Data to Decision Field Guide
Operating modelControl towerField guide

Data to Decision Field Guide

Trusted data.Better decisions.

One secure foundation for every function, from the first record to the final call.

Or scroll to see how

The shift

Every leap needed foundations. AI is the next one.

Computers: we learned to type, save and share, together.

Databases: one official record replaced piles of paper copies.

Folders and naming rules: everyone could find the same file.

AI and agents: now we need clean data, clear owners and guardrails.

The teams that set up the basics first moved the needle. This is that moment again.

  1. ComputersShared tools
  2. DatabasesOne record
  3. Folders + namingFind anything
  4. AI + agentsTrusted data, people in charge
Next: the foundation that makes it work

Executive summary · Operating model

One foundation. Five layers.

What it is: an operating system for the whole organization. Your data, AI, security and tools are packaged once, then reused by every function. It scales as you grow and stays resilient when something breaks.

Functional control towers. HR, Sales, Finance, IT, Compliance and more each run their own tower.

People + AI agents. People decide. Agents find, draft and automate for them.

Security, privacy + governance. One set of rules protects every person, agent and record.

Tools, systems + data. A governed data fabric connects the systems you already have.

Infrastructure + models. Shared cloud and approved models keep it all running.

See the full operating model

Executive summary · Control towers

See it, check it, act on it.

What it is: think of an airport control tower. It watches every plane, keeps them safe and tells each one when to go. A business control tower does the same for requests and data.

Each function gets its own control tower. Here is the HR one in action.

A manager asks: "Who on my team is out next week?"

The tower finds it. Leave dates come from Core HR. Medical reasons stay behind.

Security checks. Is this their team? Is the purpose allowed? Every access is logged.

The answer goes out. Dates only, with the source, in seconds. The manager decides.

See the HR control tower example
Security check Datafabric Payroll Core HR Cases Medical reasonheld back Manager Dates only+ source

In practice · one record

Follow Iván from entry to decision.

You enter his new job in the official HR system.Why it matters: Everything downstream copies this entry. A typo here travels everywhere.

His record is matched, checked and cleaned.Why it matters: Duplicates are merged and job codes checked, so Iván counts once.

It becomes part of a headcount data product.Why it matters: One owner, one definition. Finance and HR finally see the same number.

An AI assistant can use it, inside a fence.Why it matters: It sees his job and team. Never his health, pay history or cases.

A named manager makes the call.Why it matters: Anything that changes his job, pay or rights needs a person.

Record
Clean data
Product
Bounded AI
Decision

AI can sort, draft, answer and flag. When a decision changes someone's job, pay or rights, a person decides.

Take it with you

Two pages. The whole picture.

Why it matters

Clouds change. The framework holds.

Most organizations now run on several clouds and dozens of SaaS tools, and the mix keeps shifting through migrations, new vendors and new AI. This model keeps people, data and decisions lined up no matter where the systems live.

Open the field guide

Eleven short topics online. The PDF is the full 53-page leadership packet on data, AI and cloud migration.

The operating model

How data becomes a trusted business outcome.

Each control tower uses the same secure foundation to understand requests, find approved data, coordinate work, check results and keep people in control.

Faster, informed decisionsLower riskBetter employee experience

Infrastructure + models: cloud, compute, networking, storage, LLMs, model routing, orchestration, monitoring, logging, APIs, containers and serverless.Tools, systems + data: HRIS, payroll, ATS, LMS, ERP, CRM, ITSM, case management, policies, databases, APIs and external data. A unified enterprise data fabric connects distributed data while preserving authoritative systems.Security, privacy + governance: identity and access, role-based permissions, least privilege, data classification, PII and PHI containment, encryption, policy, human oversight, monitoring and audit, fail-safe containment.Fail-safe containment: detect, contain, pause or stop, human review, restore. Works on an agent, workflow, tool, data connection, control tower or the whole environment.People + AI agents: managers and supervisors monitor, decide, approve and coordinate. Employees access services and complete work. AI agents support both with retrieval, analysis, recommendations and automation.Functional control towers: HR, public sector sales, finance, IT, compliance, acquisition, operations and other functions. Each keeps its own boundary on one shared foundation.HRFinanceITSalesComplianceMoreoperate insideprotectssupportssupportsFunctional control towersDomain control on one shared foundationPeople + AI agentsPeople stay in control; agents supportSecurity, privacy + governanceIncluding fail-safe containmentTools, systems + dataUnified enterprise data fabricInfrastructure + modelsCloud, compute and approved models12345
1Functional control towers
HRPublic sector salesFinanceITComplianceAcquisitionOperationsOther functions

One enterprise foundation supports every control tower. Each keeps its own boundary.

2People + AI agents
  • Managers and supervisorsMonitor, decide, approve and coordinate.
  • EmployeesAccess services, complete work and get guidance.
  • AI agentsSupport both with retrieval, analysis, recommendations and automation.

People remain in control while agents support decisions and work.

