Dr. Irina Raicu.

ENTERPRISE AI TRANSFORMATION

From AI ambition to a system the organization can run.

Most large organizations have models, licenses, and a folder of pilots. What they lack is a system: where AI belongs, who owns the work, and how value is measured.

I design that system with the leadership team that runs it: operating model, use-case portfolio, adoption architecture, measurement cadence. I built it at global scale inside Microsoft.

  • 15+ years across consulting, AI product delivery, digital transformation, and organizational adoption
  • Former Microsoft Global AI Director, Enterprise AI Adoption & Support Transformation
  • PhD in AI
Start a conversation See the enterprise impact
Dr. Irina Raicu in a navy double-breasted blazer with gold buttons, photographed against a plain white background.

THE WORK / 01

AI operating models that survive contact with the organization.

An AI operating model answers the questions that stall programs at month four.

I fit it to your reporting lines, funding cycles, and appetite for central control. At Microsoft that meant a hybrid hub-and-spoke model: standards centrally, the last mile regionally.

What gets decided

  • 01

    Ownership. Which decisions sit centrally, which with the business units, and who breaks a tie.

  • 02

    Intake and gates. How an idea enters, what it must prove, and who can stop it.

  • 03

    Funding. Whether AI work is funded centrally, by business case, or by the team that benefits.

  • 04

    Review and escalation. Which forums meet, on what cadence, and with what authority.

  • 05

    Responsible AI checkpoints. Bias assessment, data-privacy gates, and production-readiness review sit inside the lifecycle.

You leave with a documented operating model, a responsibility map naming real people, and a cadence to run it.

THE WORK / 02

A portfolio you can rank, sequence, and stop.

Most enterprises are not short of AI ideas. They are short of a defensible way to rank, sequence, and stop them.

So every candidate is scored, then sequenced against the teams and budget you have. At Microsoft that process assessed 10+ use cases; four advanced to production.

How candidates are scored

  • 01

    Value. The size of the outcome, who books it, and the baseline behind it.

  • 02

    Feasibility. Technical difficulty, integration surface, and whether the receiving team can absorb it.

  • 03

    Data readiness. Whether the data exists, is clean enough, and is permitted for this use.

  • 04

    Ownership. The named business owner who runs the workflow after launch. No owner, no funding.

  • 05

    Kill criteria. The specific result, by a specific date, that ends the work.

You hold a ranked portfolio, a quarter-by-quarter roadmap, and a written basis for saying no.

THE WORK / 03

Adoption starts when the job description changes.

Adoption fails between the license and the workflow: people are trained, the tool is live, and the work continues as before.

So the work is role-level. We decide which steps an agent takes and which a person keeps, then rewrite the artifacts that govern behavior.

Champions, regional leads, and managers carry it, held to adoption in their own reviews. At Microsoft this ran across a 2,000+ person global support organization spanning the Americas, EMEA, and APAC.

What changes

  • 01

    The workflow. Step by step, with the human-agent handoff and the checking step named.

  • 02

    The role. What the job is now, what good output looks like, and what to review.

  • 03

    The manager. Adoption becomes a line-management responsibility with visible numbers attached.

  • 04

    The champion network. Practitioners with defined responsibilities and a route to escalate what is failing.

  • 05

    The feedback loop. Usage data and champion signal, used to diagnose barriers and change the rollout.

You finish with redesigned roles, a functioning champion network, and adoption owned by the line.

THE WORK / 04

A roadmap the executive team can defend.

An AI roadmap is a funding argument before it is a delivery plan. It has to survive a CFO asking what the second quarter buys.

So I sequence work across quarters with explicit dependencies, including the platform, data, and security work that lands first, then reforecast on a fixed cadence.

What the roadmap contains

  • 01

    Sequenced phases with the outcome each phase is accountable for.

  • 02

    Dependencies across data, platform, security, and the receiving business teams.

