Your complete guide — the conversation, the tools, the licensing, and what comes next.
Do not sell Cowork. Sell the management of Cowork.
Before the minute-by-minute detail, see the whole shape at once: seven beats inside the 40 minutes, then the branch you should be walking toward the moment you finish. Nothing below is optional filler — every box on this map gets its own fully-scripted stop in Section 4, and the branch at the right gets its own roadmap in Section 11.
Figure 0 — Beats 1–3 (teal) place Cowork on the map and open the governance wedge. Beats 4–5 (dark teal) survive the two objections that actually kill deals. Beats 6–7 (purple) convert the sale into a named next step — which is also where the session hands off into the six-rung ARR ladder that runs for months afterward.
Forty minutes is not enough time to teach Cowork. It's enough time to give you the four moves you need to have the conversation — and this design is built around exactly those four, in order. This guide picks up exactly where the Shadow AI discovery conversation leaves off: your customer has a Copilot license, or is about to, and now you have to decide what to say about Cowork.
(a) explain Cowork's place in the customer conversation without a demo, (b) run the co-sell talk track against a Claude or ChatGPT preference, (c) size Cowork consumption for a customer's monthly bill on paper, and (d) name the two follow-on SOWs you'll pursue after any Cowork sale.
Everything below is a building block, deliberately. Each stop in Section 4 is independently reusable in a one-on-one customer meeting, a lunch-and-learn, or a QBR — if you only retain Stop 4 (the reframe) or Stop 5 (the cost story), you still walk out able to run half the conversation.
This conversation is deliberately low-tech in delivery — no live tenant, no login risk — but it leans on four real tools you'll use live, not just cite afterward. Know each one cold before you walk in.
Microsoft's official interactive credit-forecasting tool. Enter agent type, licensing option, and expected feature usage; it returns a monthly Copilot Credit range. This is the tool of record — the Cowork Cost Simulator below is the co-branded, offline version of the same math.
The browser-based, offline, PDF-exportable partner tool — your go-to for the live moment. It runs as a guided interview: one question per screen, with the words to say sitting beside every field, and the full output held back until the end so the customer answers before they see a number. Same math as the Credit Estimator, aligned to Microsoft's published task-mix defaults, but built to print a co-branded leave-behind on the spot.
Three-column comparison — Business Premium alone, Business Premium + Defender Suite + Purview Suite, and E5 — with a line-by-line capability delta. Proves the SMB-native path (BP + Suites) covers what most Cowork rollouts need without an E5 forklift.
The partner-facing propensity portal. Weekly-refreshed, per-tenant Copilot and Cowork signals — this is where your named account for the closing exercise should come from, not a guess.
If you haven't run the Shadow AI discovery conversation with this account yet, spend thirty extra seconds on the one sentence that connects the two: “Cowork grounds in your customer's M365 data through Work IQ, and it honors whatever Purview labels are already there. Ungoverned data in, unreliable output out — at the same credit cost.”
Three acts in forty minutes — each one a building block you can redeploy on its own. The pacing is intentionally front-loaded on objection handling because that's where deals actually die: not in the pitch, in the second question from the CFO or the CIO who already has a Claude subscription.
Where Cowork sits relative to Copilot Chat and the M365 Copilot seat, and why the Shadow AI conversation had to come first. No pitch yet — just the mental model the rest of the session hangs from.
“We already use Claude” and “this is going to cost a fortune.” Both get a reframe and a number, not a slogan.
Not a wrap-up — a commitment. You leave with a named account, a booked call, and the two SOWs you'll pitch next.
Every stop below has the same three-part anatomy: what you say (the talk track), what you show (the visual or comparison), and the segue — the exact line that pulls the conversation into the next stop. Read it top to bottom once before your first live run; after that, the segues alone are enough to keep you on track.
This is a brand-new opportunity — the customer isn't running Cowork yet. So don't open with someone else's usage report; open with the question that actually got you in the room.
Their team already runs Claude — sanctioned or shadow — and someone just spotted “Cowork” surface inside Microsoft 365 Copilot.
An exec or your IT lead sat through a Cowork demo at an industry event or a Microsoft partner conference, walked away excited, and wants it running before anyone's checked the cost or whether their data's ready.
An early adopter ran a Cowork task, it impressed a VP, and now there's executive pressure to roll it out with no plan for cost or governance.
| Surface | What it is | Buying motion |
|---|---|---|
| Copilot Chat | Fast, grounded, single-conversation work — draft, summarize, ask | Included in the M365 Copilot seat |
| M365 Copilot | Copilot inside Word, Excel, PowerPoint, Outlook, Teams | Fixed per-user/month via CSP — the ARR anchor |
| Copilot Cowork | Agentic execution of long-running, multi-step work — returns a completed result, not a draft | Usage-based, Copilot Credits, PayGo or pre-purchase |
Cowork “executes complex, long-running, multi-tool tasks… and returns a completed result, not just a draft or a recommendation.”— Charles Lamanna, Microsoft 365 Blog, GA announcement, June 16, 2026
Sensitivity label inheritance on outputs · audit logging of prompts/responses · eDiscovery on Cowork content · Data Lifecycle Management & retention · Communication Compliance policies
DLP specific to Cowork (Microsoft lists it “coming soon”) · Cowork interactions in the Apps & Agents dashboard · Cowork interactions in the AI observability page
The highest-leverage eight minutes in the session.