3Security, privacy + governance
  • Access controlsIdentity, roles and least privilege.
  • PII / PHI protectionsClassification, encryption and masking.
  • MonitoringEvery access and action is audited.
  • Human oversightPeople review sensitive actions.
  • Fail-safe containmentDetect, contain, pause or stop, human review, restore. Works on an agent, workflow, tool, data connection, tower or the whole environment.
4Tools, systems + data
HRIS / HCMPayrollATSLMSERPCRMITSMCase managementPolicies and documentsDatabasesAPIsExternal dataCollaboration tools

Unified enterprise data fabric. Connects distributed data while preserving authoritative systems.

5Infrastructure + models
CloudComputeNetworkingStorageLLMsModel routingOrchestrationMonitoring and loggingAPIsContainers / serverless

Follow one request end to end

Choose a request and watch each step.

Each stage is handled by a specialist agent. Security and human oversight wrap every step.

One foundation. Many control towers.

Each business function gets its own tower. They all stand on the same data, security and AI.

Explore what supports the flow

Open a layer to see what it does and when the request uses it.

What the system can produce

What the organization gains

Infographic: The Enterprise Operating Model. One operating system, multiple control towers

See the full operating model

One operating system, multiple control towers: towers, apps, data and infrastructure on one page.

Download PDF PNG View full size
NextSee a control tower in actionFollow real people through the HR tower.

Operating modelHR control tower

The HR control tower

How people, data, AI and decisions connect, today versus the future.

Live view

Sample activity

Coming inGoing outStopped by a rule

The journey, six steps

    TodayFuture

    What's broken today

    What fixes it

    How it is built

    Press play to follow Iván, a new payroll analyst, from offer to day one.

    Infographic: Human Resources Control Tower. Challenges, outcomes and an implementation plan

    The HR control tower

    Challenges, outcomes and a tomorrow-morning plan on one page, for printing or sharing with leaders.

    Download PDF PNG View full size
    NextLearn the basics in the field guideEleven short topics, from first entry to trusted decision.

    Your path

    Eleven short modules in four chapters. Each one takes a few minutes.

    Field guide

    People data, start to decision.

    Eleven short topics in four chapters. Read in order or jump to what you need.

    01Basics · 2 min

    How people data becomes a good decision.

    You touch HR records every day. Here is where they go next.

    Every decision starts as a record.

    What you type in the official HR system is what every report, dashboard and AI answer repeats.

    1. RecordYou enter it
    2. DataIt gets cleaned
    3. AI helpInside a fence
    4. DecisionA person owns it
    Go deeper

    Scroll to read. Each topic opens with a picture, then a few short ideas. Tap the pictures; most of them move. Use Contents to jump anywhere, or download the PDF to keep a copy.

    Four words you will hear.

    Tap a word to see what it means.

    What you do

    1. Enter it right, the first time, in the official system.
    2. Fix it at the source, never in a copy.
    3. Ask who owns a field when you are not sure.

    Why it matters · 3 min

    AI is not the starting point. Trust is.

    If the record is wrong, every report and every AI answer built on it is wrong too.

    Real example

    A manager says the headcount dashboard is off by 3 people. What is the most likely cause?

    Records were entered wrong at the source. Most report errors start as entry errors in the HR record. Dashboards only show what the record says.

    Example

    A job code is typed wrong in the HR system. The headcount dashboard is now wrong. The AI assistant now tells a manager the wrong pay band. One small error travels all the way down.

    Your job
    • Enter it right the first time, in the official system.
    • Fix errors at the source, never in a copy.
    • Flag anything that looks off.

    Watch records flow through the checks

    Each dot is a record. Most pass. Some are held back for fixing.

    Illustration · rates exaggerated

    Where records drop out

    12,480 new and changed records in one month. Each check removes some before AI can use them.

    Sample data
    Show as table

    The model · 5 min

    Seven layers, one flow.

    Data moves from the front door to a human decision.

    Real example

    A manager wants a report on team skills. Which layer builds it?

    Reports and AI. Reports are built from clean data in layer 6. They read the record. They never change it.

    Layer 1

    Front door

    Ask yourself

    The experience · 4 min

    Same request. Two very different days.

    Pick a person.

    Real example

    An employee needs leave and has no idea where to start. What helps most?

    One front door that routes the request. One entry point that knows the process saves the employee from guessing between systems.

    The effort, side by side

    Lower is better. Switch the person above to compare.

    Sample data
    StepTodayConnected

    05Protecting people · 4 min

    People data needs a fence.

    AI never browses raw HR tables. It only sees what passes the filter.

    The filter decides what AI can see.

    Names, IDs and health details are removed or masked before any assistant gets the data. Pick a type to see the rule.

    What the AI is allowed to see

    Pick a data type above. The matching rule lights up.

    Example record

    Every step is logged.

    From source to action, each hand-off is recorded, so anyone can check who saw what and why.

    SourceApproved fields only
    FilterHide names and IDs
    AI assistantSees only what passed
    ActionLogged and reviewed

    Every step is logged

    When in doubt, it goes to a person.

    Real example

    A manager asks the AI assistant who on the team is on medical leave. What should happen?

