Skip to content

Copilot AI Foundation

Copilot readiness for companies whose rollout is waiting on governance and legal.

Your company has switched on Microsoft Copilot (formerly Microsoft 365 Copilot), but company-wide use is blocked by data governance, an acceptable use policy and legal sign-off. We do the work that clears it in your own Microsoft 365 tenant, from oversharing fixes, Purview labels and data loss prevention (DLP) to the AI policy, agent rules and training for each team, and hand it to your IT team with a runbook.

Updated November 2026

The problem

Why company-wide Copilot gets held up.

Copilot finds what was shared too widely

Copilot shows each person only what they can already open, and years of sites opened to the whole company, often through broad groups such as "Everyone except external users", make that far more than anyone intended. In a July 2026 survey by CoreView, a software vendor, two-thirds of organizations had delayed or cancelled Copilot, and 73% of those feared it would surface confidential information.

A label alone does not keep a file out of Copilot

A sensitivity label marks a file. Encryption on the label can stop Copilot using the content when it withholds the rights Copilot needs, and a DLP policy written for Copilot can keep labeled files and emails out of answers. That policy needs the rights that come with the Microsoft 365 E5 enterprise plan or a Purview add-on, and both approaches need testing app by app.

With no policy, legal has nothing to sign

Without an AI policy, a register of the AI tools in use and a route for approving new ones, legal and compliance usually hold back company-wide use. Department tools such as CRMs and scientific software now ship their own AI features, and they need to be governed by the same policies.

Agents appear before anyone owns them

Staff can build agents, AI assistants set up for one task, in Copilot Chat, SharePoint and Copilot Studio and share them with colleagues. Microsoft reported more than 100,000 agents in its own tenant in November 2025. It deletes personal agents when their owner leaves and asks the owners of team agents to confirm them on a schedule.

What you get

What the Copilot readiness package includes.

The package covers Purview labels, DLP, oversharing fixes and staff training, plus the AI policy and agent rules that legal and compliance ask for. One team delivers all of it as one engagement.

Discovery

A review of your tenant and how widely its sites are shared, the AI tools already in use (including AI features inside department tools such as CRMs and scientific software), and the AI terms in your client contracts where they apply. Staff get an interim AI rule early, and discovery ends with a confirmed plan.

Data foundation

Purview sensitivity labels, DLP, oversharing fixes in SharePoint, Teams and OneDrive, sensitive sites kept out of company-wide Copilot search, audit of Copilot activity, and a runbook handed to your IT team.

AI foundation

An AI policy with acceptable use rules, an AI tool register with a simple approval route, rules on who may build and share agents, an agent register, and a controls map to the EU AI Act, GDPR, NIST AI RMF and ISO/IEC 42001.

Staff training in four layers

AI basics for every employee, Copilot training per team built on that team's real work, training for the people who build agents, and handover training for IT on running the controls.

Pilot and rollout plan

A pilot group measured against how they work today, then a rollout plan with the Microsoft license and running costs your decision group needs to approve it.

Optional add-ons

A regulated track with a data protection impact assessment (DPIA), a works council pack and a written boundary for GxP, the good-practice rules of regulated life-sciences work. Run support after handover keeps the controls current.

Data foundation

The data foundation, set up in your Microsoft 365 tenant.

We fix oversharing first, because a policy cannot protect a site the whole company can already open. Your IT lead approves each change before it goes live, and DLP rules run in test mode before they enforce. When a rule blocks legitimate work, the runbook names who reviews the exception.

Purview sensitivity labels

A short label scheme, for example Public, Internal, Confidential and Restricted, with encryption on the labels that need it and labeling rules your staff can follow without a manual.

DLP policies

Data loss prevention for email, SharePoint and OneDrive, including files shared in Teams, so confidential content shows a warning or is blocked before it leaves the company.

Oversharing fixes

Anyone links, organization-wide access and broad groups removed from sensitive sites, Teams and OneDrive, with site owners confirming who should have access. Reports and site access reviews in SharePoint Advanced Management, Microsoft's advanced SharePoint admin tools, find the ownerless and inactive sites.

Sensitive sites kept out of Copilot search

Restricted Content Discovery hides named sites from company-wide search and Copilot without changing who can open them. It controls what people find, so we pair it with permission fixes, and with Restricted Access Control where only one group should open a site.

Labeled content kept out of Copilot

A DLP policy on the Microsoft 365 Copilot and Copilot Chat location can stop Copilot using files and emails that carry a chosen label. It covers supported file types and emails sent from January 2025, a matching item can still be cited, and files uploaded straight into a prompt are not scanned, so we test it in your tenant. The policy needs E5-level rights, so we tell you where your licenses allow it.

Audit and a runbook for IT

Copilot activity searchable in the audit log by named roles only, and a runbook your IT team follows for new sites, new joiners and new AI tool requests, written as we make each change.

