One buyer deducts the AI bill before it counts the saving

This reaches a CFO whose FY27 plan carries AI as an investment case rather than a cost-to-achieve one, and it arrives as a comparison problem. WTW has filed a board-approved AI plan with a margin target for 2028, the cost to achieve it, and a savings figure struck after the AI bill. A plan signed by 31 December 2026 that carries none of those lines will be read against one that does. The vendors of AI publish what they charge for it; four of these five put a number on what AI and automation are expected to take out of the cost base, and one of them states that its saving is net of the AI run cost without stating what that cost is.

01The saving is struck after the AI bill, and the bill carries no number

WTW's 8-K of 29 October 2026 puts the plan, the cost and the return in consecutive sentences. “On July 28, 2026, the Board of Directors of WTW approved and authorized management to execute Propel, an Artificial Intelligence ('AI') Acceleration Plan.” The arithmetic follows: “The Company expects to invest approximately $625 million of cash and incur approximately $25 million in non-cash charges to generate approximately $400 million in run-rate savings, delivering a cash-cost-to-achieve ratio of approximately 1.6 times. After reinvesting approximately $50 million to support growth, the Company expects to deliver approximately $350 million in net run-rate savings and approximately 30% adjusted operating margin in 2028.” The plan runs to 31 December 2028, and the margin target carries a year.

The footnote is the finding. Slide 10, headed “Propel Update: Costs to Achieve and Run - Rate Savings”, puts footnote 3 on the run-rate savings line, and footnote 3 reads in full: “Net of incremental AI run costs.” So the approximately $400 million, and the approximately $350 million that survives the reinvestment, are both stated after the incremental AI run costs have been deducted. No amount for those costs appears in the release or the slides, which means the deduction is disclosed and the bill is not. What it does give is pace: from inception to 30 September 2026, $21 million of cost incurred, $14 million of it non-cash, and $18 million of annualised savings.

Put together, the commitment is larger than a year of the benefit: approximately $650 million of cash and non-cash cost against approximately $350 million of annual net saving, about $1.86 of cost for each $1 of saving. The company's own ratio of 1.6 times is measured on cash alone and against the $400 million before reinvestment, and it is the number to quote back at a supplier. The 2028 margin bridge has four legs, and they close: 25.2% in 2025, 200 to 220 basis points from operating leverage and portfolio optimization, 300 to 330 basis points added for Propel savings, about 50 basis points subtracted for the growth reinvestments, approximately 30% in 2028.

02The other four count the saving and disclose no AI run cost

State Street filed on 14 October 2026 and put three quarters of its target in cost. Endnote 8 of the Q3 2026 presentation reads: “Medium-term targets include ~$1B of annual run-rate Transformation benefits expected to be achieved by 2029, reflecting ~$750M of productivity savings, plus ~$250M of revenue.” The same page lists AI-enabled Transformation among the strategic growth initiatives, inside the third of its three numbered items, “Wealth Tech and AI-enabled Transformation”, and the deck does not split productivity by cause, so nothing here says how much of the $750 million is AI. The quarter shows both lines moving at once: total expenses of $2,620 million up 8 per cent, information systems and communications of $594 million up 15 per cent, “largely related to technology and infrastructure investments and volume-related costs”, and compensation and employee benefits up 8 per cent, “partially offset by productivity savings”.

Conduent is the only one of the five that itemises where the money comes from. Its investor day of 30 September 2026 states: “Expect ~$120M of structural cost efficiencies from re-organization, vendor rationalization and AI-enabled operations.” The supporting slide splits the approximately $120 million across five initiatives and three functions, and the split is the useful part: technology is $43.6 million, every dollar of it in shared and corporate functions, which makes it the second largest line after people. A footnote adds that the total “includes $30 million to be realized in 2026”, and the same deck expects free cash flow to turn positive in 2028 as restructuring spend rolls off.

LabCorp, which filed on 10 September 2026, is the cleanest case of intent with no AI number attached. The key takeaways state “AI is embedded across Labcorp to improve productivity and increase operating leverage”. The financial outlook then attaches the arithmetic to a programme: productivity savings through LaunchPad of $125 million to $150 million a year, and margin expansion of 75 to 150 basis points by the end of 2029. AI is named as an enabler of a programme target, so the filing cannot tell a reader what AI contributed to it.

03CNA held the ratio under 30% while investing, and claimed nothing from AI

CNA Financial filed on 2 November 2026 with the most careful sentence in the set: “Despite continued investments in talent, technology and AI capabilities, our expense ratio has remained below 30% in five of the last six quarters.” The ratio was 29.7% for the quarter and 29.9% year to date, against 29.1% and 29.7% a year earlier, which is 0.6 points higher on the quarter and 0.2 points higher on the nine months. The company reads the level as a floor rather than as a result: “We continue to believe an expense ratio around 30% represents a reasonable run-rate for 2026.” Five quarters out of six, a ratio held under 30%, and no claim that the investment produced the hold.

