AI Use-Case Portfolio Prioritizer

AI CAPABILITY ASSESSMENT (20-person team)

TRIAGE FRAMEWORK
Evaluate each capability on: strategic importance, engineering lift, cost of buying, timeline to market

BUILD (core to competitive advantage)
→ Capability: Multi-agent orchestration engine
  Strategic importance: CRITICAL (our product differentiator)
  Lift: 8 weeks, 3 senior engineers
  Cost to build: 60K (salary)
  Cost to buy: no good off-the-shelf solution, would require adaptation 40K
  Decision: BUILD. Buying would require so much customization it's equivalent to building anyway.
  Timeline: 8 weeks, start Q1 2027

→ Capability: Domain-specific fine-tuning on customer workflows
  Strategic importance: HIGH (improves accuracy by 15%)
  Lift: 12 weeks, 2 engineers + ML researcher (contractor)
  Cost to build: 85K
  Cost to buy: API access to model provider 20K/month
  Decision: BUILD. ROI positive in 5 months; buying creates vendor lock-in.
  Timeline: 12 weeks, start Q1 2027 after orchestration engine stabilizes

BUY (commoditized, not core)
→ Capability: Text-to-image generation
  Strategic importance: NICE-TO-HAVE (feature, not differentiator)
  Lift: 6 weeks, 1 engineer
  Cost to buy: Midjourney or Stable Diffusion API 0.001 per image
  Decision: BUY. Cheap, fast to integrate, no maintenance burden.
  Vendor: Stable Diffusion (self-hosted option for control)

→ Capability: Email classification and routing
  Strategic importance: MEDIUM (improves user experience)
  Lift: 4 weeks, 1 engineer
  Cost to buy: HubSpot or custom ML 500/month
  Decision: BUY. Out-of-the-box solutions are mature.
  Vendor: HubSpot (already integrated with CRM)

→ Capability: Real-time speech transcription
  Strategic importance: LOW (nice feature, not core)
  Lift: 3 weeks, 1 engineer
  Cost to buy: Rev.ai or Deepgram 0.0001 per minute
  Decision: BUY. Cheap, reliable, no maintenance.
  Vendor: Deepgram (better latency for real-time)

TEAM ALLOCATION (20 people)
Engineering capacity: 8 engineers total
  → 3 assigned to BUILD initiatives (multi-agent engine + fine-tuning) = 20 weeks of work
  → 2 assigned to integrations (bought capabilities) = 12 weeks of work
  → 3 assigned to product features and bug fixes (no AI work)
ML/Research: 1 contractor (20 hours/week, 8 weeks)
Product: 2 to define requirements and test
Other: 7 (sales, ops, support, finance)

COST SUMMARY
Build: 85K (salaries, 10 weeks)
Buy APIs: 2,400/month (varies by volume)
Contractor: 18K (8 weeks)
Total investment: 103K + ongoing API costs
Expected ROI: 40% accuracy improvement, 2-month time-to-market vs competitors

TRAP TO AVOID
Do NOT try to build everything. You'll spread engineers too thin and ship nothing. Prioritize orchestration engine (your differentiator) and buy the rest