The Shadow AI Assessment Guide

1. Why Shadow AI Is the First Conversation

By the time a partner sits down to talk about a Copilot license, the AI conversation has usually already started — just not with IT in the room. Employees are pasting client emails into free ChatGPT accounts, dropping proprietary code into personal Cursor or Claude sessions, and running meeting recordings through consumer transcription apps. None of it shows up in a firewall log, because every one of those tools is a trusted TLS endpoint that every URL filter waves through as "productivity."

2. The Shadow AI Reality — What SMBs Actually Have Running

2.1 · The pattern you will find on nearly every engagement

2.2 · The categories you'll actually find

The following categories show up on nearly every SMB discovery scan, well before any purpose-built discovery tool is even switched on. Treat this as the starter checklist for the conversation, not an exhaustive taxonomy.

3. Microsoft's Four-Step Model — Discover, Block, Protect, Govern

Microsoft publishes a formal deployment model in Microsoft Purview called Prevent data leak to shadow AI. It's worth anchoring to because it's the reference architecture a customer's auditors, cyber insurer, and Microsoft account team already expect to see reflected in a serious engagement.

3.1 · What each step actually means for a partner

3.2 · The two-layer simplification worth teaching every customer

Microsoft MVP Nikki Chapple frames the same architecture as two decisions, and it's the cleanest version of this story for an executive conversation:

4. The Assessment Delivery Model

4.1 · The five-capability self-audit — can you deliver this today?

Before quoting a Shadow AI assessment, run the same five-capability audit on your own tenant first. Selling discovery you haven't personally survived is the fastest way to under-scope a SOW — dual-track it: run the snapshot on yourself, then sell it to customers.

4.2 · The engagement flow — from first meeting to managed governance

4.3 · Realistic timing and staffing for a 25–300 seat customer

4.4 · What the finished engagement produces

5. A Sample Engagement — 120 Users, 50 Copilot Seats

This is a composite, representative walkthrough built to teach the pattern of a common SMB deal — the numbers are illustrative of what a real 14-day scan at this seat count typically turns up, not a specific named customer's data.

5.1 · The profile

• Company: 120 employees, professional services
• Current licensing: 90 Business Premium, 30 Business Standard
• The ask: 50 Microsoft 365 Copilot licenses for finance, senior consultants, and leadership
• Doesn't yet own: Defender for Cloud Apps (standalone), the Purview Suite add-on
• Already owns via Business Premium, mostly unconfigured: Entra ID P1, Intune, Defender for Business, Purview sensitivity labels and basic DLP

5.2 · Handling "we don't own Purview or Defender for Cloud Apps"

This is the objection that stalls the deal if it's handled as a forklift upgrade to E5. It isn't one. Microsoft sells a Defender Suite and a Purview Suite specifically scoped to Business Premium, and the two can be bundled at a meaningful discount when purchased alongside Copilot. The pitch is not "buy more Microsoft" — it's "the tool that powers this assessment is the same tool that governs Copilot after it's turned on."

5.3 · Representative findings, at this seat count

• 11 distinct AI apps discovered — led by general chat assistants, meeting transcription, and personal writing tools.
• 43 sensitive-info-type hits in the DLP audit log — client emails, identifying information, engagement-letter templates.
• 12 users authenticating to AI tools with a personal Microsoft or Google account rather than corporate identity — the strongest single signal that governance, not just policy, is missing.
• 3 SharePoint sites classified "Confidential" turn out to be shared far more broadly than intended — exactly the kind of finding that becomes a hard gate before any AI grounding product goes live.

6. The Licensing Motion — Positioning the Suites

6.1 · What changed for SMB licensing

• Microsoft sells Defender and Purview Suite add-ons scoped to Business Premium — SMB-specific bundles that unlock governance capability that used to require an E5 upgrade.
• Confirm current pricing directly on Microsoft's Business add-ons page before quoting; list prices and bundling change (see the Reference Library for the live URL).
• Microsoft has run promotional pricing on the Purview Suite when purchased alongside Microsoft 365 Copilot — worth checking at the time of the deal.

6.2 · The upsell logic worth teaching

7. Sanction, Don't Just Block

The instinct after a discovery scan is to block everything unsanctioned. Resist it. Blocking without sanctioning drives usage further underground — onto personal devices and personal networks the partner has zero visibility into — and it burns the goodwill that made the customer receptive to the conversation in the first place. The credible position is narrower and more durable: bring the tools people already trust under management, rather than take them away.

The plan for a customer who won't adopt Microsoft's own AI stack

8. The Top AI Apps — Enterprise Onboarding Matrix

Keep this as a living document — expand it quarterly. Pricing and SSO/SCIM support change often enough that this table intentionally omits dollar figures; confirm current terms directly with each vendor before quoting a sanctioned-app onboarding SOW.

For each partner-sanctioned app, the onboarding pattern is the same six steps:

1. Add the app via the Entra ID enterprise app gallery, or configure a custom SAML app.
2. Configure single sign-on with the vendor's SP metadata.
3. Configure SCIM for lifecycle management — this is what makes the AI account disappear the day the employee does.
4. Configure the vendor's admin console: data retention, model access, audit export.
5. Get the data processing agreement signed and complete a privacy review.
6. Mark the vendor's domain Sanctioned in Defender for Cloud Apps, and add it as a monitored destination in Purview DLP.

