At a glance

  • Most leadership teams now budget for AI licenses, usage credits, and a pilot or two.
  • Few budget for redesigning the work around the people who will use them.
  • In McKinsey’s 2026 survey, 8 in 10 said AI raised their productivity; only 37 percent reported any EBIT impact.
  • Fund the operating model: redesigned workflows, review rules, named owners, and incentives that reward changing the work.

AI’s return now depends on the company around the technology and its people.

Microsoft’s 2026 Work Trend Index, a survey of 20,000 AI users in 10 countries, traces about 67 percent of AI’s impact to organizational factors and 32 percent to individual ones. Microsoft’s own summary is short: “People are ready. Systems aren’t.”

It’s a vendor survey, so read it as a signal. Independent research points the same way.

Individual gains need a plan to reach company results

People feel the difference before the P&L does. In McKinsey’s 2026 State of AI survey of 1,719 respondents, 8 in 10 said AI raised their own productivity. Only 37 percent reported any EBIT impact, and about 6 percent qualified as high performers.

The St. Louis Fed puts a number on the gap. In the second quarter of 2026, 39.2 percent of US workers used generative AI for work in the past week, and it saved 2.2 percent of all work hours. The saving is real and small at company scale.

BCG’s June 2026 AI at Work survey shows where the time goes. Among regular AI users in non-manager roles, 42 percent save a full workday or more each week. But 66 percent of those saving time get little or no guidance on what to do with it.

Time saved without a plan disappears into the week.

Four pieces of the operating model decide the return

The firms that see results changed the work along with the tools. Four pieces recur across these studies, and each needs someone’s time and budget.

Redesigned workflows. BCG found that organizations reshaping or inventing processes with AI rose from 22 percent in 2025 to 42 percent in 2026, and they create more value. Its headline finding: strategy matters more than tools.

Defined review. McKinsey’s high performers were far likelier to have set rules for when a person validates AI output: 65 percent, against 23 percent of the rest.

Incentives that reward change. In Microsoft’s survey, only 13 percent of people say they’re rewarded for reinventing their work when the results miss. Nobody redesigns a job they’ll be punished for experimenting with.

Named owners and governance for every agent. Microsoft’s Agent 365, generally available since May 1, 2026, registers and governs agents, and Entra Agent ID gives each one an identity.

The tools exist. Someone in the business still has to own each agent, its limits, and its results.

Training is the investment people ask for first

McKinsey’s January 2025 superagency research found employees readier for AI than their leaders assumed. Employees ranked training as the most important factor for adoption, at 48 percent. Only 1 percent of leaders called their company mature in its use of AI.

In a service business, training means a dispatcher learning when to override the agent’s draft schedule, and a parts counter pro learning which quotes to check line by line.

The unbudgeted costs show up after the pilot

Three costs beyond the license catch leadership teams by surprise.

Usage. Every Dynamics 365 agent consumes Copilot Credits. Dynamics 365 Premium licenses include 1,000 credits per user per month, pooled across the tenant; beyond that, agents bill by usage. Someone has to own that consumption like any other operating cost.

Continuous change. Microsoft no longer publishes twice-yearly release plans. New Dynamics 365 capabilities now arrive continuously on the AI at Work roadmap, and the Release Planner retires on November 15, 2026. Your system changes every month; someone has to decide what to switch on.

Redirected time. When an agent drafts the quote or the schedule, the hours it saves need a destination: more customer calls, faster billing, coaching new technicians. Decide that before go-live, or it won’t happen.

People lead the redesign, and roles change with it

McKinsey’s work on the agentic organization expects small teams of 2 to 5 people to supervise 50 to 100 agents, and about 75 percent of roles to need reshaping. Those figures are a projection. They still give a fair sense of the size of the job.

The idea is older than AI. In 1962, Douglas Engelbart argued that augmenting people means improving the whole system around them: tools, methods, language, and training together. Buying the tool is one of those four.

We run our own firm this way. Our Helios 1 map gave each of 150 opportunities across 34 functions a named owner and marked the rows that stay human, before we built anything.

When the operating model isn’t the bottleneck

Sometimes the problem is underneath. If the system of record is unstable, with data your teams don’t trust or integrations that fail, fix that first. Agents amplify whatever they’re built on; our piece on stabilizing before adding agents covers the order.

Read the numbers with their limits. The Microsoft, McKinsey, and BCG figures are surveys, self-reported and correlational.

BCG’s “frontline” figures describe non-manager office staff, not field crews. For a small team, the operating model may be one owner and a written review rule.

What leaders should do in this year’s budget

Put the operating model in the AI budget. Give workflow redesign, review, and training their own line, next to licenses and credits.

Redesign one workflow end to end before scaling. Pick one handoff, such as the call to dispatch or the finished job to the invoice, and change the whole thing.

Write the review rule for every agent. Who checks its output, how often, and what they do when it’s wrong.

Decide where the saved time goes. Name the work it moves to, and measure it.

Reward the people who change the work. Including when an honest test misses.

The redesign starts with a map of which work belongs to people, rules, and agents: mapping the jagged frontier. The people it matters most for may be your newest hires. See how Helios 2 builds the map, and our view on the human side of AI.

How we know this

This point of view comes from Ludia’s delivery work and its own Helios 1 map.

  1. Organizational vs. individual factors, rewards for reinvention: Microsoft, Work Trend Index 2026 (May 2026). 20,000 AI users in 10 countries; a vendor survey.
  2. Productivity, EBIT impact, high performers, human validation: McKinsey, The State of AI (2026). 1,719 respondents, May to June 2026; self-reported.
  3. Usage and hours saved: Federal Reserve Bank of St. Louis, Bick, Blandin and Deming, “Does generative AI save time at work?” (Aug 2026). Nationally representative US adults; self-reported.
  4. Time saved, guidance, process redesign: BCG, AI at Work: Why Strategy Matters More Than Tools (June 2026). 11,749 workers in 14 markets.
  5. Training and leader maturity: McKinsey, “Superagency in the workplace” (Jan 2025).
  6. Agent teams and role reshaping: McKinsey, “The agentic organization” (Sept 2025). Projections.
  7. Agent governance: Microsoft, Agent 365 general availability (May 2026); Microsoft Entra Agent ID.
  8. Copilot Credits in Dynamics 365: Microsoft, Dynamics 365 Licensing Guide (Sept 2026).
  9. Continuous roadmap: Microsoft, “One always-on roadmap” (Aug 2026).
  10. Augmentation as a system: Douglas Engelbart, “Augmenting Human Intellect: A Conceptual Framework”, Stanford Research Institute (1962).