How to Build an AI Readiness Assessment for Your Organization

Before deploying AI, you need to know if your organization can actually support it. 73% of enterprise AI projects fail — not because of model quality, but because of missing infrastructure, bad data, and organizational unreadiness. This assessment gives you an honest, quantified view of where you stand and what to fix first.

The most dangerous situation: leadership says “we need AI” without understanding the prerequisites. The result is a $500K investment in models that never reach production because the data is fragmented, the infrastructure can’t serve predictions, and nobody on the team knows how to operationalize a model.

The 5-Pillar AI Readiness Framework

Pillar	Weight	What It Measures
Data Maturity	30%	Quality, accessibility, governance of your data
Infrastructure	20%	Compute, storage, MLOps tooling
Talent & Skills	20%	Engineering and data science capability
Governance	15%	Ethics, compliance, risk management
Culture	15%	Leadership support, change management

Quick Self-Test

Before diving into the full assessment, answer these five questions:

Question	Yes	No
Can you access clean, documented data for your target use case?	+20	0
Do you have GPUs or cloud ML compute available?	+20	0
Is there at least one data scientist or ML engineer on staff?	+20	0
Does leadership have a budget allocated for AI specifically?	+20	0
Have you completed at least one data-driven project successfully?	+20	0

Score 80-100: Ready for production pilots. Use the full assessment to identify gaps. Score 40-60: Foundation building. Address the “No” answers before investing in AI. Score 0-20: Not ready. Focus on data infrastructure and hiring first.

Step 1: Score Data Maturity (30%)

1.1 Data Quality Audit

import pandas as pd

def score_data_quality(df: pd.DataFrame) -> dict:
    """Score a dataset's quality on key dimensions"""
    total_cells = df.size
    null_cells = df.isnull().sum().sum()

    scores = {
        "completeness": round((1 - null_cells / total_cells) * 100, 1),
        "uniqueness": round(df.drop_duplicates().shape[0] / df.shape[0] * 100, 1),
        "consistency": _check_consistency(df),
        "freshness": _check_freshness(df),
    }

    scores["overall"] = round(
        sum(scores.values()) / len(scores), 1
    )
    return scores

def _check_consistency(df):
    """Check for format consistency in key columns"""
    issues = 0
    for col in df.select_dtypes(include='object').columns:
        # Check for mixed case inconsistency
        if df[col].str.lower().nunique() < df[col].nunique():
            issues += 1
    consistency = max(0, 100 - issues * 10)
    return consistency

def _check_freshness(df):
    """Check timestamp columns for data freshness"""
    date_cols = df.select_dtypes(include='datetime64').columns
    if len(date_cols) == 0:
        return 50  # Can't evaluate
    latest = df[date_cols].max().max()
    days_old = (pd.Timestamp.now() - latest).days
    if days_old < 1: return 100
    if days_old < 7: return 85
    if days_old < 30: return 65
    return 40

1.2 Data Accessibility Checklist

Question	Score	Why It Matters
Can analysts query production data without DBA involvement?	/10	Self-service access = faster experiments
Is there a central data catalog (e.g., DataHub, Collibra)?	/10	Discoverability prevents data hoarding
Are datasets documented with schema definitions?	/10	Undocumented data = wasted weeks
Is there a self-service data access request process?	/10	Bottlenecked access kills projects
Can you join data across 3+ source systems?	/10	AI needs cross-functional data

1.3 Data Volume Readiness

Use Case	Minimum Data Required	Your Data
Classification (binary)	1,000 labeled examples per class	___
Regression	5,000+ labeled rows	___
NLP (text classification)	5,000+ labeled documents	___
Computer vision	500+ labeled images per class	___
RAG / document Q&A	100+ documents (unlabeled OK)	___
Anomaly detection	10,000+ normal examples	___

Step 2: Score Infrastructure Readiness (20%)

2.1 Compute Assessment

# Check GPU availability
nvidia-smi --query-gpu=name,memory.total,driver_version \
  --format=csv,noheader 2>/dev/null || echo "No GPU detected"

# Check available RAM
free -h | head -2

# Check Docker availability
docker --version 2>/dev/null || echo "Docker not installed"

# Check Kubernetes
kubectl cluster-info 2>/dev/null || echo "No Kubernetes cluster"

2.2 Infrastructure Scoring

Capability	Level 1 (Basic)	Level 2 (Ready)	Level 3 (Advanced)
Compute	Shared VMs	Dedicated GPU instances	Auto-scaling GPU clusters
Storage	Local/NAS	Cloud object storage	Lakehouse with governance
MLOps	Manual scripts	MLflow / Weights & Biases	Full Kubeflow / SageMaker
Monitoring	Basic logs	APM + custom metrics	AI-specific observability
Networking	Public internet	VPN/Private endpoints	Zero-trust architecture