9. Finding the Conversation — Reading Partner Signal Data

Microsoft's partner-facing propensity tooling in Partner Center — surfaced through the AI Business Solutions & Security Partner Experience (ASPX) — exposes weekly-refreshed, per-tenant signals that map directly onto a Shadow AI conversation. If your organization has Partner Center access, these are worth building into a weekly call-list routine rather than treating as background reporting.

10. The Revenue Ladder — Six Service Lines

The assessment is the land. What makes the practice durable is that every rung above it has a natural trigger from the one below — nothing has to be sold cold. Pricing ranges below are editorial estimates meant as a starting point; test them against your own cost model before quoting.

11. Objection Handling

12. Companion Assets Worth Building

Ten asset concepts worth productizing on top of this guide — each one shortens the sales cycle for a rung on the ladder in Section 10.

13. Where the Premium Sits — Regulated Verticals

Vertical-specific work in a regulated customer is genuine additional scope, not padding — and it's usually visible enough that the customer recognizes it as real. Community-observed pricing premiums of roughly 20–40% above generalist rates in these verticals reflect that extra work, not markup for its own sake.

14. "It's Not a Big Deal" — And Why That's Wrong

15. The Partner's Non-Negotiables

16. Next Steps

What to build first, in rough priority order:

1. Publish this guide as the standing Shadow AI reference page, and link it from every related asset on the site.
2. Ship the License Uplift Calculator (companion asset 9) as an embedded, interactive tool.
3. Turn the AI App Onboarding Matrix (Section 8) into a living page with a quarterly refresh cadence.
4. Package the AUP template, executive read-out deck, QBR pack, and the six SOW templates as a downloadable, email-gated partner kit.
5. Ship the "Sanction, Don't Block" briefing as an open, ungated download — it's the objection-handler most partners will forward inbound.
6. Add a regulated-vertical addendum library as expandable subsections under each vertical in Section 13.
7. Wire the partner-signal runbook (Section 9) as a partner-only page behind Partner Center authentication.
8. Cross-link every asset to its named source — Chapple, Tenant Wizards, CIAOPS — so the practice stays grounded in named, checkable authority.

17. Reference Library & Further Reading

Every linked item below was checked at the time this guide was written. Product pages, pricing, and licensing tables change on Microsoft's own schedule — re-verify before quoting a customer. Items without a live link are cited by name only; confirm the current URL before publishing them externally.

Microsoft, official

• Prevent data leak to shadow AI — Microsoft Purview deployment model
  The official four-step Discover / Block / Protect / Govern architecture this guide is built around.
• Microsoft Purview deployment models — overview
  The full family of scenario-based models, including Secure and Govern Copilot Agents.
• Data Security Posture Management (DSPM) for AI
  The in-portal visibility surface for third-party AI app interactions and Copilot activity.
• Microsoft Purview browser extension
  Endpoint DLP extended into the browser tab, for Edge, Chrome, and Firefox.
• Manage generative AI apps — Defender for Cloud Apps
  The Cloud App Catalog's generative-AI category, sanctioning, and risk scoring.
• Shadow AI discovery — Microsoft Entra Internet Access
  Network-layer detection of unsanctioned AI apps, ahead of app-level classification.
• Microsoft 365 Business — additional services, plans, and pricing
  Current pricing for the Defender and Purview Suite add-ons scoped to Business Premium. Confirm live pricing here before quoting.
• Microsoft 365 Copilot vs. ChatGPT Enterprise
  Microsoft's own comparison of grounding, permissions, and governance boundary.

MVP and community authority

• Nikki Chapple — How to govern shadow AI using Microsoft Purview, Defender, and DSPM
  Source of the "control the apps / control the data" framing used in Section 3.
• Nikki Chapple — Microsoft Purview browser extension: governing AI shadow IT
• Danny Vorst / Tenant Wizards — Shadow AI in Microsoft 365: find and block it with Purview
  A practitioner walkthrough of the same four-step model with configuration detail.
• Robert Crane / CIAOPS — Microsoft Defender and Purview Suites for M365 Business Premium
  The detailed licensing breakdown behind Section 6. Verify against the live Microsoft pricing page above.
• ShadowLock — State of Shadow AI (2026)
  Community synthesis of Gartner, Microsoft, IBM, Cyberhaven, and Netskope research; source of the ~75% and $400K figures in Section 2. Treat as directional and re-confirm before quoting.

Research and benchmark data

• Gallup — AI use at work rises (2025)
  Source for the AI-tool-preference figures in Section 2 — re-check the exact percentage in the current edition.
• McKinsey — The State of AI (2025)
  Source for the organizational AI-adoption figure.
• Verizon Data Breach Investigations Report
  Source for the AI-bolstered breach-technique figure. Confirm the current edition's exact number before quoting.
• PagerDuty — 2026 shadow AI workplace survey
  Source for the 66% policy-violation admission figure.
• Cynomi — State of the vCISO 2025
  Source for the 21%→67% year-over-year vCISO-adoption figure in Section 10.
• Kaseya — 2026 State of the MSP Report
  48% of MSPs rank AI/automation as their #1 client priority for 2026; only 13% currently monetize it meaningfully — the gap this guide exists to close.