Why Cheaper Copilot Still Won't Fix Your AI Problem

In December 2025, Microsoft cut the price of Copilot for small businesses from $30 to $21 per user per month — with promotional pricing as low as $18. That’s a 30–40% discount for organizations with fewer than 300 employees.

The timing is instructive. TechRadar ran a piece headlined “Almost 3 years later, it’s time to admit that Microsoft Copilot was a mistake.” Gartner’s 2025 survey of 187 IT leaders found that only 5% of organizations that finished piloting Copilot moved to larger deployment. Only 3.3% of Microsoft 365 users actually pay for it.

Microsoft’s diagnosis: too expensive. Their prescription: make it cheaper.

They’re solving the wrong problem.

Price Was Never the Barrier

If the problem were price, Copilot’s initial trajectory would have looked different. Companies didn’t try Copilot at $30, love it, and reluctantly cancel because of cost. They tried it at $30, found it underwhelming, and stopped using it — whether or not they cancelled the subscription.

Gartner’s analyst Dan Wilson put it plainly: “Impact, value and ROI measurements remained elusive.” Nearly half of the 127 respondents who had piloted or deployed Copilot rated it “some value, shows promise” — the polite version of “not worth the money.” Most organizations were “pausing and waiting it out.”

The MIT NANDA study found that 95% of enterprise AI pilots deliver zero measurable ROI. Not modest returns. Zero. One executive on LinkedIn described deploying 4,000 Copilot seats at $1.4 million annually. Three months later, 47 people had opened it. Twelve had used it more than once.

A 30% discount on a tool twelve people use doesn’t change the math. A tool that people actually use does. (We explored why practical AI tools beat strategy projects in detail — the research is unambiguous.)

The Three Problems a Price Cut Can’t Fix

1. The Amnesia Problem

Copilot doesn’t remember you. Every conversation starts from zero.

We covered this in depth in our honest Copilot comparison, but the core issue bears repeating: Copilot’s architecture is retrieval-first, not memory-first. It searches your Microsoft Graph for relevant documents and emails, then uses that as context. It doesn’t build a model of who you are, what you care about, or how you work.

Microsoft launched a “Copilot Memory” feature, but users report it’s inconsistent and frequently stops working. One Copilot Pro subscriber wrote: “The overall behavior is as if memory were disabled, even though it appears active in the interface.”

Cutting the price from $30 to $21 doesn’t give Copilot a memory. The same tool that forgot who you were yesterday still forgets who you are today — it just costs less while doing it. And when that forgetfulness hits organizational knowledge, it compounds the knowledge drain problem that already costs companies millions.

2. The Single-Agent Ceiling

Copilot is one generic model applied to every M365 application. The same AI that drafts your emails writes your financial reports and summarizes your sales pipeline.

There’s no specialization. A 2025 study found that domain-specific agents are 37.6% more precise than generalist agents. Google Research confirmed through controlled evaluation of 180 agent configurations that multi-agent coordination dramatically improves performance on parallelizable tasks. Enterprises using multi-agent systems spend 61.2% less time correcting AI outputs.

Microsoft offers Copilot Studio for building custom agents, but a brutally honest review described it as “a platform of contradictions” — connected agents can’t run their own tool servers, version control between Teams users is broken, and lifecycle management produces vague SQL errors.

A cheaper Copilot at $21 is the same single agent. It doesn’t gain specialization because the price dropped.

3. The Data Residency Gap

Microsoft completed its EU Data Boundary in February 2025. Customer data at rest stays within the EU. But when a German employee types a prompt, the processing happens wherever Microsoft has available LLM capacity.

The situation got worse in January 2026. Anthropic became a subprocessor for Microsoft 365 Copilot, but the Claude AI model runs exclusively on AWS in the United States. Data submitted to Copilot’s Researcher Agent is processed outside the EU — and is explicitly excluded from the EU Data Boundary and in-country processing commitments.

For companies with GDPR and AI compliance obligations, a cheaper Copilot with the same data residency gaps changes nothing. You don’t get German data processing at $21 any more than you did at $30.

What Microsoft Is Really Telling You

When a company cuts prices by 30–40%, it’s not because the product is succeeding. Winning products get price increases, not fire sales.

Microsoft’s own numbers tell the story. more than 450 million M365 commercial seats worldwide, only 15 million paying for Copilot. That’s a 3.3% attach rate — after two years of the most aggressive enterprise AI push in history.

The price cut is an admission: at $30, the value proposition didn’t hold. But the reason it didn’t hold has nothing to do with the number $30 and everything to do with what you get for it.

What the Research Says Actually Works

The Harvard Business School and BCG study that we discussed in our piece on why companies need an AI colleague, not a strategy showed consultants using AI completed 12.2% more tasks, finished them 25.1% faster, and produced 40% higher quality results. The key factor wasn’t price — it was practical utility embedded in daily work.

Stanford’s Erik Brynjolfsson found 14% average productivity gains from AI assistants, with the least experienced workers seeing up to 35%. McKinsey’s employee survey showed that the two factors driving AI adoption were formal training (48%) and “seamless integration into existing workflows” (45%).

Not cheaper tools. Better tools. Tools that remember. Tools that specialize. Tools that are already there — inside the messaging app your team has open eight hours a day, not in a separate browser tab.

The Question to Ask

When evaluating an AI assistant for your team, the question isn’t “can we afford it?” It’s “will people actually use it?”

Microsoft’s price cut answers the wrong question. The right question is architectural:

• Does it remember? Not “can it search your SharePoint” — does it genuinely remember who you are, what you’re working on, and how you prefer to communicate?
• Does it specialize? Not “can it draft in Word and Excel” — does it have purpose-built agents for different domains that deliver expert-level results?
• Does it work where you work? Not “can you open another app” — is it already there, in the conversation, before you think to ask?
• Does it stay compliant? Not “is it on a roadmap” — is your data processed in your jurisdiction today?

If the answer to any of these is no, a price cut doesn’t fix it. You’d be paying less for the same limitations. And your team — the people who already solve the shadow AI problem by bringing their own tools because the sanctioned ones aren’t good enough — will continue doing exactly that.

The Bottom Line

Microsoft cut Copilot’s price because adoption stalled. The evidence is overwhelming: 5% broader deployment rate, 3.3% attach rate, “elusive” ROI. Their response was to lower the barrier to entry.

But the barrier was never the price. The barrier was a stateless, single-agent, architecturally limited tool that doesn’t deliver enough value to justify any price — whether that’s $30 or $21 or $18.

amaiko takes a different approach: persistent memory that grows with every interaction, multi-agent specialists that deliver domain-specific precision, Teams-native integration that eliminates context switching, and German data residency — not on a roadmap, but today. It costs more than discounted Copilot. It also actually works.

The cheapest AI isn’t the one with the lowest price tag. It’s the one that delivers enough value that nobody questions the investment.