2.3 Minimum Infrastructure by AI Type

AI Application	GPU?	Minimum RAM	Storage	Network
RAG / chatbot (API-based)	No	8 GB	10 GB	Internet (API calls)
RAG / chatbot (self-hosted)	Yes (16 GB VRAM)	32 GB	100 GB	Internal
Fine-tuning LLMs	Yes (40+ GB VRAM)	64 GB	500 GB	Internal
Traditional ML (tabular)	No	16 GB	Depends on data	Internal
Computer vision	Yes (8+ GB VRAM)	32 GB	100+ GB	Internal + edge

Step 3: Score Talent & Skills (20%)

Skills Matrix

Skill Area	Minimum for AI Readiness	Assessment Method
Data Engineering	2+ engineers who can build ETL pipelines	Review recent pipeline work
ML/Data Science	1+ scientist who can train & evaluate models	Technical interview
MLOps/DevOps	1+ engineer who can containerize & deploy	Deploy a test model
Data Literacy	Managers can interpret model outputs	Run a decision exercise
AI Ethics	Someone owns responsible AI policy	Review policy document

Hiring vs Upskilling Decision

Gap	Hire	Upskill	Outsource
No data engineers	✅ Hire (core capability)	❌ Too specialized	Temporary contractors
No ML engineer	✅ Hire if AI is strategic	✅ If strong devs exist	✅ Pilot projects
No MLOps	✅ If scaling	✅ DevOps → MLOps path	✅ Managed platforms
Low data literacy	❌ Not a role	✅ Workshop-based training	❌ Must be internal

# Simple skills gap calculator
skills = {
    "data_engineering": {"current": 2, "needed": 3},
    "ml_data_science": {"current": 1, "needed": 2},
    "mlops": {"current": 0, "needed": 1},
    "data_literacy": {"current": 60, "needed": 80},  # % of managers
    "ai_ethics": {"current": 0, "needed": 1},
}

for skill, counts in skills.items():
    gap = counts["needed"] - counts["current"]
    status = "✅ Met" if gap <= 0 else f"⚠️ Gap: {gap}"
    print(f"  {skill}: {status}")

Step 4: Score Governance Readiness (15%)

Governance Checklist

Data classification policy — Is data labeled (public/internal/confidential/restricted)?
AI usage policy — Are there rules for how AI can be used with company data?
Model risk framework — Who reviews and approves models before production?
Bias testing protocol — Do you test for fairness across protected classes?
Compliance mapping — Have you mapped AI use cases to regulatory requirements (GDPR, CCPA, EU AI Act)?
Incident response — What happens when an AI system produces harmful output?
Audit trail — Can you explain any model decision after the fact?

Step 5: Score Culture & Leadership (15%)

Culture Assessment

Signal	Points
C-suite sponsor for AI initiatives	+20
Dedicated AI budget (not borrowed from IT)	+20
Cross-functional AI steering committee	+15
Pilot projects completed (even if small)	+15
Data-driven decision-making culture	+15
Willingness to fail and iterate	+15

Step 6: Calculate Your Overall Score

def calculate_ai_readiness(scores: dict) -> dict:
    weights = {
        "data_maturity": 0.30,
        "infrastructure": 0.20,
        "talent_skills": 0.20,
        "governance": 0.15,
        "culture": 0.15,
    }

    weighted_score = sum(
        scores[pillar] * weights[pillar]
        for pillar in weights
    )

    tier = (
        "🟢 AI-Ready" if weighted_score >= 75 else
        "🟡 Foundation Building" if weighted_score >= 50 else
        "🔴 Not Ready — Build Foundations First"
    )

    return {
        "overall_score": round(weighted_score, 1),
        "tier": tier,
        "pillar_scores": scores,
        "recommendation": _get_recommendation(scores)
    }

def _get_recommendation(scores):
    weakest = min(scores, key=scores.get)
    return f"Priority: Strengthen '{weakest}' (score: {scores[weakest]})"

Interpretation Guide

Score Range	Tier	Action	Timeline
75-100	AI-Ready	Proceed with production pilots	Now
50-74	Foundation Building	Address gaps, run contained experiments	3-6 months
25-49	Early Stage	Invest in data + skills before AI	6-12 months
0-24	Not Ready	Focus on digital transformation basics	12-18 months

Readiness Assessment Checklist

:::note[Source] This guide is derived from operational intelligence at Garnet Grid Consulting. Try the free AI Readiness Assessment Tool or get a Premium AI Readiness Report. :::