Cross-functional Collaboration — Khi Không Chỉ Nói Chuyện Với Engineers

PM nói: "Chúng ta cần feature này ASAP"
Nghĩa là: Có competitive pressure hoặc customer demand

PM nói: "Đơn giản thôi, thêm 1 nút"
Nghĩa là: PM chưa hiểu technical complexity

PM nói: "User cần real-time updates"
Nghĩa có thể: "User cần thấy update trong 5 giây"
               (không phải milliseconds)

Cách respond:
  Đừng: "Cái này phức tạp lắm" (PM nghe = "engineer resist change")
  Nên: "Để tôi break down thành options với trade-offs..."

1. Ask "Why" trước "How"
   PM: "Thêm feature export CSV"
   Bạn: "User dùng CSV để làm gì? Có thể API integration 
         tiện hơn cho power users?"
   → Đôi khi solution tốt hơn ≠ solution PM yêu cầu

2. Đề xuất simpler alternatives
   PM: "Build recommendation engine"
   Bạn: "MVP có thể là 'most popular items' — 
         đủ tốt 80% cases, build trong 2 ngày thay vì 2 tháng.
         Sau đó iterate dựa trên data."

3. Quantify trade-offs
   "Option A: 2 tuần, cover 90% use cases
    Option B: 6 tuần, cover 99% use cases
    Recommend A trước, nếu 10% bị miss → iterate"

4. Flag risks sớm
   "Feature này require schema migration.
    Risk: 2-4 giờ migration window.
    Cần plan trước, không thể đưa vào sprint cuối."

Ceremonies hiệu quả:
  → Weekly sync: PM + Tech Lead (30 phút)
    Agenda: priorities, blockers, technical insights
  
  → Sprint planning: PM define WHAT, Engineers define HOW
    Bạn (Staff): "Approach này có technical debt implications, 
    tôi suggest allocate 20% sprint capacity cho refactor"
  
  → Quarterly planning: Align technical roadmap + product roadmap
    Bạn (Staff): Present technical roadmap, explain 
    WHY technical investments matter cho product goals

Collaborate sớm:

  ❌ Designer design UI → Frontend build → 
     "API không support layout này" → redesign

  ✅ Designer + Frontend + Backend cùng discuss sớm:
     "UI cần data gì? API shape nào phù hợp?"
     → Design system alignment
     → Avoid N+1 API calls
     → Agree on loading states, error states

Designer: "Infinite scroll với real-time search"

Bạn: "Infinite scroll: ✅ cursor-based pagination, straightforward.
      Real-time search: phụ thuộc latency target.
      
      < 100ms: Elasticsearch, cần setup cluster ($)
      < 500ms: PostgreSQL full-text search, free
      
      Recommend: Start với PostgreSQL. 
      Nếu search traffic > 1000 QPS → migrate Elasticsearch."

→ Đừng nói "không được". Nói "được, với conditions/trade-offs..."

Data Scientist build model. Bạn (Staff Engineer) cần:

  1. Serve model trong production
     → Real-time inference: gRPC/REST endpoint
     → Batch inference: scheduled job
     → Streaming inference: Kafka consumer

  2. Feature pipeline
     → DS cần features từ production data
     → Feature store: Feast, Tecton
     → Online (real-time) + Offline (batch) feature serving

  3. A/B testing infrastructure
     → Canary deployment cho models
     → Metrics collection cho model performance
     → Rollback nếu model performs worse

  4. Monitoring
     → Model drift detection
     → Prediction latency
     → Feature data quality

DS cần từ Engineering:
  → Clean, well-documented data (events, schema)
  → Feature pipeline infrastructure
  → Model serving infrastructure
  → A/B testing framework

Engineering cần từ DS:
  → Model requirements: latency, throughput, size
  → Input/output schema
  → Offline evaluation results
  → Monitoring criteria (when model is "broken")

Common friction:
  DS: "Notebook works fine, why can't you just deploy it?"
  Eng: "Notebook ≠ production. Cần error handling, scaling,
        monitoring, rollback, versioning..."
  
