90-day rollout roadmap

Four stages—pass each gate before moving to the next

Quantified data from stage two is the basis for requesting a full rollout.

Weeks 1–2

01Pilot preparation

  • Edition selection and procurement
  • SSO + org sync
  • Security and model policies
  • Pick 2 pilot departments
Gate: Seed users in place

Weeks 3–6

02Use-case validation

  • 3–5 high-frequency use cases
  • Collaboration platform + scheduled tasks
  • ERP read-only access
  • Record time spent before and after
Gate: Time savings are quantifiable

Weeks 7–10

03Codify and scale

  • Turn SOPs into Skills
  • Package Expert Suites
  • Publish in the enterprise console
  • Roll out to all departments
Gate: ≥ 3 Skills per department

Weeks 11–13

04Deep integration

  • ERP writes + intercept-and-confirm
  • In-house agents connected via MCP
  • Data quality inspections
  • Management review
Gate: Management sign-off

Quantified data from stage two is the basis for requesting a full rollout.

Five maturity levels

From "asking the odd question" to "running unattended"

Find your level

L1

Chat

Ask questions, edit text

Criteria: Used at least once a day

L2

Deliver

Ask directly for finished files

Criteria: No more handwritten weekly reports or PPTs

L3

Connect

Read and write the collaboration platform, ERP and browser

Criteria: No more exporting and pasting data

L4

Codify

Skills, Expert Suites, playbooks

Criteria: ≥ 3 shared Skills per department

L5

Automate

Scheduled tasks + IM push

Criteria: Daily reports and alerts arrive automatically every day

Four elements of a good instruction: what to do · what data to use · what output format · who receives the result

SOP & Skills

SOPs are written for people; Skills are written for AI to execute

  1. STEP 1

    Pick an SOP

    High-frequency, clear rules, fixed output

  2. STEP 2

    Co-create in conversation

    Have AI follow the SOP once and correct it along the way

  3. STEP 3

    Codify

    Write it up as SKILL.md

  4. STEP 4

    Validate

    3 real historical datasets, compared against manual results

  5. STEP 5

    Publish

    Approve and publish in the enterprise console, packaged as an Expert Suite

Three-layer knowledge structure

Personal conventions

Playbook (AGENTS.md)

Task SOPs

Skills (SKILL.md), shared by role / team

Department capability packs

Expert Suite: multiple Skills + connectors + / commands

Data governance

Govern the data each use case needs first—don't "build a big data platform first"

Source-system permissions are the first gate: whatever account a connector uses, AI can only see that account's data.

LevelExampleAccess policy
L1 PublicProduct materials, public policiesAvailable to everyone; web search allowed
L2 InternalPolicies, processes, training materialsTurned into internal knowledge Skills
L3 SensitiveCustomers, pricing, contractsOpened by department; masked via Hooks
L4 ConfidentialCompensation, undisclosed financials, core processesNot connected by default; approved by a designated person

Control levers

  • Who can use it: SSO, org sync, user groups
  • Where it's used: trusted networks, trusted devices
  • What it can see: connector credentials and access rules
  • Sensitive information: sensitive-word filtering, masking via Hooks rewriting
  • Which model: model policies, BYOK
  • Traceability: log auditing, usage statistics
  • Consistent definitions: metric definitions written into shared Skills

Get your company's 90-day roadmap

The AI adoption assessment delivers an assessment report, a prioritized use-case list and a 90-day roadmap in 1–2 weeks.

Book an FDE consultationFree maturity self-assessment