The briefing

November 17, 2026

A five-minute read for the people putting AI to work in support, sales, finance and operations: one story explained, two more in brief, what else happened, one number, one field note. Written by Wren, every source linked.

GXO, which calls itself the world's largest contract logistics provider, expects more profit from matching paid hours to work done than from AI and robots combined. It told investors on Monday that its warehouses track labour mostly in Excel. Today is about the plain number that sits beside the AI one: LegalZoom cut 13% of its staff and lowered its profit forecast three months later, and KPMG pulled a report after 40 of its 45 citations failed a check.

By the numbers

  • $50M to $70M a year By 2030, about 0.4% to 0.5% of GXO's 2025 revenue of $13.2 billion; its own target, not a result.
  • $70M to $90M a year From software that matches paid hours to work done; GXO says labour is about 60% of warehousing cost.
  • $10M lower Lowered three months after a 13% staff cut that cost $6.0 million; its filings point to dearer customer acquisition.
  • 5 of 45 GPTZero's count, per The Register; the report was public for about eight months before KPMG removed it.
Measure labour first

GXO expects more profit from tracking warehouse labour than from AI and robots

GXO Logistics, which runs more than 1,000 warehouse facilities with about 150,000 people, told investors on 16 November 2026 that automation, warehouse AI and retrofit robots should add $50 million to $70 million a year to its profit by 2030. It expects more, $70 million to $90 million, from a labour management system: software that matches the hours it pays for to the work completed. Both figures are targets in the presentation GXO filed with the US Securities and Exchange Commission that day.

Why it matters: GXO's own slide on labour reads "No standardized system" and "Predominantly Excel-based". The presentation says labour is about 60% of warehousing cost and that labour efficiency should be 5% to 15% higher at sites with the system, against about $5 billion of labour spend. The AI and robots line comes to about 0.4% to 0.5% of 2025 revenue ($50 million and $70 million ÷ $13.2 billion).

The catch: Every figure is a target for 2027 to 2030, not a result. GXO lists six warehouse AI modules, among them placing stock to cut the distance pickers walk, and gives no saving for any one of them. The figures it does give are for hardware: $120 to $250 saved per container by machines that unload trailers, and 15% to 35% higher productivity from collaborative robots, with no baseline or site count. The presentation does not say who reviews the labour system's flags on "paid time with no activity behind it".

What to do: Ask your site managers this week whether paid hours are matched to tasks completed on every shift, and in what tool. If the answer is a spreadsheet, get a price for fixing that before you price any AI module. If a vendor is quoting dock or picking robots, ask for the saving per container and the installed price, then work out the payback: GXO's test is under two years "on every deployment". If you outsource warehousing, ask at the next review which of these tools run on your account and how the savings are shared.

Looking ahead: GXO plans to have the labour system at about 80 sites in 2027 and more than 300 by 2030.

Our coverage →GXO 2026 Investor Day presentation (Form 8-K, Exhibit 99.2, filed with the SEC 16 Nov 2026) ↗GXO Announces 2030 Financial Targets at Investor Day (Form 8-K, Exhibit 99.1, filed with the SEC 16 Nov 2026) ↗

The cut, then the forecast

LegalZoom cut 13% of staff to go "AI-native", then lowered its profit forecast by $10 million

LegalZoom, which sells company formation and legal services to small businesses, committed in August 2026 to cut about 13% of its workforce as part of becoming "a more agile, AI-native company", its quarterly report says. The cut cost $6.0 million in the third quarter, mostly severance. On 4 November it lowered its forecast of 2026 adjusted profit to $180 million to $185 million, from $190 million to $195 million.

Why it matters: Pay did fall. Technology and development payroll was $7.8 million lower than in the same quarter of 2025, though it had been falling before August. Finding customers got dearer faster. The filing says Google's changes, including "the prioritization of AI-generated summaries", reduced unpaid search traffic, and customer acquisition spending rose to $51.1 million from $44.2 million. LegalZoom says more than half of the forecast change is spending on new channels, "including partnerships, affiliates and AI channels".

The catch: The filings do not say which roles went, what work AI took over, what the AI tools cost or the yearly saving expected.

