The CDO's AI strategy must be grounded in organizational reality, not technology possibility. The most sophisticated AI capabilities mean nothing if your organization can't absorb and operationalize them.
Level 1, AI-Aware: Leadership understands AI's potential. Occasional POC projects. No production deployments. Most organizations are here.
Level 2, AI-Experimenting: Multiple POCs, some reaching production. Data infrastructure being built. ML talent being recruited. Results are inconsistent.
Level 3, AI-Scaling: Several production AI systems. MLOpsMLOpsMachine Learning Operations: combining ML and DevOps practices to industrialise, deploy, monitor, and retrain models reliably in production.View full definition → practices emerging. Feedback loops between model performance and business outcomes established. 15-20% of large enterprises.
Level 4, AI-Native: AI is embedded in core business processes. Culture of decision-making. Continuous experimentation. Significant derived from AI. Less than 5% of organizations.
The path from Level 1 to Level 4 takes 5-7 years for most organizations. CDOs who set realistic expectations and build systematically outperform those who promise Level 4 in 18 months.
Knowledge check
1. According to the lesson, what should fundamentally ground a CDO's AI strategy?
2. An organization runs multiple POCs, some of which reach production, is actively building data infrastructure and recruiting ML talent, but its results remain inconsistent. Which maturity level best describes it?
3. What distinguishes an AI-Native (Level 4) organization from an AI-Scaling (Level 3) one?
4. Select ALL statements that reflect the lesson's view on realistic AI roadmaps and maturity progression.
Select all the correct answers.
5. Select ALL correct statements about why hiring only data scientists is a common failure mode.
Select all the correct answers.
AI capability requires a specific team structure that most organizations don't yet have:
Data Scientists: Model development, feature engineering, experimental design. Build models. Typically PhD or MS in quantitative field.
ML Engineers: Production ML systems, model serving infrastructure, MLOpsMLOpsMachine Learning Operations: combining ML and DevOps practices to industrialise, deploy, monitor, and retrain models reliably in production.View full definition → pipelines. Take models to production and keep them there. Software engineering + ML expertise.
Data Engineers: Data pipelineData pipelineETL (Extract, Transform, Load) is a data integration process that pulls data from sources, reshapes it into a consistent format, and writes it into a target system.View full definition → infrastructure that ML models depend on. Feature pipelines, training data pipelines, monitoring pipelines.
Applied AI Researchers: For organizations doing cutting-edge work, researchers who develop new methods. Usually only in large tech companies or organizations with significant AI competitive differentiation.
AI Product Managers: Define AI product requirements, prioritize use cases, translate between business and technical. Critical but often missing.
The common failure: hiring data scientists without the complementary roles. Data scientists without ML engineers can't get models to production. Models without data engineers have unreliable training data. Use cases without AI PMs drift from business needs.
An AI strategy requires decisions about which capabilities to build internally vs. partner for vs. buy off-shelf:
Build internally: Proprietary models trained on proprietary data where the AI capability is central to competitive differentiation. Example: a fintech building a proprietary credit scoring model trained on unique transaction data.
Partner (fine-tune foundation models): Use a foundation model provider (OpenAI, Anthropic, Google, Mistral) and adapt their models with your data. Best balance of capability and investment for most enterprise use cases.
Buy (SaaS AI): Use pre-built AI products that solve specific problems. Salesforce Einstein for CRMCRMCustomer Relationship Management: software and strategy to manage and analyse customer interactions throughout their lifecycle.View full definition → intelligence, Adobe Sensei for marketing personalization, Workday AI for HR analytics. Fast deployment, limited customization.
Most organizations need all three tiers. The strategy decisions are about which tier each use case belongs in.
AI investment is not a one-time project, it's an ongoing capability investment:
Infrastructure (30-40% of AI budget): Data platform, compute (GPU infrastructure or cloud), MLOpsMLOpsMachine Learning Operations: combining ML and DevOps practices to industrialise, deploy, monitor, and retrain models reliably in production.View full definition → tooling. The foundation everything else depends on.
Talent (40-50%): The most constrained resource. Competitive salaries, meaningful work, and career development are the retention factors for ML talent. Underpaying is false economy.
Vendor/platform fees (10-20%): LLMLLMA Large Language Model is an AI system trained on vast text data to predict and generate language, enabling tasks like writing, summarizing, and answering questions.View full definition → APIAPIApplication Programming Interface: a standardised interface that lets applications communicate and exchange data without knowing each other's internal workings.View full definition → costs, MLOpsMLOpsMachine Learning Operations: combining ML and DevOps practices to industrialise, deploy, monitor, and retrain models reliably in production.View full definition → platforms, tools, vector databases.
Governance and risk (5-10%): Model auditing, bias testing, legal review, output monitoring. Consistently underinvested and consistently the source of expensive incidents.
The organizations that get best results from AI invest in infrastructure and talent first, not in model novelty. A great model on weak infrastructure, maintained by a struggling team, performs worse in production than a simpler model on solid foundations.
1. Quel est le niveau de maturité IA le plus courant parmi les grandes entreprises aujourd'hui ?
A) Niveau 4 (AI-Native)
B) Niveau 3 (AI-Scaling)
C) Niveau 1-2 (AI-Aware / AI-Experimenting)
D) Niveau 5 (AGI-Ready)
Réponse: C
2. Quel rôle est critique mais souvent manquant dans les équipes AI enterprise ?
A) Data Scientist
B) ML Engineer
C) AI Product Manager, qui traduit entre les besoins métier et les capacités techniques
D) Applied AI Researcher
Réponse: C
3. Quelle répartition budgétaire est recommandée pour le talent dans un budget AI mature ?
A) 10-20% du budget AI
B) 40-50% du budget AI
C) 70-80% du budget AI
D) 5-10% du budget AI
Réponse: B
These actions are compiled in the role's Playbook.