Licensing

What Business Premium covers, and what needs E5 or a Purview add-on.

Our reading of Microsoft's service descriptions, checked in November 2026. Microsoft changes licensing often, so we confirm each row in your admin center and with your reseller. Purview add-ons must cover the users who benefit from each control, so discovery counts them before you buy. We do not resell licenses.

ControlMicrosoft 365 Business PremiumWhat adds it
Manual sensitivity labels with encryption, plus default and mandatory labeling set in label policiesIncludedNothing more
A default label for a whole SharePoint document libraryNot includedMicrosoft 365 E5, or the Purview Suite for Business Premium
Automatic labelingNot includedMicrosoft 365 E5, or the Purview Suite for Business Premium
DLP for email, SharePoint and OneDrive, including files shared in TeamsIncludedNothing more
DLP for Teams chat messages, and endpoint DLP on devicesNot includedMicrosoft 365 E5, or the Purview Suite for Business Premium
DLP that keeps labeled files and emails out of CopilotNot includedE5-level rights: Microsoft 365 E5, the Purview Suite, or the Purview Suite for Business Premium
Audit of Copilot activityAudit (Standard) with 180 days of retention, which records Copilot interactions once Copilot is licensedAudit (Premium), whose retention policies can keep Copilot audit records longer: Microsoft 365 E5, or the Purview Suite for Business Premium
Restricted Content Discovery, Restricted Access Control, site access reviews and data access governance reportsNot part of Business Premium itselfOne paid Copilot license in the tenant turns them on for a Copilot rollout. For Copilot Business tenants, confirm in your admin center. Restricted site creation by apps still needs SharePoint Advanced Management Plan 1
Agent Registry in the Microsoft 365 admin center: the agent inventory, publish approvals, blocking, deletion and owner changesIncluded, as in every Microsoft 365 planAgent 365, a separate per-user license, adds monitoring of agent activity, Conditional Access for agents and Purview and Defender protection for agents
Copilot Studio environment routing and sharing limits (Managed Environments)Not part of Business PremiumA Power Platform premium governance feature, so the people who build in those environments need qualifying licenses. We confirm which before you buy

AI foundation

Rules your compliance lead can sign, for AI tools and for agents.

Legal and compliance sign off on documents and evidence, so we write them with your compliance lead. Each rule maps to a setting in the tenant that enforces it or, where no setting can, to a procedure with a named owner.

AI policy and acceptable use

Which AI tools are approved, which data may go into which tool, personal accounts, AI features inside department software, disclosure to clients, and how staff report a mistake. Leadership signs it, and it replaces the interim rule.

AI tool register and approval route

Every AI tool and AI feature in use, with an owner, a risk level and the data it may handle, plus a one-page route for requesting a new tool and a checklist for vendors.

Who may build and share agents

Rules for each kind of agent, set in the tenant: who may share agents built in Copilot Chat (Microsoft calls the tool Agent Builder) with the whole company, Copilot Studio builders routed by default to governed workspaces that Microsoft calls managed environments, and connector policies that limit which outside systems an agent may reach.

Open

Agent register

Every agent with an owner, a purpose, the data it reads and a review date, reconciled with the Agent Registry in the Microsoft 365 admin center, which every Microsoft 365 plan includes. Agents nobody owns are blocked or deleted on a set schedule.

Open

Client AI terms turned into controls

Where client contracts restrict AI, discovery lists each term and we map it to a label, a DLP rule, a register entry or a line in the policy, with evidence you can show the client.

Controls map

A record of how each requirement is met and who owns it, linked to the EU AI Act, GDPR, the NIST AI Risk Management Framework (NIST AI RMF) and ISO/IEC 42001, the international standard for AI management systems, so an auditor or a client can follow it.

Open

Staff training

Staff training in four layers.

The EU AI Act requires measures that support AI literacy, sized to the role and the risk (Article 4, which has applied since February 2025 and was amended in July 2026). The European Commission says no certificate is needed and suggests keeping internal records, so every layer leaves one.

AI basics for every employee

What Copilot and other AI tools do with data, what your policy allows, and how to check an answer before using it. Short sessions in English and German where you need both, with a completion record.

Copilot training per team, on its own work

Each team learns Copilot on its own documents and tasks, so finance learns on its month-end files and sales on its proposals. The UK government's cross-government Copilot experiment (2025) concluded that training was essential to get the benefits, and HMRC users asked for more practical, tailored training (2026). No public study yet compares training per team with generic sessions, so we measure the difference in your pilot.

Training for agent builders

The people allowed to build agents learn the guardrails: which environment to use, which connectors are allowed, how to share safely, and how to register an agent with an owner and a review date.

IT handover training

Your IT team learns to run what we set up: the label and DLP policies, Restricted Content Discovery, the audit search, the agent inventory and the runbook.

