A bank named AI agents for 400 job cuts, a grocer counted £251m of savings with AI inside, and an airline refused to let anyone else's AI near its prices.
Investment. DNB, with a cost/income ratio of 38% against an EU average of 58%, will cut about 400 jobs by the end of the year as AI agents take over three use cases: customer data checks, Know Your Customer and coding. One-off charge expected in the Q4 2026 results, full run-rate saving from Q2 2027. Unions push back: "will not accept dismissals in this process".
Return. Tesco saved £251m in H1 2026/27 and is on track for its £500m target this year. Drivers: optimised picking of online orders, "AI-led improvements to in-store replenishment processes", reduced store energy consumption and lower inventory loss. The savings were "more than offsetting further investments in the customer offer and operating cost inflation"; adjusted operating profit still rose 6.3%.
Risk. Delta's chief executive says AI shopping agents are "trying to commoditize your product and your premiums". At stake, in the words of his chief commercial officer, is "a defining strength of Delta": premium cabins, loyalty and other high-margin streams that bring in 61% of revenue, including more than $9bn this year from American Express.
"We are already seeing considerable gains, and are therefore adapting our organisation to a new reality"
On Tuesday 6 October DNB announced "a reduction in the workforce of about 400 full-time equivalents" in its Technology & Services unit: about 350 in Norway, from a pool of about 1,500 people being assessed, and about 50 at its office in Riga, Latvia. The stock exchange release, which quotes group chief executive Kjerstin Braathen, is specific about the work: "Control of customer data and the work relating to the Know Your Customer (KYC) process, as well as technology development and coding, are areas where the bank is already seeing that AI agents can contribute substantial efficiency gains." In summer 2025 DNB had already cut 100 roles in Technology & Services in Know Your Customer, anti-money laundering and operational security, moving part of the work to the Riga office. This is the third round since autumn 2024, about 1,000 positions in all, in a group of 10,829 full-time equivalents. At a cost/income ratio of 38.0%, DNB is well below the EU average of 58.3% and ahead of Nordea (45.0%) and Handelsbanken (41.5%); among the large Nordic banks only Swedbank (36%) is lower. DNB will publish the restructuring cost and the saving "later this year".
The unions want the arithmetic. "We support neither the selection pool nor the reduction figure," says Lillian Hattrem, Finansforbundet's chief representative at DNB. A second union, Fagforbundet FinTech, says management "chose not to listen to our input and warnings".
"AI can deliver large gains. But how the gains are taken is a choice."
Arne Fredrik Håstein, Deputy Leader, Finansforbundet (Finance Sector Union of Norway), 6 October
When job cuts are attributed to AI, weigh value against total cost. Value is the saving or the capacity freed for growth. Cost is the restructuring charge, the technology, from licences to tokens and running costs, and the people who build and supervise the agents. Risk-adjust for union action, a late saving or an agent getting a check wrong. Keep asking until the numbers are clear.
See also: Fair Isaac restructuring filing, 6 October [Fair Isaac, US, credit scoring]: the company behind the FICO credit score is cutting about 15% of jobs at a one-off cost of $27.0m, and gives "integrating AI-driven product development" as one of four reasons; Tarifrunde private Banken startet ohne Arbeitgeberangebot [AGV Banken, Germany, employers' association of private banks]: the unions want protection from AI job cuts written into the wage deal, the employers refuse to negotiate it, next rounds on 2 and 23 November; C.H. Robinson to Acquire RXO [C.H. Robinson, US, freight logistics]: its "Lean AI operating model" is to take about $300m a year of cost out of RXO, a US truck broker it is buying for $5.8bn
"Delivered a further £251m of savings in the period and on track to deliver £500m Save to Invest target for the full year"
Tesco's interim results on Thursday 8 October credit AI with improvements to in-store replenishment, one of the four Save to Invest drivers. The programme delivered about £535m last year and, since it began in June 2022, "over £2.2bn of savings" against £316m of restructuring cost booked as adjusting items. For customers, chief executive Ken Murphy reports "strong progress on AI-enabled personalisation": personalised "Your Clubcard Prices" now reach "around 2.5 million customers", and an AI meal planning assistant was trialled with about 280,000 colleagues before reaching customers in September.
Where the savings went is specific: about 700 Aldi Price Match products, "more than 10,000 Clubcard Prices" and "over 800 products" launched or improved in H1 2026/27. Group sales rose 1.6% at constant rates to £33.8bn, adjusted operating profit to £1,783m, and the full-year profit range was raised to £3.15bn to £3.30bn, from £3.0bn to £3.3bn in April. Capital expenditure rose £65m to £732m in the same period.
