AI ROI
Every Monday. What European boards and global investors can verify in reported numbers. Subscribe by email
A postal group will spend €0.7bn to gain €0.7bn a year and names AI among the savings, a car retailer sold 7.6% more cars in a market up 0.6% and credits AI, and an internet group’s AI bill took 20.9% off its profit
Companies do not publish everything their boards see. This issue sets out what an outside reader can verify.
Reported, not published, and what to ask
| Company | Reported | Not published | Board test |
|---|---|---|---|
| Poste Italiane Investment | €0.7bn one-off cost of integrating Telecom Italia; at least €0.7bn a year of synergies; AI named in two of six groups of cost savings | How many people are redeployed, the AI share of the saving and the cost of the AI | Total cost of the AI, and the redesign of structures, processes and roles behind the saving |
| Motorpoint Return | Retail cars sold up 7.6% in a market up 0.6%; margin per car up 19.1% to £1,266; profit before tax up 80.6% to £6.5m; results headed “driven by AI and digital capabilities” | The sales its AI agents added, the part of the margin gain due to AI pricing, and the cost of the technology | The return on AI as a number, set against what the AI cost |
| NAVER Risk | Revenue up 10.7%, operating profit down 20.9%, attributed to investment “to secure AI competitiveness” | The amount spent, and how it splits between computing capacity, content and data | How much of a fall in profit is AI spending, and what revenue it is meant to bring |
Poste Italiane: integrating Telecom Italia will cost €0.7bn, and AI is named in two of the six ways the savings will come
Reported. Poste Italiane will spend about one year of the integration’s benefits to obtain them. On Thursday 12 November it gave investors the bill for integrating Telecom Italia (TIM), the Italian telecoms group in which it held 85.8% of the share capital when its tender offer closed in September.
| Cost | “€0.7bn pre-tax one-off integration costs to be expensed mainly in 2026-27” |
| Benefit | At least €0.7bn a year once complete: at least €0.5bn of cost savings by 2028, and more than €0.2bn of operating profit from additional revenue by 2029 |
The AI link. The plan lists six groups of savings behind the €0.5bn of cost synergies, and AI is named in two of them. Under technology, Poste plans “One IT & Operations platform powered by Poste Italiane AI and hybrid cloud infrastructure”. Under human resources, it lists “AI/integration-enabled people re-deployment across Poste+TIM”, and says it is still “Defining functions’ target size, AI-enabled efficiencies, FTE up-skilling and re-skilling needs”.
Not published. Poste gives no figure for the people to be redeployed across Poste and TIM. It does not say how much of the €0.5bn depends on AI, or what the AI will cost. The same slide repeats four targets that Poste announced on 24 July for its own business, among them the redeployment of “Up to 20%” of overhead staff within five years. Poste presents those as separate from the TIM synergies and does not tie them to AI.
The groups of cost savings in Poste’s plan for TIM that name AI: the shared technology platform and the redeployment of people. Poste publishes no figure for either.
“AI/integration-enabled people re-deployment across Poste+TIM”
Poste Italiane, nine-month results and TIM integration update, presentation slide 13, 12 November
Board test. Look for the total cost of the AI and the redesign of processes, structures and roles, written down beside any saving described as “AI-enabled”. Poste names AI as an enabler and is still defining the size of the functions.
Read Poste Italiane’s integration update →
See also: Nerdy [US, online tutoring] states the kind of number Poste has not published. Its shareholder letter of 9 November says its consumer sales, support, engineering and product teams “are leaner by more than 300 people, as AI now handles work that previously required additional headcount.” It also reports the running cost of the AI, which few companies do: “our annualized AI cost run-rate declined approximately 60% from its peak in June and July to September”. Nerdy gives the fall and does not give the cost itself.
Motorpoint: “Record H1 volumes driven by AI and digital capabilities”
Reported. Motorpoint sold more cars at a higher margin in a nearly flat market, and it credits its technology, AI included. The UK used-car retailer published half-year results on Wednesday 11 November under the heading above. In the six months to 30 September it sold 35,430 retail cars, 7.6% more than a year earlier, in a used car market that “grew by 0.6%”. The margin on each car, which Motorpoint calls metal margin, rose 19.1% to a record £1,266. Revenue rose 6.0% to £686.7m and profit before tax rose 80.6% to £6.5m.