    It refuses and points to the HR partner. Health data is fenced off. Initials still identify people. The assistant routes to a human.

    Go deeper

    Even initials or a team of three can identify someone. That is why health and case data stay fenced, and the assistant routes the question to the HR partner instead of guessing.

    What you do

    1. Never paste employee data into a public AI tool.
    2. Use test data, not real people, when trying things out.
    3. Ask for access only for the task in front of you.

    Using AI · 4 min

    Start with the work. Earn the autonomy.

    The bigger the impact on a person's job, pay or rights, the less AI should do on its own.

    Real example

    Recruiting has 400 applicants this week. Where should AI help?

    Schedule interviews and draft emails. AI takes the busywork. Choosing or rejecting people stays with a recruiter and hiring manager.

    Impact vs AI autonomy

    Each dot is an HR task. The further right, the more it affects a person. The higher, the more AI may do alone.

    Sample data
    Show as table
    Low risk

    Policy and benefits questions

    Answers from approved policy. Hands off the hard cases.

    Human gate: disputes and exceptions

    Moderate

    Skills and staffing

    Spots skill gaps from clean data.

    Human gate: planner checks and decides

    Moderate

    Recruiting tasks

    Drafts outreach and flags slow steps.

    Human gate: who gets an offer

    High impact

    Hiring, pay, relations

    AI may sort evidence. It never picks.

    A person decides. Always.

    Bars show how much the AI may do on its own, from 0 to 5.

    Who does it?

    Drag each task to where it belongs. On a phone, tap a task, then tap a box.

    AI can help

    A person decides

    The infrastructure · 6 min

    Many clouds. One record.

    HR data lives across several vendors' clouds.

    Real example

    Iván's Teams account was not created on day one. Which system do you check first?

    Entra ID, which gets his record from HCM. Accounts come from the identity system, which is fed by the official HR record. Check that link first.

    Walk Iván to day one

    Eight stops across five vendors. Tap a stop or press Next.

    0 / 8

    The full system map. Stops light up here too.

    Step 0 of 8

    Follow a new hire

    Iván accepts a job offer. Watch his data move, one system at a time.

    SaaS partners

    Applicant trackingCandidates and offers
    Payroll providerPays people
    Benefits providerPlans and enrollment

    Oracle Cloud

    Oracle Fusion HCMSystem of record
    Oracle IntegrationMoves data between systems
    APEX + ORDSForms and APIs

    Microsoft Azure

    Entra IDWho you are, what you can open
    Microsoft 365 / TeamsWhere people work

    Google Cloud

    Pub/Sub + DataflowCarry changes in
    BigQueryClean data products
    DataplexCatalog, quality, tags
    LookerDashboards
    Vertex AI / GeminiBounded AI assistant

    People

    HR coordinatorChecks and approves
    Hiring managerOwns the team decision
    Private links between clouds Identity sync (SAML / SCIM) APIs, events and secure files

    Illustrative pattern. Your organization's systems may differ. Hosting a copy in another cloud never moves authority away from the system of record.

    Match the system to its job

    Drag each system onto the job it does.

    Holds the official record

    Decides who can log in

    Stores clean data for reports

    Shows dashboards

    Vendor collaboration · 4 min

    Vendors run the clouds. We own the data.

    Each vendor is responsible for its own service.

    Real example

    Oracle HCM is down for a planned patch. Whose job is it to bring it back?

    The vendor. Vendors own uptime and patching. Your job is to know who to call and to log the impact.

    When something breaks
    1. Find the source. Which system holds the official value?
    2. Check the move. Did the integration run? Look at its status.
    3. Open a ticket with the team that owns that system.
    4. Never patch the copy. It gets overwritten on the next run.

    Who is responsible

    Filled dots own the task. Rings support it.

    Show as table
    TaskVendorOur ITHR
    Keep the service running and patchedOwnsMonitors
    Network links between cloudsProvidesOwns
    Who gets accessSets upApproves
    What a field meansSupportsOwns
    Fixing bad dataSupportsFixes at source
    Decisions about peopleNamed leader

    Governance · 3 min

    The center sets the rules. Teams build inside them.

    Five controls are shared by everyone.

    Real example

    A team wants to send a salary report to every manager. Is that allowed?

    Only if access rules allow it by role and purpose. Shared rules travel with the data. Internal does not mean open to everyone.

    Shared
    guardrails

    Pick a control to see what it means.

    Scorecard · 3 min

    Measure what people feel.

    Every good number has a guardrail next to it.

    Real example

    AI now finishes 74% of tasks without rework. Is that good news?

    Check how often people override it first. A good number needs its guardrail. High overrides would mean people do not trust the output.

    Leadership scorecard

    Each measure trends over eight months. The guardrail beside it must hold steady.

    Sample data
    Show as table

    Wrap up · 2 min

    Five rules to keep.

    1. One record. The official system always wins.
    2. Fix at the source. Never patch a copy.
    3. Fence the data. AI sees only what is approved.
    4. Bigger impact, less AI. People decide on jobs and pay.
    5. Know the owner. Every system and field has one.

    That is the whole guide. Keep a copy to hand.