04The party that loses is the supplier whose invoice is the cost line

Two buyers can now defend a lower cost base with a date on it. WTW has a plan running to 31 December 2028, a target of approximately 30% adjusted operating margin in 2028, and 150 basis points of adjusted operating margin expansion in the quarter it began implementing the plan, on organic revenue growth of 4% and adjusted diluted EPS growth of 18%. State Street has 2029, and a target split 3 to 1 toward cost. Both can now answer the question a board asks after a productivity claim: how much, by when, and at what cost.

The first loser is the third-party supplier whose invoice sits inside the buyer's take-out. Conduent puts “vendor rationalization” in the same sentence as AI-enabled operations and puts $43.6 million of an approximately $120 million target in technology, every dollar of it in shared and corporate functions, a line realised through vendor contracts at renewal rather than at headcount. Procurement is a separate initiative in the same table, at $6.2 million of the approximately $120 million. The second loser is the operator still funding AI as an innovation budget with no margin target and no disclosed run cost: at the next plan it will be compared with a peer's approximately 30% margin in 2028 and will have no equivalent line to show.

05What to decide before 31 December 2026

• CFO: put three lines in the FY27 plan that WTW publishes and the other four do not, the cost to achieve with the years it falls in, the run-rate saving before reinvestment, and the run-rate saving after the AI run cost. State the ratio between the cost and the net saving, because WTW states its own at 1.6 times on cash and the total is about 1.86 times. A plan without those lines cannot be compared with a peer's, and it will be compared. By 31 December 2026, when the plan is signed.
• COO: take the saving to the renewal rather than to the budget round. Conduent's technology line is $43.6 million of approximately $120 million, and its own wording puts “vendor rationalization” alongside AI-enabled operations, so that third of the take-out lands only where a contract can be re-papered. By 31 December 2026, so that renewals falling in 2027 are priced against it.
• CIO: forecast the AI run cost in the technology line instead of netting it inside a benefit. State Street's information systems and communications line rose 15% to $594 million in the quarter it reported those medium-term targets, so the run cost and the saving will not be found in the same place. By 31 December 2026, alongside the plan.

06The questions a board should put to management

One decision in this cycle is the board's and not the functions': approval of the FY27 operating plan on a cost-to-achieve basis, with the annual saving stated before and after the AI run cost and the ratio between the cost to achieve and the net saving printed in the plan, signed by 31 December 2026. WTW's board approved Propel on 28 July 2026, and the cost to achieve and the 2028 margin target were published numbers in the same filing. A plan approved without those lines leaves the board holding a productivity claim where a peer has a ratio.

• What is our cost to achieve for AI, in which years does it fall, and what annual saving does it buy?
• Is the saving we report before or after the AI run cost, and who signed the definition?
• Which contracts carry the run cost, what is the committed amount, and when do they renew?
• If a peer reports approximately 30% operating margin in 2028 on a disclosed cost to achieve, what is our equivalent number, and will we publish it?
• How much of our cost take-out is procurement and how much is headcount, and which of the two needs a decision from this board?

07Five filings, and one footnote between them

• Willis Towers Watson, Form 8-K filed 29 October 2026, accession 0001171843-26-006942, Exhibit 99.1 and Exhibit 99.2.
• State Street, Form 8-K filed 14 October 2026, accession 0000093751-26-000592, Q3 2026 earnings presentation.
• Conduent, Form 8-K filed 30 September 2026, accession 0001677703-26-000118, Exhibit 99.1.
• CNA Financial, Form 8-K filed 2 November 2026, accession 0000021175-26-000055, Exhibit 99.4.
• LabCorp, Form 8-K filed 10 September 2026, accession 0000920148-26-000201, Exhibit 99.2.

08What to watch, and when

• 28 February 2027: WTW's fourth quarter 2026 filing. The plan had $21 million of cost incurred and $18 million of annualised savings at 30 September 2026, against approximately $400 million targeted. If cumulative savings are not materially above $18 million at the year end, the plan is behind its own arithmetic and the compression reading fails.
• 30 June 2027: whether a second buyer among these five attaches a run-cost deduction to a savings figure. If none does, this is one company's disclosure practice rather than a change in how buyers report.
• The next two quarters: CNA's expense ratio. Below 29.1%, its level in the third quarter of 2025, while it keeps investing, would mean a benefit has arrived without a number being claimed, and the counter-case here fails.