What to do: Before you approve a headcount cut justified by AI, ask marketing for three numbers: the share of new customers arriving from unpaid search, how that moved in 12 months, and the cost per customer of replacing it with paid channels. Put payroll saved and acquisition cost on one page for your CFO.

Our coverage →LegalZoom quarterly report for the period ended 30 September 2026 (Form 10-Q, filed with the SEC 4 Nov 2026) ↗LegalZoom third-quarter 2026 earnings release (Form 8-K, Exhibit 99.1, 4 Nov 2026) ↗LegalZoom second-quarter 2026 earnings release (Form 8-K, Exhibit 99.1, 5 Aug 2026) ↗

5 of 45 citations

KPMG pulled a report on AI agents after a review found 5 of its 45 citations matched their sources

KPMG removed a report on AI agents in June 2026 after GPTZero, a company that sells AI-detection software, said only five of its 45 citations correctly pointed to the source they named, The Register reported on 12 June. The report had been public since October 2025, about eight months. GPTZero alleges roughly half its factual claims were false, unsupported or attributed to the wrong source, including case studies about UBS, Swiss Federal Railways and Transport for London.

Why it matters: KPMG had the policy. Its statement to The Register says it expects staff to follow guidelines that include "human oversight to validate content and verify independent sources". Fortune reported on 22 September that KPMG's marketing department now vets such reports with an in-house checker called Hawk.

The catch: KPMG has not said that an AI tool wrote the report, who approved it or what Hawk misses. GPTZero sells the kind of checking it performed.

What to do: For each report, white paper or customer case study, name one person who opens every citation and confirms the page says what your text claims. Record the name and date with the sign-off, and ask for that record on the last three pieces your team published.

Our coverage →The Register: KPMG's AI report becomes an accidental demo of AI hallucinations (12 Jun 2026) ↗Fortune: Exclusive: KPMG reorgs AI division to create new incubator (22 Sep 2026) ↗

What else

  1. 01Apollo Global Management plans to grow headcount 3% to 5% a year instead of its typical 12% to 15%, with AI the "primary driver", co-president Scott Kleinman said, according to Business Insider on 16 November; he also said Apollo has not reduced its hiring yet. Our coverage →
  2. 02Omnisend, a 246-person marketing software company, gave more than half its staff raises of 2% to 4% in the first quarter of a scheme that pays for AI work with a measurable business result, Fortune reported on 9 June 2026; its finance team earned one for automating up to 85% of an invoice process. Our coverage →
  3. 03Gartner's survey of 1,303 senior functional leaders found that 27% of finance functions reported an AI-related cut in staff through April 2026 and 6% an increase, CFO.com reports; of 5 hours a week saved with generative AI, 1.7 go to work that improves results. Source ↗
  4. 04Uber told Oregon's attorney general it notified 149,346 people that names, addresses, birth dates and driver's licence numbers were taken in a breach at Fireworks AI, a supplier of AI models, TechCrunch reports; Uber said "no alerts were triggered" because the attackers used Fireworks' own credentials. Source ↗

Stat of the day

6% to 11%

That is the share of the work AI agents could do that is complex enough to typically need a vendor's newest and most expensive model, in each of 19 industries, according to a Wharton School and Accenture report covered by Fortune on 16 November 2026. The researchers scored 11,339 tasks. These are scores on a framework, not measurements of live deployments, and Accenture sells AI consulting. Ask each vendor of an AI agent you buy which model it runs on and whether a cheaper one would lower your price. Fortune: Most work doesn't need the Ferrari of AI, Wharton and Accenture find (16 Nov 2026) ↗

Field note

Damin Mutti, founder of the bookkeeping software company ZeraBooks, writes in CPA Practice Advisor that data pulled from PDF bank statements, by software or by a clerk, should pass two checks in Excel before it reaches the books. First, the opening balance plus every transaction must equal the closing balance the bank printed. Second, each month's opening balance must equal the prior month's close. A gap of twice one transaction usually means a flipped sign, he writes, such as a refund booked as a withdrawal. Mutti sells software that runs these checks.

One question

Shift by shift, do you know how many of the hours you pay for match work completed, and is that record in a system or a spreadsheet?

Reply to Wren → One line is plenty. Answers shape what the newsroom covers next.

— Wren

This issue went to subscribers at 7:00 ET. Get tomorrow’s by email, free →