66% got results from AI. 48% scaled them. Fund the middle

A team redesigning a workflow on a wall chart in an office

A pilot that works is the easy part of AI in a business. The hard part starts the week after, when the workflow has to change for everyone, the review work lands on a named person, and the training has to reach staff who never asked for a new tool. Most firms fund the pilot and starve that second stage. The Frontier Firms research puts the gap in plain figures. In a survey of more than 5,200 senior leaders across 28 industries, 66% say AI has delivered substantial or transformative results, and 48% are scaling that value across multiple business units. The distance between those 2 numbers is the deployment gap, and it is where a Canadian firm’s 2027 AI budget will be won or lost.

The direct answer: The deployment gap is the distance between an AI result in 1 team and the same result running across the firm. It closes with funded middle work: workflow redesign, review time with a named owner, agent governance and a skills plan. Licence money alone does not close it. A firm planning its 2027 spend should fund the middle work as its own line, at the same table as the tool bill, before it buys 1 more seat.

  • The gap, in the research’s own figures
  • The 4 pieces of middle work
  • The funding order for a Canadian firm

The gap, in the research’s own figures

Start with the honest limits of the numbers. The research is IDC work commissioned by Microsoft, the returns are reported by the leaders surveyed, and the study itself states that it shows association rather than cause. Read it as a map of where firms stall, and it earns its place. Agents show the stall clearly. 36% of organizations are still piloting agents, months into the loudest agent year on record. Piloting is a fine place to visit. It is a poor place to hold a budget.

Advertisement

The top of the scale shows what the other side looks like. Frontier Firms, the top 10% on the maturity scale, report an average 4x return on AI, against 2.7x for slow adopters in the bottom 30%. 39% of Frontier Firms report returns of 5x or more, against 4% of slow adopters. Hold those figures at arm’s length, because self-reported returns deserve it. Then look at the behaviour gaps underneath them, because those are checkable in any office. 93% of Frontier Firms are highly confident they can measure AI’s value, against 61% of organizations overall. 87% of Frontier Firms run a process to govern agents, against 51% of slow adopters. 80% of Frontier Firms hold an organization-wide skilling strategy, against 3% of slow adopters and 20% of organizations overall. The pattern is consistent. The firms at the top spend on the work around the tool.

Canada starts this race from a measured base. 19.2% of Canadian firms use AI. Read the definition before you cite it set out the country’s own count on this site: under 1 firm in 5 used AI to produce goods or deliver services in the year to Q2 2026. A Canadian owner reading vendor returns from global surveys should keep that base in view. The question for a Canadian firm is narrower and more useful. Of the AI work already running inside the building, how much of it has crossed from 1 team into the way the whole firm works?

The 4 pieces of middle work

The first piece is workflow redesign. An agent dropped into an unchanged process speeds up the old steps and keeps the old bottlenecks. The research makes the same point in its own language: agents create the most value when work is redesigned around them. Redesign is meetings, drafts, arguments and a new standard procedure. It bills in staff hours, and it never appears on a vendor quote.

The second piece is review time. Every scaled AI workflow produces output that a person has to check, and that checking load grows with the rollout. A firm that funds generation and skips review builds a queue of unchecked work and a quiet quality problem. Review time belongs in the plan as hours with a name attached. The 5 jobs you keep putting off. Hand 1 to AI this week showed the small end of this discipline: 1 finished workflow, checked end to end, teaches more than a shelf of pilots.

The third piece is governance that a working person can run. 87% of Frontier Firms govern their agents under a set process, and the gap to slow adopters is 36 points. Governance at office scale is plain work. Give your AI agent an 8-line job description before access gives an agent its access, its limits and its off switch in writing. Your agent’s memory needs an editor, not a bigger limit gives its memory a named owner. Neither costs a licence dollar. Both cost management hours, and both decide whether a scaled agent stays safe at month 6.

The fourth piece is skills. This is the widest gap in the research, and the cheapest to start closing. 67% of organizations say their workforce reports high trust in AI, and 20% hold a comprehensive skilling strategy. Trust without a plan curdles. Staff improvise their own methods, the careful people fall behind the reckless ones, and the firm ends up with 12 private workflows and no shared standard. A skills plan at a small firm is a short list of named skills tied to real tasks, taught in working hours, with a person accountable for each one.

The funding order for a Canadian firm

71% of organizations in the research expect AI budgets to rise 10% or more in the next 12 months. Rising spend is the default, so the order of the spend is the decision. One more figure from the same research sharpens the point. 90% of Frontier Firms exceeded their planned AI spend. Read plainly, the firms at the top did not hold their first budget. They spent past it, on purpose, as the work showed them what scale actually costs. A Canadian owner should hear that as permission and as a warning. Permission, because a 2027 AI line that grows past its first draft is normal at the top of the scale. A warning, because the overrun at a Frontier Firm buys redesign, governance and skills, while the overrun at a stalled firm usually buys more seats for a pilot that never crossed the building. The difference is not the size of the spend. The difference is what the spend is for. Fund the middle in this order. First, pick the 1 workflow that already proved itself in a pilot and pay for its redesign across the firm. Second, budget the review hours that workflow will create, and name the reviewer. Third, write the governance lines, access, limits, memory owner, before the rollout widens. Fourth, attach the skills plan for the staff the workflow touches. Buy new tools last, with whatever the middle leaves behind.

The recap fits on 1 line. Results are common, scale is rare, and the distance between them is a set of ordinary jobs that nobody budgeted.

Do this week: take your strongest pilot, write the 4 middle costs beside its licence cost on 1 page, redesign, review, governance and skills, and bring that page to the 2027 budget meeting. A pilot with its middle funded is a plan. A pilot without it is a demo that repeats every quarter at full price.

Total
0
Shares
Prev
Fable leaves your plan tonight. Credits start at $57.03 CAD
A finance worker reviewing a billing statement at a desk

Fable leaves your plan tonight. Credits start at $57.03 CAD

Tonight, November 16, 2026, at 11:59 pm PT, the Fable launch week ends

Next
Qwen4-Max output costs $6.42 CAD. Route the easy steps, keep client data home
A developer routing requests between servers on a monitor

Qwen4-Max output costs $6.42 CAD. Route the easy steps, keep client data home

Alibaba’s Qwen team released Qwen4-Max to the public API today, November

You May Also Like