Claude Implementation Timeline: What to Expect in Weeks 1–12 of Deployment

Week 1 is the foundation. A two-day discovery workshop with the project owner, IT lead, a representative from Legal/Compliance, and two or three subject matter experts from the primary use case area. The workshop outputs are: a confirmed use case scope and success criteria, a data classification inventory (what data will Claude have access to, how is it classified, what are the handling requirements), an inventory of existing tools and systems that Claude may need to connect with, and a preliminary governance framework.

Week 2 is typically the most technically intensive week in Phase 1. The governance framework is finalised — data handling rules, acceptable use policy, escalation procedures, and (for regulated industries) regulatory compliance mapping. In parallel, the system prompt architecture is designed: how many user roles or personas need separate system prompts, what firm-specific context needs to be embedded, what boundaries need to be enforced. A system prompt design session with subject matter experts is usually three to four hours of structured work. The output is a specification, not the final prompt — that comes in Week 4.

Week 3 finalises the technical architecture. For a Claude Enterprise deployment: SSO configuration spec, admin console setup plan, licence provisioning schedule. For API or agentic deployments: the full technical architecture document — API integration patterns, MCP server design for each required data connection, authentication and authorisation design, and the testing plan. Security review by the client's InfoSec team typically happens in Week 3 — schedule this proactively, as InfoSec calendars can add weeks of delay if not planned.

Week 4 is system prompt engineering week. The specification from Week 2 is turned into tested, validated prompts through an iterative process: draft → test with representative inputs → evaluate output quality → refine. Each role profile typically requires 3–5 iterations to reach production quality. Platform configuration runs in parallel: SSO setup, admin console configuration, usage policy upload, and test account provisioning. By end of Week 4, the platform is configured and the system prompts have passed internal quality review.

Weeks 5 and 6 are integration build for deployments with MCP connections or custom API work. MCP servers are developed, tested, and documented. Integration testing validates that Claude can access the required data sources correctly, that data handling rules are enforced, and that the system behaves correctly at edge cases. For simpler Claude Enterprise deployments without custom integration, these weeks focus on advanced prompt testing against real use case scenarios and the development of training materials.

Week 7 is UAT with a small group of five to ten end users from the target population. UAT tests the complete user experience: platform access, system prompt quality, integration functionality (if applicable), and the training materials. UAT outputs: a structured feedback form covering output quality, usability, and training adequacy; a list of required changes before pilot go-live; and a pilot readiness sign-off from the project owner. The most common UAT finding is system prompt gaps for edge case scenarios — build in buffer to iterate before the pilot.

Pilot go-live with 10–20% of the target population, chosen to represent the range of roles and use cases. All pilot users complete the training programme before access is granted. A Slack or Teams channel is set up for real-time pilot support (we typically staff this with one implementation team member for the pilot duration). Usage dashboards go live; the first daily usage report is reviewed by the implementation team and project owner at end of Day 1.

Weeks 9 and 10 are structured data collection. Daily usage review, twice-weekly structured feedback surveys from pilot users, and weekly review calls with the project owner. System prompt iterations based on real usage patterns are typically made 3–5 times during the pilot phase. Usage data at Day 14 of the pilot should show daily active user rate above 50%; if below 40%, there is a change management issue that needs addressing before full go-live. The Claude change management guide covers the interventions that work.

Full rollout training for the remaining population is delivered in the first half of Week 11 (live sessions, segmented by role). Accounts are activated in batches as users complete training. A hypercare period of 72 hours follows full go-live: the support channel is staffed, usage is monitored every four hours, and minor system prompt adjustments are made within 24 hours if quality issues emerge at scale. Day 3 after full go-live is when the daily active user rate stabilises — this is the first meaningful adoption metric.

Week 12 closes the initial engagement with three deliverables. First, a 30-day performance report: usage analytics (DAU rate, sessions per user, query patterns), adoption by department, and qualitative feedback themes. Second, an ROI measurement: actual time saved versus baseline, extrapolated to annual value, compared against the pre-deployment business case. Third, a 90-day roadmap: the next phase of capability expansion — new use cases, deeper integrations, agentic workflows, Claude Cowork expansion — with a prioritised sequence and rough effort estimates.