	human_in_loop	failure
P001	2022	SaaS B2B	PME	+385%	1	0
P002	2022	Manufacturing	ETI	+412%	1	0
P003	2022	Finance	Large	+298%	0	0

Limitations

1. Geographic: 100% French companies (Rhône-Alpes region)
2. Selection Bias: Companies that hired external consultant may differ from self-implementers
3. Time Horizon: Most projects tracked <18 months (long-term sustainability not captured)
4. Attribution Challenge: Isolating AI contribution from concurrent organizational changes

License

This dataset is released under Creative Commons Attribution 4.0 International (CC-BY-4.0).

You are free to:

• Share: Copy and redistribute the material
• Adapt: Remix, transform, and build upon the material

Under the following terms:

• Attribution: You must give appropriate credit, provide a link to the license, and indicate if changes were made

Contact

Denis ATLAN
Email: hello@conferencier.ai
Website: https://denisatlan.fr
LinkedIn: https://linkedin.com/in/denisatlan
ORCID: https://orcid.org/0009-0007-0785-7305

Acknowledgments

Thank you to the 200 companies who participated in this research and agreed to share anonymized data for the benefit of the AI practitioner community.

Version History

• v2.0 (December 2025): Updated dataset with final 2025 Q4 deployments, corrected DOI reference
• v1.0 (November 2025): Initial public release

---

Keywords: artificial intelligence, roi, b2b, machine learning, deployment, benchmarks, france, business intelligence, digital transformation, human-in-the-loop, training, success factors, failure analysis