Build vs Buy Enterprise AI

The decision that determines your 3-year TCO, switching cost, and competitive differentiation. Seven dimensions, scored, with cost benchmarks from 60+ projects.

TL;DR: Buy when the use case is generic (Copilot for Office, ChatGPT Enterprise) and your data is in M365/Google Workspace. Build when (a) compliance demands on-prem / data residency, (b) integration with proprietary CRM/ERP/EMR is required, or (c) the AI is the product differentiator. Hybrid is most common: buy the LLM, build the orchestration + RAG + agent layer.

7-Dimension Rubric

Dimension	Buy (off-the-shelf)	Build (custom)	Winner
Time to first value	1-4 weeks (SaaS)	3-6 months (MVP)	Buy
Year-1 cost	$20-200K (seats × $20-30/user/mo)	$80-450K (impl + ops)	Buy
3-year TCO (100-500 users)	$300K-$1.5M (subscription)	$250K-$900K (decreasing slope)	Build (often)
Data residency / compliance	Vendor-controlled, US/EU/UK only	Your choice (on-prem, sovereign cloud)	Build
Custom integration depth	Standard connectors (M365, Slack, etc.)	Any system with API or DB access	Build
Vendor lock-in / switching cost	High — data + prompts + workflows tied to vendor	Low — you own the IP, can swap LLM provider	Build
Speed of capability evolution	Fast — vendor R&D rolls out monthly	Your roadmap, your speed	Buy

3-Year Total Cost (300-user organization, mid-complexity use case)

📦 Buy: Microsoft Copilot for Business

$1.08M

300 seats × $30/mo × 36 mo = $324K subscription
+ 1 FTE governance/training ($150K × 3 = $450K)
+ 1 FTE integration developer ($120K × 3 = $360K — half time, 0.5 FTE × 3 × $120K)
— Subtract $60K customizations not built
Total: ~$1.08M over 3 years

🔨 Build: Custom AI Platform (RAG + chat + agents)

$720K

Implementation Y1: $180K (vendor build, 4-6 months)
Operational Y1: $80K (LLM tokens + infra + 0.5 FTE maintenance)
Y2-Y3 operational: $120K × 2 = $240K
Internal AI champion: $150K × 0.5 FTE × 3 = $225K
Y2-Y3 enhancements: $100K
Total: ~$720K over 3 years

🔀 Hybrid: Buy LLM, Build orchestration

$560K

LLM API costs (Anthropic/OpenAI): $4-8K/mo = $200K over 3y
Implementation Y1: $120K (orchestration + RAG layer)
Operational Y2-Y3: $90K × 2
AI champion: 0.4 FTE × $150K × 3 = $180K
Y2-Y3 enhancements: $30K
Total: ~$560K over 3 years

When Each Path Wins

📦 Buy off-the-shelf when:

Your data is already in Microsoft 365 or Google Workspace (Copilot integrates natively)
The use case is generic productivity: meeting summaries, email drafts, Excel formula assistance
You have <100 users and a small IT team
Regulatory pressure is low (no medical, legal, financial data)
You expect to iterate slowly (vendor pace = your pace is fine)

🔨 Build custom when:

Compliance demands on-prem deployment or data residency (medical, legal, defense, government)
Integration with proprietary CRM/ERP/EMR/MES is core to the value
The AI capability is your product differentiator (you're selling AI-powered services, not using AI for back-office)
You have >500 users (subscription costs cross over)
You have specialized domain language/workflows that Copilot doesn't model

🔀 Hybrid (buy LLM, build orchestration) when:

You want vendor R&D velocity for the model layer, but compliance/integration needs custom layer above
You expect LLM provider commoditization (multi-vendor flexibility matters)
Your differentiation is in the workflow + data, not the model
This is the most common path for mid-to-large enterprises in 2026

Hidden Costs Most Buyers Miss

📦 Buy (Copilot/ChatGPT Enterprise/etc.):

Governance overhead. 0.5-1.5 FTE for policy, training, exception handling, audit.
Per-seat sprawl. "Just buy more seats" mentality → 2-3x your initial estimate by Y3.
Switching cost when you outgrow it. Workflows + chat history + tuning locked to vendor.

🔨 Build:

Internal champion bandwidth. 0.5-1.0 FTE on your team for direction, prompts, eval, prioritization.
Compliance evolution. EU AI Act, US state regs, industry rules — your roadmap, your responsibility.
Eval infrastructure. You need test sets, regression detection, prompt versioning.

Want this scored on your specific stack?

30-minute architecture review — no sales push. We'll score your data residency, compliance, integration, and team posture against the rubric and give a recommendation in writing.

📞 Book architecture review 📐 Or run the Project Estimator