  • 03

    Build, buy, and partner decisions attached to the phase where they become urgent.

  • 04

    Gate conditions written as pass criteria, so a review ends in a decision.

  • 05

    Capability and hiring implications, including what the organization must be able to do itself.

The output reads to the board as an investment case and to delivery teams as a plan.

THE WORK / 05

Value that survives the finance review.

Most AI value claims collapse under one question: compared to what. Without a baseline captured before launch, the number is an assertion.

So measurement is designed in before the use case ships, with the baseline captured while the old workflow still runs.

Reporting then runs on a cadence with a named owner.

The three questions, kept separate

  • 01

    Is it being used? Daily active use, depth of use within the workflow, and usage by team.

  • 02

    Did the work get better? Cycle time, quality, rework, and customer-facing outcomes against a pre-launch baseline.

  • 03

    Did the money move? Capacity released, cost avoided, or revenue enabled, agreed with finance in advance.

You leave with baselines, instrumentation, a reporting cadence, and a value case finance has agreed to recognize.

SELECTED ENTERPRISE IMPACT

Three programs, and what changed because of them.

Each is described by the decisions it required and the results it produced.

LEAD CASE · MICROSOFT

Microsoft

Copilot and agentic AI adoption across a 2,000+ person global support organization.

At that spread, adoption has to be built as a structure, not issued as an instruction.

I designed and ran a hybrid hub-and-spoke operating model connecting a central AI Innovation Hub with regional enablement leads, managers, practitioners, and a 65-person champion network.

I also built the roadmap and portfolio process: 10+ use cases assessed; four advanced to production.

  • 90%+Daily active use within nine months.
  • Contributed to $3M+In documented savings, 2024.
  • 4.1 to 4.6CSAT improved from 4.1 to 4.6.
  • 7+ days to approximately three daysComplex case resolution time reduced.

CLOUD AND SECURITY

Making production the default state for a trained model.

Model deployment was the bottleneck, not model development. I redesigned the path from a trained model to a compliant production system.

  • 3-6 months to under six weeks, deployment cycle across 10+ production models
  • 90% improvement in threat-detection response
  • Contributed to an estimated 80% reduction in incident-response time

FINANCIAL SERVICES AND ESG

AI operating models inside regulated institutions.

At Sustainalytics and Morningstar, the constraint was analyst trust, not model accuracy. At Société Générale, I co-founded an AI Center of Excellence and led a cloud financial platform transformation.

  • 70% reduction in manual document processing
  • 40% faster report turnaround, with 90% analyst adoption
  • 2M+ daily users and 500M+ euros processed daily on the platform

WAYS TO WORK

Six ways engagements start.

Most engagements begin with one of these and grow into the next.

  • 01

    Executive advisory

    A standing relationship with the leadership team, working the live decisions: what to fund next quarter, how to restart a stalled rollout.

  • 02

    AI operating-model design

    I map how AI work moves through your organization today, then redesign it: intake, ownership, funding, review gates, and escalation.

  • 03

    Use-case portfolio and roadmap

    We inventory every AI initiative in flight, score candidates on value, feasibility, and data readiness, then sequence them into a funded roadmap.

  • 04

    Leadership workshop

    A working session for an executive team, run on your own use cases and numbers. It ends with prioritized bets and a named owner for each.

  • 05

    Applied workflow lab

    A working team rebuilds one real workflow around human-agent collaboration, using your data and tools. You finish with a prototype and a measurement baseline.

  • 06

    Management enablement program

    A structured program for the managers who own adoption inside their teams: setting expectations, redesigning roles, coaching behavior change, reading adoption data. They finish able to run the rollout themselves.

Keynotes and executive education are covered on the speaking page.

Discuss an engagement

NEXT STEP

If the pilots have stopped compounding, that is a system problem.

Tell me where the program is stuck: the operating model, the portfolio, adoption, or the measurement.

Work with Irina Speaking and executive education