Synergised Consulting
Method

AI Spending Is at Record Levels and Layoffs Get Blamed on It. The Return Is Hiding Somewhere Else

Synergised Consulting Ltd

7 min read
Title card reading Record AI spending, flat returns. Where the ROI actually shows up, with an orange rule on a light background, for a First 100 Days Operations post

Last Updated: 2026-11-21

If you want to know where AI's return on investment actually shows up, stop watching headcount and start timing tasks. The 2026 data does not show a labour market being rebuilt by automation: AI-attributed layoffs peaked in May and collapsed by October, total announced job cuts are down 44 percent on last year, and the share of firms reporting any profit impact from AI has not moved in a year. The measurable return, where it exists, shows up as hours that a specific task stops consuming, and that number is only visible to someone who measured the task before touching it.

What the 2026 layoff data actually shows

The AI-layoff story of 2026 is mostly an attribution story, not a displacement story. US employers cited AI in 121,401 announced job cuts through October, about 20 percent of all announced cuts, according to Challenger, Gray & Christmas's October report [1]. That sounds dramatic until the surrounding numbers arrive: total announced cuts were 617,189, down 44 percent from the 1,099,500 announced in the first ten months of 2025, and October's 43,994 was the lowest for that month since 2023 [1]. A year in which overall cutting fell by nearly half is not a year of mass technological displacement.

The timing pattern inside those numbers is the more telling detail. AI peaked as the stated reason in May, at 38,579 cuts, roughly 40 percent of that month's total, then fell to 1,265 by October, the lowest monthly figure of the year and under 3 percent of announcements [1]. By October, market and economic conditions had overtaken AI altogether, cited in 130,531 cuts for the year [1]. Challenger counts announcements and the reasons employers give, not confirmed job losses, and a reason that collapses that quickly looks more like an excuse investors wanted to hear than a cause of anything.

Independent labour economists read it the same way. According to Glassdoor's chief economist Daniel Zhao, quoted by CFO Brew, at the macro level there is no evidence AI is driving layoffs and no increase in layoffs at all: the latest Bureau of Labour Statistics figures show 1.6 million people laid off or discharged in September, slightly below the 1.8 million monthly average before the pandemic [2]. Zhao's reading of the announcements is resource reallocation rather than replacement: a company cuts in one product area to fund AI work in another, and says AI because that is what the market rewards.

Why record spending has not moved the number that matters

The spending side of the ledger is genuinely historic, and the profit side has not budged. According to J.P. Morgan Global Research, the five large US hyperscalers are on track to spend roughly $900 billion on AI capital expenditure in 2026 alone, a figure revised upward from $736 billion earlier in the year [3]. Against that, McKinsey's State of AI survey found 37 percent of respondents attributing at least some EBIT impact to AI, essentially unchanged from a year earlier, and the share of high performers attributing 5 percent or more of EBIT to AI stuck at about 6 percent [4]. McKinsey fielded that survey between May and June 2026 with 1,719 participants across 97 countries, so this is the most complete measurement available of whether the spending is landing.

One comparison inside McKinsey's data explains most of the gap. Nearly three-quarters of the high performers report fundamentally redesigning workflows because of their AI use, while about one-quarter of everyone else does [4]. Most organisations are inserting AI into processes built for people, buying individual productivity without changing the shape of the work. The same survey found 80 percent of respondents reporting AI improved their own individual productivity [4], which is exactly the pattern you would expect if the gains are real but sitting at the level of tasks rather than showing up in the company accounts.

Why a role is the wrong unit to measure the return in

A role is too big and too mixed to show a return, which is why cutting one proves nothing about automation. A typical small-business operations role bundles customer emails, invoice chasing, data entry, scheduling and a long tail of judgement calls that no model should own. Automating the repetitive tasks inside that role can free real hours every week, which is a genuine return expressed as capacity, but it never adds up to removing the person, so the announcement never gets made and the return never gets counted. Conversely, cutting the role removes the hours and the judgement together, and the work quietly returns somewhere else.

Consider what happens in the two cases side by side. A firm that cuts a role to demonstrate AI ROI has optimised for the announcement, not the operation: the tasks still exist, distributed now across people who were already busy, and nothing was measured. A firm that instead times the tasks first, automates the repetitive bundle, and keeps the person for the judgement calls, ends up with the same headcount, hours of freed capacity every week, and a record of what changed. Only the second one has produced evidence that would survive a buyer's or a lender's questions.

Where the return actually shows up: at task level

The unit where AI's return becomes countable is the individual recurring task, not the job, the department or the company. This is where the discovery work we do visibly diverges from the layoff narrative: in discovery workshops, owners rarely arrive with a list of what to automate, and the list gets built by walking through a normal week task by task, at which point the repetitive screen-based bundles, the weekly report nobody enjoys assembling, the copy-paste between two systems, separate themselves almost immediately from the work that genuinely needs a person. Those bundles are where the hours sit, and hours are a thing you can actually measure.

The measurement itself is deliberately unglamorous. Pick one recurring task, count how long it took last week, and write the number down with a date on it. Then, if automation is attempted, run the task again a month later and compare. According to McKinsey's survey, the organisations reporting real financial impact are the ones redesigning the workflow rather than decorating it [4], and you cannot redesign a workflow you have never measured. The dated number, unremarkable as it looks in a notebook or a spreadsheet row, is what turns a vague sense that "AI is helping" into a fact about this business.

What to measure before you automate anything

One number, captured before any tool is bought, is the entire prerequisite: how many hours per week does this specific task consume, and who owns it. The reason owners so often end up with a flat ROI story is that the baseline never existed, so any improvement is unfalsifiable and any spending is unjustifiable. McKinsey's finding that enterprise-level impact has stayed at 37 percent for a year, while 80 percent of individuals report personal productivity gains [4], is precisely what unmeasured, unredesigned adoption produces: everyone busier and faster, nothing provable at the company level.

The working sequence is short enough to fit in a week. Choose the one task that repeats most and irritates most. Record its hours for one or two normal weeks, and note who currently owns the output. Only then decide whether to automate, simplify or leave it alone. When the change is made, measure the same task again and keep both numbers. The owner of the workflow, whoever that named person is, keeps the dated record, because in a year's time that pair of numbers is the only artefact that will tell anyone, including a prospective buyer or a lending manager, what the technology actually did here.

The gap between record AI spending and flat measurable returns is not a paradox, and it will not be closed by better headlines. It closes one task at a time, by owners who measure the hours before they change the work, and keep the evidence of what changed.

Sources

  1. [1] Challenger, Gray & Christmas, October 2026 Job Cuts Report:
  2. [2] CFO Brew, "AI-related layoffs may not be all they appear", 17 November 2026:
  3. [3] J.P. Morgan Global Research, "What unprecedented AI capex means for markets and investors", 19 October 2026:
  4. [4] McKinsey, "The state of AI in 2026: On the road to ROI", 25 August 2026: