AI Hype vs Reality – By the Numbers
Here’s a number that should make you uncomfortable: 55% of knowledge workers use AI weekly. That sounds like transformation underway. It isn’t.
According to Section AI’s 2026 AI Proficiency Report, only 15% of those use cases are generating any meaningful business value. The rest? People prompting their way through tasks that look like productivity but don’t move the needle. And if you zoom out to the advanced, organization-level automations the headlines keep promising — only 2% of workers are actually there.
This is the adoption-proficiency gap. And it’s a pattern I’ve seen before: high activity numbers, modest real-world results, and a leadership class convinced the rollout is going well. The hype and the reality are running on separate tracks — and the distance between them tells you exactly where to focus.
The Four Levels of AI Proficiency
Think of AI proficiency less like a pass/fail test and more like a spectrum with four distinct tiers. Here’s where the workforce actually sits:
Level 1 — Non-User (26% of workers)
Not using AI for work at all. This is smaller than people assume, but it’s still a quarter of the workforce.
Level 2 — Experimenter (~55% of workers)
This is where most people live. You’re prompting ChatGPT the way you used to Google things. You’ve replaced search, occasionally drafted an email, maybe summarized a document. The tasks are real. The time savings are real. But they’re small, discrete, and personal — they don’t change how work gets done at any meaningful scale.
Level 3 — Practitioner (~15% of workers)
This is where actual business value starts. Practitioners have found specific, repeatable AI use cases tied to their core function. They’re saving hours per week, not minutes. They’re producing outputs — analysis, communications, workflows — faster and at a quality that’s hard to replicate manually. This isn’t AI as a party trick. It’s AI as a work system.
Level 4 — Advanced (~2–3% of workers)
This is the tier the demos live in. Automations that run without human hands. Agents that handle workflows end-to-end. Organizational ROI that shows up on a spreadsheet. Real, but rare.
The uncomfortable truth: most “AI adoption” is Level 2. And Level 2 isn’t nothing — but it’s not what the C-suite is measuring against when they talk about transformation.
The C-Suite vs. IC Problem
Here’s the subplot that makes the gap worse: leadership thinks the gap isn’t there.
C-suite executives are 31% more likely than individual contributors to believe their organization has achieved widespread, successful AI adoption. Meanwhile, their ICs are having a very different experience.
Why the disconnect? Because the access isn’t equal. In most organizations:
- 80% of C-suite have access to AI tools. 32% of ICs do.
- 81% of C-suite receive AI training. 27% of ICs do.
And predictably: 68% of ICs feel overwhelmed or anxious about AI at work. Only 26% of C-suite feel the same way.
So the people who actually do the work — the ones whose daily tasks AI could most directly transform — are the least equipped and least trained to use it. The people making adoption decisions are working from a data set that confirms what they want to believe.
This isn’t unique to AI. I watched the same dynamic play out during the RPA rollout years ago. Leadership sees the demo, declares the initiative launched, and measures success by licenses purchased. The people on the floor figure out — usually the hard way — that the demo and the day job are very different things. The hype cycle runs on a different clock than actual implementation.
Finance Gets the Short End
If you work in finance or accounting, the picture is even more pointed.
CFO Connect’s 2026 State of AI in Finance report found that finance and accounting ranks last among all business departments in AI deployment maturity. Behind engineering. Behind legal. Behind customer success.
The reason isn’t that finance people are resistant to change. It’s structural:
- Fragmented data. Most finance teams are still reconciling across ERPs, spreadsheets, billing systems, and legacy tools. AI can’t automate what it can’t reliably read.
- Lack of process standardization. A foundational requirement for automation of any kind is a standardized, repeatable process. In finance, getting the process right — manually, consistently — is often the harder problem. You can’t reliably automate what you haven’t yet reliably standardized.
- No time to fix the problem. Finance teams are buried in the work that AI could help them escape. The monthly close doesn’t pause for process improvement. The capacity to experiment requires protected time that almost never gets carved out.
- Regulatory and audit requirements. Finance work requires showing your work. When external auditors or regulators arrive, they need to trace how the numbers were calculated. Transparency has to be designed into how AI is used from the start — it can’t be retrofitted.
- Security anxiety. Feeding financial data into public AI models is genuinely risky, and the policy guidance from most organizations is either absent or so restrictive it’s useless in practice.
The result: a split between what CFO Connect calls “Tinkerers” (using ChatGPT for summaries and emails, manual copy/paste, no integration) and “Integrators” (AI connected to ERPs, automating close processes, delivering new analytical insight). Most finance teams are Tinkerers who aspire to be Integrators — with no clear path between the two.
The Self-Diagnostic
Here’s a quick way to figure out where you actually sit.
Answer honestly:
- When you use AI at work, does it change the output — or just the speed of reaching the same output you’d have gotten anyway?
- Do you have three or more AI use cases you could describe to a colleague that would make them want to try the same thing?
- Have you used AI to do something in the past month that you couldn’t have done at all without it — not just faster, but at all?
- Have you built anything (a prompt chain, a template, a workflow) that you or your team uses repeatedly?
Scoring:
- 0–1 yes: You’re a Level 2 Experimenter. You’re using AI. You’re not yet getting value from it.
- 2 yes: You’re on the boundary between Level 2 and Level 3. Getting closer.
- 3–4 yes: You’re at Level 3 or above. The gap is behind you.
What It Takes to Move Up
The research is consistent on what separates Level 2 from Level 3. It’s not smarter prompting. It’s specificity.
Experimenters use AI for generic tasks. Practitioners have identified the 3–5 specific tasks in their actual job where AI produces consistently better or faster results — and they’ve built habits around those tasks.
For finance professionals, that means not asking “how can AI help me?” but asking: “What are the three most time-consuming, repetitive analytical tasks I do in a given month — and could AI handle the drafting, formatting, or first-pass analysis of any of them?”
The Integrators didn’t get there through inspiration. They got there by getting uncomfortable with the specifics of their own process workflows, running experiments, failing a few times, and building something that actually fit their daily work.
That’s the path. It’s not glamorous. But it’s the difference between using AI and getting value from it.
The Section.ai AI Proficiency Report (2026) and CFO Connect’s State of AI in Finance (2026) were used as primary sources for the data in this post.