"Our strong performance enables us to keep investing in the customer offer and the capabilities that will drive future growth."
Ken Murphy, Group Chief Executive, Tesco, interim results, 8 October
Tesco’s savings pay for Aldi price matching, Clubcard Prices, new and improved products and operating cost inflation, every year. Put a figure on what AI saves you, and decide as deliberately as Tesco does whether it lands in profit or goes back into the business.
See also: Deutsche Telekom boosts growth, efficiency and quality through the use of AI [Deutsche Telekom, Germany, telecoms]: about €2.5bn of savings targeted by 2030 against 2023 from AI and automation together; Nordic Capital and Anthropic collaborate to accelerate AI adoption across portfolio companies [Nordic Capital, Sweden, private equity]: more than 16,000 hours of work expected to be saved in the first year at its portfolio company Duco, no money figure; Booking Holdings spends nearly $9 billion a year to reach travelers. Its CFO explains how AI is changing trip planning [Booking Holdings, US, online travel]: customer service costs down slightly while bookings grew at a high single-digit rate, no money figure
"We're going to be a little cautious about who we give access to our inventory"
On Delta's results call on Thursday 9 October, chief executive Ed Bastian called AI shopping agents, which search and book on a customer's behalf, "a danger that we need to be mindful of": agents "looking to shop you" will "compete us all against ourselves". He sees room only for "a known agent" working with corporate customers, where Delta had "the highest corporate sales week in our history" in early September.
Only 39% of Delta's revenue now comes from standard economy tickets, the product an agent compares most easily. Premium and loyalty are "each growing nearly 20%", loyalty revenue rose 18% in the quarter, SkyMiles "membership is growing faster than capacity", and Delta is on track for "a fifth consecutive year of 1 million new card acquisitions". Commercial chief Joe Esposito: "Our diversified, high-margin revenue streams remain a defining strength of Delta, enhancing the quality and durability of our earnings."
Delta is not against AI. Delta Concierge, its own agent, is now rolled out to all SkyMiles members. It also prices with AI, and has had to defend it: in September it told US Congressman Frank Pallone, who had raised concerns about "surveillance pricing", that its AI pricing tool is "a decision-support tool" that is not given "personal information about our customers".
"I do think this question of brand will be very important in the agentic economy for the future and brand loyalty, and preference is going to be one of the ways in which the strong will stay strong."
Ed Bastian, Chief Executive Officer, Delta Air Lines, third-quarter earnings call, 9 October
Your customers' AI agents now search, compare and recommend; in business-to-business they draft the tender and weigh the quotes. The risk is a contest you did not choose, judged on criteria you cannot see, in which agents may "compete us all against ourselves". Delta keeps them away from its prices; Zalando meets them with its range and its data. Decide which contest you enter, and who owns that decision.
See also: USA TODAY Co. v. OpenAI, complaint filed 8 October [USA TODAY Co., US, newspapers]: seeks more than $250m and quotes OpenAI's own line that once ChatGPT answers there is "no good reason to click"; Hostelworld trading update, 7 October [Hostelworld, Ireland, online travel]: part of a 2% fall in transactions came from "lower organic web traffic, across the industry, as AI answers more searches"; Fresh redundancies at Future as tech brands hurt by AI Overviews [Future PLC, UK, media]: UK audience of Tom's Guide down 55% since Google added AI answers
"75% of employees use AI on a recurring basis, saving an average of 2.4 hours per week on repetitive, lower-value tasks." Software developers "have improved their productivity by around 28%", voice agents "successfully resolve close to 90% of interactions" in pilots in Peru and Mexico, and the time to register a complaint is down "by nearly 80%". No figure in euros, no cost and no consequence for headcount. In the same week DNB turned its hours into 400 fewer jobs; BBVA has not said what its hours are for, and that decision is where the return is made. BBVA
"Ecosia founder clarifies move away from Mistral – It's about open models, not China". Five months after choosing Mistral, the search company is leaving it. The founder, Christian Kroll, says the AI costs fell by roughly half, and gives three reasons besides price: quality, overloaded servers and Mistral's foreign investors. The new supplier is "probably" Melious, a German company that runs open models on green power, and he says the move is "not a move towards Chinese technology". The cost figure is his own and has no base. An AI supply contract can now be replaced within one budget year, so the lock-in a board fears is weaker than it looks, and so is the supplier's hold on price. EU-Startups
