What’s new since our first edition: the five data points leaders need.
Is AI paying off yet? For a few companies. Not yet for most. That was our answer in August, and it still is. The short version: fix the work before you buy more tools.
companies say AI has added anything to their profits, about the same as a year ago.
That gap is the headline of the last two months. BCG now says nearly half of companies are getting real value from AI, by their own account. But when the question is profit, only about a third see any, and in a separate survey just 6% of executives saw AI pay for itself within a year. Most said two to four years.
Don’t promise your board a profit number this year. Promise a plan with milestones you can measure, and plan for a payback of two to four years.
BCG’s Applied AI Index puts nearly half of companies in its “future-built” and “scaling” groups, up from 40% a year earlier; this is value as companies report it, not profit.58128 McKinsey’s 2026 survey of 1,719 respondents: 37% attribute any EBIT impact to AI, about 6% attribute 5% or more.118 Deloitte, 1,854 executives in Europe and the Middle East: 6% saw payback within a year.67 Bain, 951 large companies: nearly 40% of those measuring AI savings got 10% or less against targets of 11% to 20%.103 PwC: the top fifth capture 74% of AI-attributed value.95 The clearest returns still go to the sellers, with Nvidia’s data-centre revenue up 117%, while no basket of AI-using companies has beaten the market.82119 The September debate about slowing frontier AI, from Anthropic’s chief executive to the OECD’s risk warning, does not change the arithmetic of tools already in use.124127
more likely to see real value when leaders redesigned the work, instead of only adding tools (32% vs 6%).
MIT studied two organizations with the same technology. A medical centre has 141 AI tools in daily use; a law firm has 3. The difference was not the software. The medical centre trained the people building the tools, helped them check the results, and gave them credit for the work. At the law firm, most of the experts gave up.
Pick two processes. Redesign them end to end, including who decides and who checks, before you buy more licences.
McKinsey panel of 608 leaders who use AI (not a representative sample): 32% vs 6% report enterprise value with and without workflow redesign.109 PwC, 1,217 executives: leaders were twice as likely to redesign workflows as to add tools.95 Only 26% of organizations define which work is human-led, AI-assisted or AI-executed.125 The two-organization study, a working paper; counts as reported by MIT Sloan: more than 80% of the law firm’s experts dropped out.100 MIT’s fieldwork at 20+ companies, and a randomized trial at Procter & Gamble, both find one person with AI can cover work that used to need a mixed team.92106
more output per worker at firms using AI, measured across 12,000 companies in Europe and the US.
That is the first large, independent measurement, and it is a real gain. It is also far from the 50% to 80% that case studies like to report, which usually come from a few hand-picked teams describing their own results.
When someone shows you a 50% gain, ask three questions: measured or estimated, by whom, and compared to what?
EIB Working Paper 2026/02, 12,000+ firms, productivity rather than profit; gains concentrate in larger firms that also invested in data, software and training.81 NBER: at US public firms, faster growth in AI-skilled hiring went with about one point a year of extra sales per worker.120 US productivity rose 1.4% in the second quarter with no AI attribution, and Kansas City Fed economists find adoption explains little of the gain.83115 The 50% to 80% figure is a McKinsey case of three front-runner teams at a code-tools company, self-reported.123 CFOs report more value when AI improves decisions (31%) than when it saves time (17%).113
of executives say AI creates extra work nobody recognizes: checking, fixing and explaining its output.
Two in three workers who use AI say checking its output takes real effort. And the bill is hard to see: most AI spending now sits outside the IT budget, spread across departments.
Count the checking time as a cost of every AI tool. One hour a week across 2,000 people at $70 an hour is $6.4 million a year. Then ask finance for one number for total AI spending.
IBM, 1,500 executives: 80% say adoption creates unrecognized work.125 SHRM, 5,875 US workers: 65% report moderate to significant effort reviewing and editing AI output.102 BCG: AI spend reached 3.3% of revenue, with more than 80% outside the IT budget.128 KPMG: only about a quarter of US leaders see AI run costs in real time.108 The $6.4 million is a worked example: 1 hour × 46 weeks × 2,000 people × $70.
of companies leave the head of HR out when they set their AI plans.
That matters because the hard part of AI is the people part: who does what, who checks, who gets credit, and what happens to the time saved. Meanwhile, 72% of companies use little or no AI inside HR, which makes it hard to lead by example.
Ask for a seat where AI money is decided. Then make one HR process your proof that redesign works.
IBM, 1,500 executives and 8,800 employees: 46% do not involve the CHRO in AI strategy; 73% of CHROs struggle to align the C-suite; 72% make limited or no use of AI in HR; only 23% are increasing entry-level hiring; 43% of employees say the blame lands on them when AI goes wrong.125 Only 11% of companies have an enterprise-wide skills architecture (BCG and WFPMA, 7,115 leaders),89 and 88% of HR leaders have not yet seen significant value.66 Middle managers carry the most anxiety, and come in five distinct types.116122 EU rules for AI in hiring and performance decisions apply from December 2, 2027.114
Canadian businesses use AI to make their products or deliver their services.
Most Canadian businesses have no plans to start using AI in the next year.91121 Only 8% use it in a significant way, and the Bank of Canada’s Governor says most firms are still experimenting.117126 The Statistics Canada survey that tracked this has ended, so start tracking your own.
What would change it: solid proof that the typical company, not only the leaders, is making money from AI. We haven’t seen it yet. We will look again in November.
The best argument against us: more companies say they are getting value, and the first independent measurements are positive. If profits follow, we will say so.
The one big 2026 study written for heads of HR. Read: Pages 16 to 23, and 33 to 35.
The medical centre and the law firm. The best story of the quarter. Read: The summary.
The money problem in plain terms. Send it to your CFO. Read: All of it.
The numbers your board will quote. Read: Pages 2 and 16.
The strongest case against our answer. Read: How they define “getting value”.
David has spent over 15 years in management consulting, working with clients across the world on organizational design, total rewards, people analytics, operating model and talent management. In-house he has led teams driving transformation grounded in both technology and people strategy.
That is the perspective these briefs are written from. He has priced the jobs, designed the role architecture, and run the systems that are supposed to turn saved time, innovation, and collaboration into business value.
Agentic HR is David’s personal project, built around a single thesis: AI changes the doing, not the deciding. That makes it an operating-model question, not a technology question.
How this update was made. This page summarizes the October 2026 review of Brief No.001. The 28 sources behind it are coded, dated and listed in the full brief, and every number here links to its row. AI did the searching, sorting and first drafts; David checked the sources that carry weight and made the call on what they add up to.
Corrections. David.herrera@agentichr.ca