  → Solution: ML Platform (MLflow, Vertex AI, SageMaker)
     Bridge gap giữa notebook và production

GDPR technical requirements:

  1. Right to Access (data export):
     → API: GET /users/{id}/data-export
     → Return: tất cả PII user, format machine-readable (JSON)
     → Timeline: 30 ngày

  2. Right to Deletion ("forget me"):
     → Soft delete → hard delete after grace period
     → Cascade: tìm TẤT CẢ systems chứa user data
       ├── Primary DB
       ├── Analytics warehouse
       ├── Log files
       ├── Backups (tricky!)
       ├── Cache (Redis)
       └── Third-party integrations (Stripe, SendGrid)
     → Audit trail: log deletion request + completion

  3. Consent Management:
     → Track what user consented to
     → Version consent (terms change)
     → Withdraw consent = stop processing

  4. Data Minimization:
     → Collect CHỈ data cần thiết
     → Retention policy: tự động delete sau X tháng
     → Anonymize khi có thể (analytics doesn't need PII)

Legal nói: "We need to be GDPR compliant"
Bạn cần hỏi:
  → "Data nào classified là PII trong hệ thống?"
  → "Retention requirements cho từng loại data?"
  → "Cross-border data transfer rules?"
  → "Timeline cho compliance?"

Bạn present:
  → Data flow diagram: PII đi đâu trong system
  → Gap analysis: đang compliant chỗ nào, chưa chỗ nào
  → Implementation plan: milestones, effort estimate
  → Ongoing: automated compliance checks trong CI/CD

Sales: "Customer ABC says API is slow, they might churn"

Bạn (Staff):
  1. Get specifics: Which endpoints? How slow? Since when?
  2. Check data: Customer-specific metrics, logs
  3. Identify root cause: Tenant isolation? Query pattern?
  4. Communicate fix + timeline
  5. Follow up with customer (CC Sales)

Template response to Sales:
  "Đã investigate issue. Root cause: customer có 500K products,
   search query không optimized cho catalog size này.
   
   Short-term fix (2 days): Add index, expect 5x improvement
   Long-term fix (2 weeks): Pagination + caching layer
   
   Tôi sẽ update khi fix deploy. CC customer success."

Khi Sales cần technical input:

  → RFP/RFI responses: "Hệ thống support SSO không?"
  → Security questionnaires: SOC2, encryption specs
  → Architecture diagrams cho enterprise customers
  → Capacity/scalability documentation

  Staff Engineer role:
  → KHÔNG nói "yes" cho everything (đó là Sales' job 😅)
  → Honest assessment: "Chưa support, nhưng có thể build 
     trong Q2 nếu customer commit"
  → Flag: "Custom feature cho 1 customer = dangerous. 
     Nên generalize solution."

Cho Engineering team (Slack/standup):
  "PR #234 merged, auth refactor 60% done.
   Blocked on: need DB migration approval."

Cho Engineering Manager (weekly 1:1):
  "Auth refactor: on track, ETA next Thursday.
   Risk: migration window cần coordinate với DevOps.
   Need: 30-min sync với DevOps team trước Thursday."

Cho Director/VP (monthly update):
  "Auth modernization: 60% complete.
   Impact: Will unblock SSO feature for enterprise launch.
   Timeline: Complete by April 30.
   Risk: Low. Mitigation: rollback plan ready."

Cho CTO/CEO (quarterly):
  "Infrastructure investments reduced auth latency 80%,
   enabling enterprise features. $200K ARR impact projected."

Rule: Audience càng senior → message càng ngắn + business-oriented.

Cross-functional Collaboration:

  PM:     Ask "why" before "how", quantify trade-offs
  Design: Collaborate on API shape sớm, feasibility check
  Data:   Feature pipelines, model serving, A/B testing
  Legal:  Data flow diagrams, gap analysis, privacy by design
  Sales:  Technical escalations, honest assessments
  All:    Tailor communication to audience level