Rolling Out AI to a 500-Person Company: A 90-Day Plan

500-person companies have a specific AI rollout problem. They are too big to operate informally — five business units, three geographies, mixed regulatory exposure, a real compliance function. They are too small for an enterprise-grade transformation program — no AI department, no Center of Excellence, no executive sponsorship full-time on AI.

The result: most 500-person companies have either no AI strategy or a fragmented one where individual teams have signed up for ChatGPT Teams and built shadow workflows. The 90-day plan below gets the foundation right, ships the first agents, and creates the governance posture that scales.

Days 1-15: leadership alignment and scope

Decision needed by day 7: executive sponsor (typically COO, CTO, or CEO). Without an executive sponsor with weekly time committed, the rollout will stall by month two when difficult cross-functional decisions are needed.

Decision needed by day 15: initial scope. Pick 2-3 functional domains for the first wave, not 10. The domains where AI delivers obvious value at this scale are usually:

• Customer support (highest volume, clearest deflection metric)
• Marketing operations (creative-leveraged work, structured outputs)
• Finance close acceleration (high-value, well-defined work)
• Sales research and outreach (revenue-leveraged)
• HR recruitment or knowledge management

Pick the two where leadership is already convinced and one where you want to learn. The first wave establishes the operational pattern, not the maximum scope.

Activities in this period:

• Executive briefings on EU AI Act and GDPR implications
• Quick inventory of existing shadow AI usage (everyone has more than they admit)
• Vendor selection or build-vs-buy decision (do this once, not per agent)
• Compliance assessment with the DPO and legal
• First-pass org structure for the AI function (often a small 2-3 person team initially)

Days 16-30: platform setup and governance foundation

Platform stand-up. Whether you bought a platform or building, the first 15-30 days are platform configuration:

• Identity and SSO integration
• Data residency configuration
• PII redaction rules
• Audit log retention policies
• Per-team workspace setup
• Budget caps and alerts
• Integration with the first few source systems

For a managed platform like AgentWorks, this is typically 2-4 weeks of work. For an in-house build, this phase is months 1-6, not days 16-30; the timeline assumes a platform approach for 500-person scale.

Governance baseline. The minimum governance that needs to be live before the first agent:

• AI policy document approved by exec leadership (acceptable use, transparency, oversight expectations)
• DPIA template tailored to your context
• Risk classification process per agent
• Incident response runbook
• Vendor sub-processor list approved

These are documents that exist in your governance repository, not slides. They are referenced when agents are created and reviewed.

Activities in this period:

• Platform technical setup
• Governance documents drafted, reviewed, approved
• First-line training for solutions architects (typically 3-5 people)
• Stakeholder communication: announcement, FAQ, channel for questions

Days 31-60: first agents and stakeholder engagement

Agent 1: ship the highest-confidence agent first. Usually customer support intake or a similar high-volume, well-bounded use case. The first agent is not about maximum value; it is about proving the operational pattern.

Specific milestones:

• Day 35-40: agent design complete with the business unit
• Day 40-50: agent built, tested in sandbox
• Day 50-55: agent live with one team in pilot mode
• Day 55-60: pilot review, iterate on feedback

Agent 2 (parallel to 1): starts at day 35-40, ships at day 65-70. Lower stakes if needed, but uses the same operational pattern as agent 1.

Stakeholder engagement parallel track:

• Weekly exec update with metrics from agent 1
• Monthly all-hands inclusion on AI rollout progress
• Skip-level conversations with team leads in scope domains
• Open-door sessions for questions and concerns

The communications matter as much as the agents. At 500-person scale, narrative matters; people need to understand what is happening and how it affects them.

Days 61-90: scale and operational pattern

Agent 3 onward: the operational pattern is now repeatable. Agents 3-5 ship faster than agent 1 because the platform, the governance, and the team have learned.

Operational discipline:

• Weekly portfolio review with executive sponsor
• Monthly retrospective with the platform and solutions architect team
• First quarterly business review with the executive team

Compliance evidence checkpoint: by day 90 you should have the audit evidence for the agents that are live, on a representative pull. If the evidence is thin, fix it now before the estate grows.

Adoption measurement: by day 90 you should have baseline metrics on adoption (who is using the agents, how often), quality (override rates, escalations), and value (deflection, time saved, outcomes). If any of these are weaker than expected, diagnose now.

What the 500-person scale gets wrong most often

Doing too much too fast: 10 agents in 90 days is overreach at this scale. 3-5 agents shipped well teach the team how to do 20 agents in year two. 10 agents shipped sloppily damages credibility and may collapse the program.

Skipping governance to ship faster: shipping agents without DPIA, without risk classification, without audit evidence is the recipe for a year-two compliance crisis. The 5-10% time cost of governance upfront is worth far more than the cost of rebuilding evidence on agents already in production.

Under-investing in change management: 500-person companies have strong informal communication networks. If users do not understand what the agents do and how they affect them, the rumour mill fills the gap badly. Communication is part of rollout, not an afterthought.

Building the platform instead of buying: at 500-person scale the in-house build economics rarely work (see hidden costs of building AI in-house). The build vs buy decision should be settled in the first 15 days, not revisited every six months.

Treating it as an IT project: the AI rollout is a business transformation that uses technology, not a technology deployment that affects business. The executive sponsor should be a business leader, not just a technology leader.

What success looks like at day 90

A 500-person company at day 90 of a well-run AI rollout has:

• 3-5 agents in production, with documented compliance evidence per agent
• A platform team of 1-2 engineers operating the platform
• A governance practice that the DPO and legal can sign off on
• Visible adoption in the pilot domains (with metrics)
• A roadmap for the next 3-5 agents in the next quarter
• An organisation that knows what AI is being used for and where

This is the foundation. From here the program scales with discipline. Without this foundation, year two is mostly remediation.

The next 90 days after this 90 days

Quarters 2 and 3 extend the model:

• Q2: 5-10 additional agents, mostly in the original domains
• Q3: 5-10 agents into new domains (HR, finance, ops)

By the end of year one a 500-person company should have 15-25 agents in production, the operating pattern proven, and the foundation for the next year of organic expansion as business units identify their own opportunities.

What AgentWorks provides

The AI workforce platform is built for the 500-person scale: enough flexibility for the variety of use cases a mid-size enterprise has, enough governance for the compliance posture required, simple enough to operate with a small team. The rollout pattern above is the one we recommend and we have operational support to help customers execute it.

The honest summary: 90 days is enough to get the foundation right and ship the first agents. It is not enough to transform the company. The transformation happens over years; the 90 days set up the conditions under which it can happen.