Why AI productivity is not hitting your bottom line (yet)
McKinsey found 80% of workers feel more productive with AI but only 37% see EBIT impact. Here is why individual gains stall at the company level, and what the top 6% do differently.
In this week's headlines roundup I flagged McKinsey's State of AI 2026 survey. The number I cannot stop thinking about is not the spend chart. It is the gap: 80% of people say AI makes them more productive, but only 37% of organizations can point to any enterprise EBIT impact. That ratio barely moved in a year of heavier investment.
If you run a business, you have probably seen the first half of that story in your own office. People draft faster. Research takes less time. Code ships sooner. Then you look at the quarterly numbers and wonder where it went.
This is not an AI failure. It is a translation failure.
McKinsey's framing is precise: individual gains have not yet translated into broad financial impact. That is different from "AI does not work." It means the gains are trapped inside tasks instead of flowing through redesigned workflows, pricing, staffing, and cycle times.
Think of it like giving every salesperson a faster car but never changing the route, the territory map, or the commission plan. Everyone feels quicker. Revenue per mile might not move.
What the top 6% do differently
McKinsey calls about 6% of organizations AI high performers. They attribute meaningful EBIT impact to AI. Patterns that show up again and again:
1. They redesign the workflow, not the interface
High performers are more than three times as likely to say they intend fundamental business transformation through AI. They do not ask "where can we add a copilot?" They ask "if we rebuilt this process from scratch with agents and humans, what would we stop doing entirely?"
2. They deploy across functions, not in one sandbox
A single brilliant pilot in marketing does not move EBIT if operations, finance, and customer service still run on manual handoffs. High performers spread deployment with shared data boundaries and shared logging standards so wins compound instead of conflicting.
3. Senior leadership stays in the loop
This is not an IT science project. The leaders who win treat AI as a operating program with milestones, kill criteria, and a named executive owner. When the CEO stops asking about it, the pilot drifts back to demo mode.
4. They budget for transformation, not tokens alone
High performers are more than twice as likely to commit 15% or more of ICT budget to AI, and more than half expect to increase AI investment by 10% or more. The spend includes process mapping, change management, and production hardening. Not just API keys.
Why build-vs-buy is making the gap worse for some teams
McKinsey also reports 32% of organizations skipped buying software because agentic coding tools made an in-house build look cheap. That can be smart. It can also create a graveyard of unmaintained internal tools that never connect to billing, CRM, or audit trails.
Before you build: name the owner, the shutdown plan, and the metric you will move in 90 days. If you cannot, buy or partner until you can.
Why costs are now part of the ROI conversation
About 20% of respondents said AI operating costs constrained usage. Once agents run continuously on live data, your bill stops looking like a $20/month subscription. Track cost per completed workflow. If saving ten minutes of labor costs $8 in tokens and creates a compliance review, you have not found ROI. You have found a expensive habit.
A 30-day plan to close the gap on one workflow
- Week 1: pick one process with a number attached ( average response time, rework rate, margin on a SKU ). Baseline it honestly.
- Week 2: map the current steps. Circle handoffs where humans wait on other humans. That is where AI usually stalls.
- Week 3: redesign one segment end-to-end. Fewer steps, not faster steps. Add logging and a human approval gate on external actions.
- Week 4: measure the same number. Keep, fix, or kill. Do not add scope until the metric moves.
Boring repetition beats another strategy deck. The companies that will look smart in 2027 are not the ones with the flashiest demo today. They are the ones who turned individual speed into fewer steps, lower rework, and clearer ownership.
How Swift Media thinks about this
We run agents in production with scope fences, dispatch logs, and human dispatch for a reason. AI that touches real customers on real servers needs rules, not vibes. If you want help picking one workflow and designing it so productivity actually shows up on the P&L, talk with us. We will keep it practical.
Matt Potter · Swift Media