“Keep Claude. Bring it under management. Cowork is what runs the work your Microsoft 365 tenant already knows about.”The line to use three times per Cowork conversation
The underlying model is the same family — Copilot Cowork runs Anthropic Opus 4.8 and Sonnet 4.6 at GA, with GPT 5.5 in Frontier and a lower-cost Cowork 1 model coming. Don't argue “our model is better” — you'll lose that argument. Instead: Microsoft's own disclosed internal testing showed Copilot Cowork averaging 30–40% cheaper per prompt than Claude Cowork with its Microsoft 365 connector, on the same model, same tasks (125 runs, 12 prompts, both on Opus 4.8, June 2026).
Outside the trust boundary. No Purview labels, no Defender for Cloud Apps visibility, no Cost Management, no Entra identity binding. The model may be excellent — the operations around it don't exist.
Inside the Microsoft 365 trust boundary. Purview, Defender for Cloud Apps, Entra, and Cost Management all apply automatically. Don't out-argue model quality — out-argue enterprise operations.
Want the full dimension-by-dimension teardown (billing, regional constraints, partner monetization)? That's Section 5 — keep it in your back pocket for the CIO who wants receipts.
| Persona | Light tasks/mo | Medium | Heavy | Credits/user/mo | Est. $/user/mo |
|---|---|---|---|---|---|
| Corporate Knowledge Worker | 22 | 11 | 5 | 14,250 | $142.50 |
| Customer-Facing | 17 | 13 | 5 | 14,625 | $146.25 |
| Technical | 12 | 9 | 14 | 22,800 | $228.00 |
| Manager / Senior Leader | 13 | 6 | 3 | 8,225 | $82.25 |
Monthly Cost Management + Cowork Usage review, spending-policy tuning, per-user limit adjustment, concentration-risk check. The recurring hinge of the whole ladder.
The SharePoint/Teams cleanup that makes Cowork's grounding trustworthy in the first place — priced as its own line, not folded into the license quote.
AI Acceptable Use Policy authoring ($1K–$5K), sanctioned third-party AI onboarding under Entra SSO ($3,500–$5,000/app), and a vCISO retainer for regulated verticals ($2,600–$15K/mo).
Stop 4 gives you the reframe. This section is what's behind it — pull it out when a colleague, or a customer later, wants the full dimension-by-dimension teardown.
| Dimension | Microsoft Copilot Cowork | Claude Cowork (Anthropic) |
|---|---|---|
| Model | Opus 4.8 & Sonnet 4.6 at GA, GPT 5.5 in Frontier | Anthropic Claude models |
| Deployment | Cloud, integrated with M365 tenant | Desktop (macOS/Windows/Linux) + web + mobile |
| Prerequisite | M365 Copilot USL (~$30/user/mo) + usage | Claude Pro at ~$17/mo (annual) |
| Effective monthly cost | $30 license + variable usage — typically higher for active users | $17 (Pro) — agentic use may push toward Max |
| Data grounding | Work IQ — automatic across email, meetings, files, chats, people | Explicit per-task connection; M365 connector available |
| Governance | Purview (DSPM, labels, audit, eDiscovery, IRM, DLM, Comm Compliance); Defender for Cloud Apps; Entra | SOC 2 II, ISO 27001/42001, HIPAA; Cowork activity not yet in audit logs per Anthropic |
| Cost controls | Spending policies at tenant/group/user; alert thresholds; PayGo or P3 | Flat subscription; no partner-managed spend controls |
| Billing | Copilot Credits via Azure/CSP; PayGo $0.01/credit or P3 discount | Direct subscription with Anthropic |
| Partner monetization | Recurring CSP margin + governance retainer + adoption + AgentOps | One-time implementation; no CSP resale; no tenant to manage |
| Regional constraints | EU/EFTA/UK/gov-cloud tenants face Anthropic sub-processor constraints; gov cloud blocked | Broader macOS/Linux support, but no M365-native controls |
Same model family under both. Cowork's edge isn't the model — it's what wraps it: governance, cost controls, and a tenant you already manage.
Microsoft's disclosed internal testing showed Copilot Cowork 30–40% cheaper per prompt on the same model, same tasks. Offer to run the comparison live.
Fine for one person. Ask who manages spend, audits prompts, and enforces DLP across the team — that's the tenant story only Copilot Cowork has.
True — but Cowork activity isn't yet captured in Anthropic's own audit logs or Compliance API. Purview captures it today, inside your existing estate.
Simpler until spend needs a cap, an owner, or an auditor. That's exactly the governance retainer the partner is built to run.
Don't claim Cowork is universally cheaper — the 30–40% figure is specific to Microsoft's tested prompt set.
Don't claim DLP is available for Cowork today — it's “coming soon.”