The AI link. Motorpoint names AI in two places. On margin, it says the results “reflect embedded technology innovation throughout the Motorpoint vehicle lifecycle, including increased effectiveness of dynamic AI pricing models”. On volume, it says “AI agents are also creating incremental sales by identifying and re-engaging customers with previously closed or inactive enquiries”. John Walden, the chair, writes that the record margins and volumes are “attributable in large part to continued improvement in the Group’s strategic technology capabilities”.
Not published. Motorpoint gives no number for the sales its AI agents added. It does not say how much of the £203 rise in margin per car comes from AI pricing. It gives no separate cost for the technology: capital expenditure of £5.0m covers a new store in Leeds, a freehold property bought for £2.0m and “continued investment in technology, data and AI capabilities”. Motorpoint says it has “introduced new technology KPIs to track the adoption of these capabilities and will report on our progress against them going forward.”
The counter-evidence. The same announcement names causes other than AI. Motorpoint opened a store in Leeds in July, and it lowered the interest rate on its car finance to 10.9% from 12.9% “to help customers with affordability”. It does not say how much of the growth came from each.
The rise in Motorpoint’s margin on each car, to a record £1,266. The company names its AI pricing models among the reasons and does not say how much of the rise they account for.
“Our integrated platform is improving profitability per vehicle, delivering a record metal margin in the half and reducing customer acquisition cost.”
Mark Carpenter, Chief Executive Officer, Motorpoint, interim results announcement, 11 November
Board test. Look for the return on AI as a number set against its cost: the sales the AI agents added, the margin the pricing models added, and what the technology cost to build and run. Motorpoint credits AI in its heading and has promised the measures.
Read Motorpoint’s half-year results →
See also: Siemens [Germany, industrial software and automation] shows the same gap from the side of a company that sells AI. Its preliminary full-year results of 12 November report “revenue from digital business up 16 percent to €10.9 billion” under the heading “driven by Industrial AI”. On the results call Roland Busch, the chief executive, said “AI revenue was already trending to around 5% of DI software”, a business with €7.2bn of revenue, and worked it out himself: “then you end up with EUR 350 million.” The remarks are in the call transcript published by Investing.com. Siemens does not report AI revenue as a line, or its gross profit or cost.
NAVER: revenue up 10.7%, operating profit down 20.9%, and AI spending given as the reason
Reported. NAVER’s revenue grew and its profit fell, and the company itself gives spending on AI as the reason. South Korea’s leading search engine, which also runs shopping, payments and content businesses, reported on Thursday 12 November that third-quarter revenue “rose 10.7 percent year over year to KRW 3.4724 trillion”, while operating profit “came to KRW 451.3 billion, down 20.9 percent from a year earlier”. Its own summary reads “operating profit declines due to investment in AI competitiveness”, and the text says the fall came “on increased investment in infrastructure, content, offline data, and other areas to secure AI competitiveness.”
Not published. NAVER names infrastructure, content, offline data and “other areas”, and gives an amount for none of them. It does not say how much went on computing capacity, how much on content and how much on data, so a reader cannot say what part of the fall is AI. It gives no figure for AI revenue.
The counter-evidence. The spending is already bringing in some revenue, and NAVER names two places. The first is advertising. Revenue from its search, shopping and services platform rose 7.3% to KRW 1.95 trillion, and NAVER credits part of that to selling more advertising with the help of AI (“advertising revenue growth from expanded AI monetization efforts”). The second is a new business. NAVER is building an AI computing venture with Brookfield, the Canadian investor, and NVIDIA, the chip maker, which it calls its “AI factory”, and says it has signed contracts “worth approximately KRW 2 trillion”. That is about 17% of last year’s revenue, but none of it has been earned yet, because NAVER expects the revenue to start “in the first half of next year”.
