Microsoft Spent Billions on Copilot. 3.3% of Licensed Users Pay to Use It.

Microsoft Copilot's adoption rate sits near 3.3% of the 450 million-plus Microsoft 365 users who technically have access to it, and most government pilots land closer to 30-40% active use, well under the 65-85% needed to justify the license cost. If your agency is weighing whether to expand Copilot seats next fiscal year, that gap between "licensed" and "used" is the number to model first, not the sales deck.

Guilford County, North Carolina didn't theorize about this. It piloted 50 Microsoft 365 Copilot licenses starting October 1, 2026, at an annual cost of roughly $18,450. That works out to about $30.75 per user, per month. Fifty seats. One county. A real, dated, board-approved number, not an analyst's projection.

Scale that math up and the picture gets less comfortable. A 5,000-seat deployment runs close to $1.8 million a year in licensing alone, before a single hour of training or change management. And Microsoft raised Copilot pricing again effective July 1, 2026, with blended per-seat costs now running $23.50 to $90 a month depending on the base plan underneath it. Layer Copilot onto an E5 license and per-user cost climbs 53%. Layer it onto Business Basic and it climbs 500%.

The Microsoft Copilot Adoption Rate Nobody Puts in a Slide Deck

Here's the part procurement officers should ask about directly, because it rarely shows up unprompted. The UK government ran a 20,000-user Copilot trial and peaked at 23-24% adoption. Only 8% of employees prefer Copilot over ChatGPT or Gemini when given a choice. Four complaints repeat across the enterprise market: low voluntary adoption, hallucination risk in compliance-sensitive workflows, security exposure amplified by permissions nobody cleaned up first, and a persistent gap between license cost and actual return.

That third one matters more than it sounds. Copilot doesn't grant new access, it retrieves through whatever permissions already exist, so a SharePoint site shared to "everyone except external users" back in 2019 becomes instantly findable the moment Copilot can see it. EPC Group's readiness audits find an average of 150-300 overshared SharePoint sites per enterprise tenant, and fixing that before go-live runs $50,000 to $150,000 depending on tenant size. Our SharePoint permission audit checklist covers that cleanup work if security review is what's actually slowing your rollout.

Why Adoption Stalls: The Training Burden

Training alone runs $200 to $500 per user. Change management eats another 5-10% of the deployment cost every year it's live, not just at launch. None of that shows up in the per-seat sticker price, and none of it is optional if you want staff to actually open the tool instead of ignoring it after the rollout email.

Part of why that burden is so heavy: Copilot is a general-purpose assistant. It drafts documents, summarizes meetings, writes code, and searches internal content, all inside one interface, which means every employee needs at least baseline fluency across all of those modes before the tool earns its keep. That's a lot of surface area to train against for a 2,000-employee agency with a 3-person IT shop.

The Real Copilot ROI Government Agencies Should Model

Even well-run Copilot programs model out to 1.5x-3.0x Year 1 ROI, not the order-of-magnitude return implied in a lot of vendor messaging. That's not a bad number, it's just a different number than the one usually pitched in the room. Model it against the 65-85% adoption threshold needed to hit even that range, then check it against your agency's actual pilot usage data, not the vendor's enterprise-wide average. None of this makes Copilot the wrong tool for drafting and summarizing. It means a general-purpose assistant and an internal search tool solve different problems, and budgeting for them identically is where agencies get burned.

A Narrower Tool Carries a Narrower Training Burden

Keyspider's Workplace Search exists for one job: an employee asks a plain-English question and gets an answer with a citation, pulled from across SharePoint, Drive, Confluence, ServiceNow, and the other systems staff already touch. It doesn't draft a board packet or write code, and that narrower scope is why the training burden drops. One interaction pattern to learn, not five.

We don't have a published Keyspider-specific adoption percentage, and we're not going to invent one to win an argument. What we can point to directly: deployment runs about two weeks with no migration or consolidation project first, and agencies running it report saving staff 30-45 minutes a day they used to spend hunting for documents, with ticket volume down 30-40% within 90 days. It also won't fix a SharePoint site overshared since 2019, no search layer does that, Keyspider included. What it does do is mirror whatever permissions the source system enforces and log every search and click across all 40-plus connected systems, a real answer to the records-retention and subpoena questions county commissioners are already asking out loud. If you're sizing up Workplace Search against a Microsoft-centric plan, that audit trail belongs on the comparison sheet.

On price, we won't pretend there's a published workplace-search-only number either. Workplace Search ships inside Keyspider's Enterprise tier, and like most of this market, that tier is quote-based. Check current pricing and ask for a number sized to your seat count, the same way you'd ask Microsoft for one. Agencies already buying through a cooperative vehicle, NASA SEWP V, NASPO ValuePoint, OMNIA Partners, TIPS, or Sourcewell, can run it through Carahsoft without opening a new procurement just to get a quote on the table next to Copilot's.

Model the adoption rate your agency actually gets, not the one in the enterprise average. The gap between those two numbers is where the budget case either holds up or doesn't.

For the procurement side of this, our RFP question bank for workplace search includes the specific cost and permission questions worth putting in writing before you sign anything, whether the vendor on the table is Microsoft or anyone else. It compares well against the cost-of-silos argument we've made before in our piece on workplace search ROI, which tackles the other side of this budget conversation: what it costs *not* to fix internal search at all.

See how Keyspider Workplace Search prices out against the licensing math above. Book a demo and bring your current Copilot seat count.