Don't claim Cowork appears in every Purview surface today — Apps & Agents and AI observability don't show it yet.
Don't sell Cowork into EU/EFTA/UK/gov-cloud tenants without checking Anthropic sub-processor status.
Don't sell Cowork before the customer has a CSP-channel Azure subscription — they'll hit the billing wall.
Stop 6's objection “we don't own Purview or Defender for Cloud Apps” only lands cleanly if you have the current SMB licensing math memorized. Here it is.
Business Premium customers hear this as “double my Microsoft bill” — deal stalls.
“The Copilot AI engine is identical on Business Premium and E5” — Microsoft states this plainly. The difference is the governance stack in the base SKU, not the Copilot feature set.
Business Premium + Copilot + Purview Suite add-on + optional Defender Suite add-on. Same Copilot experience, right governance stack, no E5 forklift.
DSPM for AI ships in the Purview Suite for Business Premium add-on for third-party AI monitoring, but full Copilot data assessments may still require E5 Compliance in some configurations. Confirm against current licensing tables before quoting.
This is exactly what the License Uplift Calculator (§2) is built to walk a customer through line by line — three columns (BP alone / BP + Suites / E5), each Shadow AI capability marked ✓ / partial / ✗, so nobody has to take the delta on faith.
| Objection | The reframe | The evidence |
|---|---|---|
| “Our staff isn't using AI.” | The bet is on you naming exactly how many are — before an audit does. Most AI-using employees are on personal, unmanaged accounts. | Free Copilot MAU from ASPX + a 14-day Defender for Cloud Apps scan closes this in three weeks |
| “We already use Claude — why Cowork?” | Same-caliber model, different governance, different grounding, different identity boundary. Keep Claude. Add Cowork. | Stop 4 script · §5 deep comparison |
| “Copilot is too expensive.” | You're already paying for AI — it's just not on your P&L. Employees are on personal subscriptions today. Governance is the difference between exposure and asset. | Independent unmanaged-AI exposure research (§13) |
| “Cowork credits are unpredictable.” | Set caps and per-user alerts before the pilot. Month one gives you the baseline; treat it as an operating discipline, not a surprise bill. | The month-one admin field report (§13) |
| “We'll just block ChatGPT and Claude.” | Don't. Blocking without sanctioning drives Shadow AI further underground — to personal devices you can't see at all. Sanction one or two, monitor the rest. | The two-layer governance model (§13) |
| “We don't own Purview / Defender for Cloud Apps.” | The Business Premium Defender + Purview Suite add-ons close the gap at a fraction of E5 cost, discounted alongside a Copilot purchase. | Section 6 licensing snapshot |
These are the bridge from this conversation to your first customer call. Better still: don't start from a blank slate. Pull your top account through ASPX first (§2) so Question 8 already has a data point behind it before the customer even answers.
The forty-minute design is deliberately spare. When the agenda allows a longer block, these five exercises turn the same building blocks into hands-on practice.
The index card from Stop 7 is the first rung of a ladder, not a one-off sale. Six service lines compound off the same 4-week discovery engagement that starts with the call you just booked — training runs underneath every rung as the recurring evidence layer that keeps audit and insurance renewals painless.
Every dollar on this ladder sits in one of three buckets: Project/SOW (Rungs 1, 2, 4), Managed Governance/AgentOps (Rungs 3, 6), or Outcome-Based (Rung 5). Naming them once keeps the ladder legible and stops partners from accidentally price-anchoring on hourly billing.
Once a customer is live on Cowork, the Managed AI Governance retainer (Rung 3) has concrete monthly work behind it — pulled straight from the field-tested Cowork admin playbook:
| Admin view | Best for | Date behavior | Refresh behavior |
|---|---|---|---|
| Copilot → Cost Management | Credits, limits, policies, users/groups, agents/services, prepaid vs. PayGo | Supports an exact calendar month | Overview every 4 hrs; Consumption every 2 hrs |
| Copilot → Cowork → Usage | Active users, tasks, scheduled vs. user-initiated, retention, adoption | Rolling 7/28/90/180 days; default 28 | Check the visible “Last updated” timestamp |
The monthly scorecard you should ship on that retainer: Cowork Credits used (exact calendar month), active users and sessions, median/top-5 concentration, % tasks scheduled vs. user-initiated, retained users week-over-week, and users at ≥50%/≥80% of their limit as the early-warning list.
| Horizon | Partner motion |
|---|---|
| Day 30 | First call made on the named account; Azure plan live on at least one customer tenant; Cost Management review scheduled. |
| Day 60 | First Managed AI Governance retainer sold; oversharing remediation scoped or completed on the pilot customer. |
| Day 90 | The retainer motion repeatable across three or more accounts; seller team drilled on the Claude objection; practice revenue reported against the base-case model. |
Every figure, quote, and talk track in this guide traces to one of the sources below. Cite these by name when someone pushes back — the credibility is earned, not asserted.
This guide is designed to stand alone as a downloadable resource on copilotplaybook.com, cross-linked to its companion documents so the credibility chain — Microsoft official, named MVP, internal playbook — stays visible to any partner who opens it cold.