NAVER’s operating margin this quarter and a year earlier, worked out from its figures. The company attributes the fall to spending for AI and does not say how much it spent.
“on increased investment in infrastructure, content, offline data, and other areas to secure AI competitiveness.”
NAVER, third-quarter 2026 results release, 12 November, on why operating profit fell
Board test. Look for the amount behind any single line such as “investment for AI competitiveness”, and for the revenue it is expected to bring. NAVER names the spending and the contracts it expects to follow, and gives no amount for the spending.
See also: Cisco [US, networking equipment] showed the same pattern from the supplier’s side on 12 November. Revenue rose 24% to a record $18.5bn, while the adjusted gross margin on products fell 350 basis points to 63.7%, which its chief financial officer put down to “higher hardware mix and elevated memory costs”. Operating costs rose 2%, so adjusted operating income still rose 37%. Cisco does not publish the margin on the AI sales themselves.
- Rakuten puts a figure on AI in customer service. Its third-quarter presentation of 13 November has a slide titled “Raptor: Generative AI-based Customer Service is Reducing Costs”, showing “3.5B JPY” of cumulative cost reduction for the first three quarters of 2026, with the tool “Live on 9 Services”. The slide gives no baseline and no cost for the AI. Read the presentation.
- Vodafone lists AI results and keeps them out of its savings figure. Its half-year presentation shows “85% of total contacts resolved by our AI chatbot” and “>99% of tenders with AI”. The “>€200 million gross savings in H1 FY27” in the same presentation covers all cost measures in Europe and shared operations, and Vodafone does not say how much of it is AI. Read the presentation.
- The UK reporting regulator counted AI in 100 annual reports. The Financial Reporting Council’s annual review of 12 November found that 95 mention AI, 68 place it in the risk framework, and “10 companies discussed their use of AI in their operations without also explicitly discussing AI governance and risk management.” Read the review.
- On total cost: Where a plan describes a saving as “AI-enabled”, what is the total cost of the AI: licences, the model and computing fees charged for every use (billed per token), integration, data work, training and the people who supervise it? Which processes are redesigned end to end, which structures and layers go, and what work will the redeployed people do?
- On attribution: Where our results credit AI for higher sales or margin, how much of the gain is due to the AI, and how was that measured? What did the AI cost to build and run, and which measures will we report to show it?
- On AI spending: How much of the change in our operating profit this year is AI spending, split by computing capacity, content and data, and people? What revenue is it expected to bring, and by when?
The evidence file. Everything in this issue comes from what the companies published in the week of 9 to 15 November: results, filings, announcements and their results calls, plus one regulator’s review. Where the issue does a sum of its own, it is labelled “The maths”.
A correction. Issue 8, as first sent, said NAB wrote off A$1.35bn of software “because AI can rebuild it”. That was wrong and was corrected the same day: the write-down results from a change in NAB’s software capitalisation policy and is not an AI cost or an AI saving.
What we are building. Accelr8™ helps a business recover enterprise value and reduce risk. It checks what the business has committed, in contracts, plans and entitlements, against what happens in invoices, results and use, across six Value Pools, from Revenue Quality to Margin Integrity to Capital Velocity, and shows the board where value is lost, what has been recovered, and the evidence for both. All three questions above are the kind it answers from a company’s own records. Patent pending; a founding pilot is open.
The glossary. AI, in board terms. AI terms explained for the people who decide, with what to ask management for under each, including run-rate, used in this issue.
Let’s talk. Board discussions about AI are moving from vendor demos to where the money goes and what it returns. If what you see on your own board or in your portfolio matches or contradicts this issue, we would like to hear it.
The content is checked against the companies’ own filings before publishing, but mistakes can slip through. If you spot one, please let us know and it will be corrected in the next issue. For information only, not investment, legal or tax advice. Published by Agent & Capital Ltd, 71-75 Shelton Street, Covent Garden, London WC2H 9JQ. Subscribe here.
Get AI ROI by email every Monday at 07:00 UK time. One story each on investment, return and risk. Unsubscribe in one click.