"Where agents shop, we are there." Delta keeps shopping agents away from its prices; Zalando lets them in on its own terms. Co-chief executive Robert Gentz told Handelsblatt that Zalando is testing advertising in ChatGPT, connecting its range to AI providers' interfaces and piloting Google's Universal Commerce Protocol, where "the customer buys directly in the AI chat, but remains a Zalando customer with their Zalando account". Agents bring "one percent of visitors, not of buyers" so far. Its own AI models put "13 percent more items in the basket", and size advice avoided "eight percent of size-related returns". The AI models can be swapped; "what cannot be swapped belongs to us: our knowledge of fashion, our data and the relationship with our customers." Delta and Zalando agree on what to protect, the customer relationship. They disagree on whether agents threaten it. Handelsblatt
"Booking Holdings spends nearly $9 billion a year to reach travelers. Its CFO explains how AI is changing trip planning". Ewout Steenbergen gives two measures. In engineering, its roughly 9,000 technologists put about 30% more merge requests, batches of finished code, into production, counting only code that "cleared quality controls". In customer service, booking units grew at a high single-digit rate while total costs "declined slightly". Back in August, on the results call, he put traffic from large language models at "still significantly below 1% of our room nights". On AI's return: "I think in this area there isn't a perfect answer." Output by function, and where possible against cost, is a comparison any board can ask its own management for; the money is the next question. Fortune
"Lawmakers raise alarm at Google plan to acquire Spirit Airlines data for AI models". Google won an auction in August for Spirit's internal records, including 100 million emails and 500 million Teams messages, for $10m, to train AI. Passenger data was excluded. A bankruptcy judge decides on Wednesday 14 October. A company's internal record is now an asset with a price, and in a sale or an insolvency someone else may own it. Reuters
"Treasury Announces Enforcement Penalty for Violation of Outbound Program". The first penalty under the US rules on investing in Chinese AI, chips and quantum computing: $200,000 on Amidi, LLC for not reporting a $92,478 investment its Chinese fund made in an AI and robotics company. The fine was more than twice the investment, and Congress has already passed a law widening the rules. Any fund with US partners or investors now has a price for an unreported AI investment in China. Treasury
Three questions from this issue for your next board meeting, each answerable with a number, a date or a name.
Ilona
- For any job cut we attribute to AI, what is the run-rate saving against the total cost of the change, and in which quarter does it show, allowing for what could delay it?
- How much did AI save us this year, in money, and who decided whether it went to profit or back into the business?
- Which of our customers' AI agents already compare us, on what criteria, and who decided how we compete there?
Four issues in.
Thank you for reading, and for writing back. I started this because I could not find a weekly read that took company numbers seriously and left the opinions out. Two things help me most: which item was the best one for you this week, and which sectors, countries or kinds of company you would like to see more of. If the issue is useful to you, pass it to one colleague who would put the three questions to their board.
AI, in board terms.
A glossary of 89 AI terms from proposals, results calls and regulators' letters, explained for the people who have to decide. Under each term: the policy, register, process or report to ask management for. Version 1 is on the website, with search and a PDF to download, and you can request terms.
AI ROI.
This newsletter shares its name with the book I am writing on how companies and their owners get a return from AI. Write to ilona@agentandcapital.com to hear when it is out.
A career technologist turned board director and advisor.
In executive roles at Porsche, DHL Supply Chain, Aston Martin, Innogy and ZF Foxconn I delivered ERP migrations, global infrastructure and cybersecurity builds and product-led transformations, each in under 20 months. I cut technology costs by 10 to 25%, including ahead of an IPO, and quadrupled the EBITDA of a business within a year ahead of its trade sale. I closed cyber and technology risks that due diligence had missed. As founder of Agent & Capital I advise private equity sponsors and executive teams on value creation and AI governance, and work with chief executives of technology companies on what their customers need from AI. I am a Harvard Certified Corporate Director and an ISACA-certified IT auditor with its Advanced in AI Risk credential.
Get AI ROI by email every Monday at 07:00 UK time: airoi.beehiiv.com/subscribe
AI ROI is written by Ilona Simpson and published by Agent & Capital Ltd, 71-75 Shelton Street, Covent Garden, London WC2H 9JQ. Sources are linked in each item; figures are as stated by the companies and institutions named. Every figure is checked against its source, but mistakes can still happen. I correct any error in the next issue. Not investment advice. © 2026 Agent & Capital.
Get AI ROI by email every Monday at 07:00 UK time. One story each on investment, return and risk. Unsubscribe in one click.