Distribution is a requirement now
Why AI startups without channels are about to get squeezed — and what enterprise buyers should do about it.
- 1.The era where a better model wins is over — distribution and channel access now determine enterprise AI winners.
- 2.Three use cases consistently generate positive ROI: customer service automation, code generation, and compliance document processing.
- 3.Enterprise AI token spend is up 3.4x YoY, but 62% of organizations have no centralized governance over it.
- 4.McKinsey’s $4.4T AI value estimate obscures the reality that most enterprise AI is still in pilot purgatory.
This week DeepSeek released R2, Google announced Gemini 2.5 pricing that undercuts everyone, and Microsoft quietly rolled Copilot into 40% of Office 365 enterprise seats. If you’re an AI startup without distribution, you watched three companies with distribution eat your lunch in a single week.
The era where a better model wins is over. The era where a better model plus a channel wins is just starting. For enterprise AI buyers, this means the vendor landscape is about to consolidate fast — and the ones left standing will be the ones already embedded in your workflows.
More on this in the deep dive below.
“The era where a better model wins is over. The era where a better model plus a channel wins is just starting.”
What happened this week
The cost story matters more than the capability story at this point. For internal enterprise workloads where “good enough” is the bar — document processing, summarization, first-draft generation — the cheapest model that clears the threshold wins. This is deflationary for the entire AI services market.
This is the distribution play in action. Microsoft doesn’t need Copilot to be the best AI product. It needs to be the most convenient. When it’s already in your Office license, the switching cost to a point solution is enormous. Heads of AI should be auditing which Copilot features their org is actually using vs. paying for.
The features no one talks about on Twitter are the ones that close enterprise deals. SSO, SCIM provisioning, data residency, audit trails. Anthropic is clearly going after regulated industries. If you’re in financial services or healthcare, this is the release that matters most this week.
Why McKinsey is wrong about enterprise AI
Their $4.4T estimate misses the structural shift happening underneath
McKinsey’s widely-cited estimate that AI will add $4.4 trillion in value to the global economy reads like a pitch deck, not research. It aggregates productivity gains across every industry and function into a single eye-popping number that’s impossible to falsify or act on.
The more useful question — the one enterprise AI leaders are actually asking — is not “how much value will AI create?” but “where is AI creating measurable ROI right now, and where is it still burning cash?”
From our early research conversations with F500 AI leads, the answer is surprisingly narrow. Three use cases consistently generate positive ROI: customer service automation (30–50% cost reduction in tier-1 support), code generation (15–25% developer productivity gains, measured in PRs merged), and document processing in regulated workflows (60–80% time reduction in compliance review).
| Use Case | ROI Metric | Range |
|---|---|---|
| Customer Service Automation | Cost reduction in tier-1 support | 30–50% |
| Code Generation | Developer productivity (PRs merged) | 15–25% |
| Compliance Doc Processing | Time reduction in review | 60–80% |
Source: Arcana Enterprise AI Buyer Interviews, Q1 2026 (n=23)
Everything else — AI-powered sales, marketing content generation, “strategic decision support” — is still in the experimentation phase at most organizations. The ROI is either unmeasured or negative when you account for the full cost of tokens, engineering time, and change management.
This is the gap between McKinsey’s top-down projection and bottom-up reality. The $4.4T number assumes broad adoption across all functions. The reality is that enterprise AI is concentrated in a handful of proven use cases, with everything else still in pilot purgatory.
For AI leaders, the implication is clear: double down on what’s working, kill the science projects, and benchmark your spend against peers. Which is, of course, exactly what we’re building.
Based on early data from our enterprise AI buyer conversations, organizations are spending 3.4x more on AI inference tokens compared to the same period last year. The increase is driven primarily by internal productivity tools (chat assistants, code generation) rather than customer-facing AI products. Notably, 62% of respondents report having no centralized governance over token spend — individual teams are provisioning and paying independently.
Source: Arcana Enterprise AI Buyer Interviews, Q1 2026 (n=23, preliminary)
Claude Projects for enterprise research
If you’re doing competitive analysis or market research, Claude Projects is the most underrated tool available right now. Create a Project, upload your company’s latest 10-Ks, competitor filings, and analyst reports. Set a system prompt that says “you are a senior research analyst at [company]. Always cite specific page numbers and data points.” Then ask it questions like you would a junior analyst.
The key is the system prompt. Without it, Claude gives you generic summaries. With a well-crafted persona and clear instructions, it gives you the kind of analysis that used to require a $200k/yr hire. We use this internally for every enterprise research brief we write.
Get the next issue delivered every Thursday.
I, Ian Kar, certify that the views expressed in this research report accurately reflect my personal views about the subject companies and their securities. I further certify that no part of my compensation was, is, or will be directly or indirectly related to the specific recommendations or views expressed in this report.
This report is published by Arcana Advisors for informational purposes only. It does not constitute investment advice, endorsement, or recommendation of any product or company mentioned herein. The information contained in this report has been obtained from sources believed to be reliable, but its accuracy and completeness are not guaranteed.
© 2026 Arcana Advisors. All rights reserved. This report may not be reproduced, distributed, or transmitted without the prior written consent of Arcana Advisors.