Pilot and add-ons

A measured pilot, then the add-ons you choose.

Pilot group

Real users from a few teams, measured against how long some of their own tasks take today. Before it starts we agree which tasks to measure and what result would justify a wider rollout, and time spent checking and correcting Copilot's work counts. If the pilot misses the criteria, the report says what to change or recommends holding the rollout.

Costed rollout plan

Rollout waves, go or no-go criteria from the pilot, and the Microsoft license and running costs, so the CFO sees what company-wide use will cost before approving it.

Regulated track

A DPIA with your data protection officer, a works council pack for German sites, and a written GxP boundary that keeps decisions and signatures with people and never lets AI become the source of GxP data. The life-sciences version of this package has the detail.

Open

Run support after handover

A regular control report, register updates, reviews of new AI tools and training for new joiners, for companies that want us to keep the foundation current.

Open

How it runs

Discovery, then three tracks at the same time.

A policy does not protect data while sites are overshared, and staff trained on an overshared tenant learn to find the wrong files, so the policy, the Microsoft 365 controls and the people work move together. Each phase has acceptance criteria you sign before it starts.

  1. Starts first

    Discovery

    The tenant and oversharing review, the AI tools in use, the client AI terms, the interim AI rule for staff, and a confirmed plan with acceptance criteria for each later phase.

  2. Track one

    Policy and governance

    The AI policy, the tool register and approval route, the agent rules and register, and the controls map, written with your compliance lead.

  3. Track two

    Microsoft 365 controls

    Labels, DLP, oversharing fixes, Restricted Content Discovery, Copilot and agent settings, and audit, each change approved by your IT lead and recorded in the runbook.

  4. Track three

    People and training

    The four training layers, with a named contact in each team who collects questions and tells us which tasks to train on.

  5. Closes the work

    Pilot and rollout plan

    The measured pilot, the costed rollout plan, and the handover of every control to your IT team with the acceptance tests they signed.

From your side

What it takes from your side.

We do most of the work, and four things have to come from you. Your IT lead agrees with us which admin roles we need and for how long, and removes our access at handover. The same person approves each change before it goes live. Site owners confirm who should see what, because only they know whether a finance folder open to the whole company is a mistake. A quality or compliance reviewer reads the policy and the controls map as we write them, and a pilot group of real users gives us time to measure their work before and after.

Microsoft licenses are bought through your reseller, and legal sign-off on the policy comes from your counsel. In Germany we involve your data protection officer and, where you have one, the works council from the start.

Proof

Where this work comes from.

Since 2014 we have delivered more than 200 production deployments on AI and the Microsoft platform. We have been a Microsoft partner since 2015, with teams in Texas and Munich.

How it works

What you have at the end.

Controls running in your tenant

Labels, DLP, permission fixes and Copilot settings in use, each with an acceptance test your IT lead signed.

A runbook for IT

Step by step, for new sites, new joiners, new AI tool requests and new agents.

A signed AI policy

With the acceptable use rules, and a German addendum where you need one.

Two registers

AI tools and agents, each entry with an owner and a review date.

Training records

Who completed which layer, and when, kept as the European Commission suggests.

A costed rollout plan

Waves, go or no-go criteria and license costs, ready for the decision group.

Our method

From one scoped workflow to an agent in daily use.

  1. Day one

    Scope and sign

    We define the workflow and the acceptance criteria it has to meet, and agree them in writing before the build starts.

  2. The build

    Build

    We build the agent in your environment, with Claude Code and automated evaluation against your own data, and test it against the criteria you signed.

  3. After launch

    Run

    We run, monitor and improve the agent after launch, measured against those criteria, with a named person on your side approving what it produces.

Compliance & security

  • Built to SOC 2, HIPAA and GDPR standards
  • EU AI Act-aligned delivery
  • Microsoft Solutions Partner

How we engage

Four ways to work with us.

Discovery comes first and ends with a confirmed plan. Each later phase has acceptance criteria agreed in writing before it starts, and Microsoft licenses are bought through your reseller.

Advisory

Strategy, roadmap and readiness for an AI program.

Build

One scoped workflow at a time, with acceptance criteria agreed in writing before the build starts.

Run

Managed operations after launch, measured against those criteria.

Outcome-linked terms

Commercial terms can be tied to the acceptance criteria you sign.

Every engagement starts with one scoped workflow and a written definition of done. Commercial terms are set in the statement of work.

Frequently asked

Questions buyers ask.

Which license do we need for Copilot?

Staff who use Microsoft Copilot need a paid Copilot license: Microsoft 365 Copilot, or Copilot Business for organizations with up to 300 users. Microsoft Copilot Chat is included with Microsoft 365 business and enterprise plans, so many companies start there. For the governance controls, Business Premium covers most of the data foundation, and a few controls need E5 or a Purview add-on, as the licensing table on this page shows.

We have Microsoft 365 Business Premium. Can you do the data foundation, AI policy and training on it?

Yes. Business Premium includes manual sensitivity labels with default and mandatory labeling set in label policies, DLP for email, SharePoint and OneDrive, and 180 days of audit. One paid Copilot license in the tenant turns on Restricted Content Discovery and the other SharePoint Advanced Management tools for oversharing, which we confirm in your admin center for Copilot Business. The policy, the registers and the training need no extra license. Automatic labeling and the DLP rule that keeps labeled content out of Copilot need E5 or the Purview Suite for Business Premium. Label encryption, which Business Premium includes, can also stop Copilot using a file, and we tell you which route fits before you buy anything.

Does Copilot see everything in our tenant?

No. Copilot uses only what the signed-in person can already open, although an agent built in Copilot Studio can connect to other systems with its author's credentials, so we check each agent's connections too. The bigger risk is content shared far wider than intended, such as sites opened to the whole company through broad groups. Before Copilot goes wider, we run the oversharing reports in SharePoint Advanced Management and, where your licenses include it, Purview Data Security Posture Management (DSPM), and fix the worst exposure first.

Can we keep one sensitive site out of Copilot?

Yes. Restricted Content Discovery hides a named site from company-wide search and Copilot without changing who can open it, and people with access can still open the site and its files. It controls what people find, so the lasting fix is still the site's permissions, and Restricted Access Control can limit a site to one group. Restricted SharePoint Search, the older control, is retiring and can no longer be newly turned on.

Some of our clients forbid AI on their data. How is that data protected?

Microsoft states that prompts, responses and the work data Copilot reads are not used to train its foundation models. Client terms usually ask for more, so discovery lists each AI clause and we map it to controls we then test: a client label whose encryption withholds the rights Copilot needs to use the content, a DLP policy that keeps that label out of Copilot where your licenses allow it, and the client's sites kept out of Copilot search. Where no setting can enforce a term, that client's data stays out of the AI workflow and the policy records the rule.

Can staff use approved agents without everyone building their own?

Partly. Microsoft documents no switch that stops people creating agents in Agent Builder. Admins can decide who may use agents and who may share them with the whole company, and can block or delete any agent; setting company-wide sharing to no users still lets people share with named colleagues. In Copilot Studio, makers can be routed to managed environments with sharing limits and connector rules. We pair those settings with a written rule, maker training and an agent register.

Who builds agents after you leave?

Your own staff, trained to build agents, under the rules we set up together. They build in approved workspaces with sharing rules, new agents go into the register with an owner and a review date, and your IT team approves any agent before it is shared company-wide and runs the reviews from the runbook. Run support can keep the registers current after handover, and we can build agents on the foundation as separately scoped work.

What does the EU AI Act ask of a company that uses Copilot?

For office use, mainly AI literacy. The AI Act requires measures that support AI literacy, sized to the role and the risk, and that duty has applied since February 2025. The European Commission says no certificate is needed and suggests keeping internal records. Office use of Copilot is not a high-risk use under the Act, although AI that screens job candidates or assesses staff would be.

What about Germany and the works council?

Copilot and the controls around it can monitor staff, so a German company with a works council needs its agreement before go-live, and German data protection authorities say a DPIA under GDPR will often be needed for AI. In the regulated track we prepare the DPIA with your data protection officer, the works council information pack and German versions of the documents, and your counsel gives the legal sign-off. Our teams in Texas and Munich cover the US and German sides of the rollout.

How is this different from a Copilot readiness assessment?

Most readiness assessments end with a report and a roadmap, and your team does the work afterwards. We do the hands-on work in your tenant: we fix the sharing, set up labels and DLP, write the policy and registers with your compliance lead, train each team and run the pilot. Then we hand every control to your IT team with a runbook and the acceptance tests they signed.

Is the free Copilot Chat safe to leave on for everyone?

It can be, once a few settings are chosen on purpose. When staff sign in with their work account, Copilot Chat runs under Microsoft's enterprise data protection, and prompts are not used to train the foundation models. What needs deciding is web search, what staff may paste in and which agents they may use. Early in discovery we set those, add a DLP rule that blocks web search when a prompt matches the sensitive information types we configure, and give staff an interim AI rule.

Research paper

Copilot after rollout: agents and guardrails, without agent sprawl

Who may build Copilot agents, the guardrails that need E5-level rights, and how to keep the agent count under control.

Read the paper
Get the designed PDF

Ready to ship AI that actually works?

Every engagement starts with one scoped workflow and a written definition of done. Commercial terms are set in the statement of work.

  • Built to SOC 2, HIPAA and GDPR standards
  • EU AI Act-aligned delivery
  • Microsoft Solutions Partner