This Week in AI in Marketing & Management (6th Jul 26)

Claude Fable 5 Returns, AI Overviews Split the Data, and Agentic Commerce Goes Live in Europe

This was the week the cost of running AI agents dropped sharply, the cost of paid clicks kept rising, and the argument over what AI search does to organic traffic got a lot more complicated. Anthropic released Claude Sonnet 5 as a deliberate play on price rather than raw capability, confirming that the model race has moved from “who is most powerful” to “who is cheapest to run at scale.” Meanwhile, two studies landed on Google AI Overviews in the same week and reached almost opposite conclusions, which tells you something important about how you need to interpret that data. On the commerce side, Visa completed live AI agent-powered transactions across Europe with more than 30 banks and named participants including Barclays, HSBC UK, Lloyds, NatWest and Revolut, while Stripe launched its Agentic Commerce Suite in Germany with reach into 195 markets. On the management side, KPMG research of 2,000+ global leaders found that 42% still cannot see where their AI money goes, and a CIO analysis set out the four reasons AI projects fail that have nothing to do with technology. Here is what marketers and managers need to do about it.

In this week’s round-up

1. AI News, Tech & Tools

Anthropic launches Claude Sonnet 5 as a cheaper way to run agents

Sources: Anthropic | TechCrunch | 30 June 2026

Anthropic released Claude Sonnet 5 on 30 June, positioning it not as a frontier capability upgrade but as a cost reduction play for organisations already running agents at scale. The model can “make plans, use tools like browsers and terminals, and run autonomously at a level that, just a few months ago, required larger and more expensive models,” according to Anthropic’s blog post. TechCrunch notes that the framing mirrors what OpenAI and Google have said about their own recent releases, with all three labs converging on the same message: agentic capability is now the baseline expectation at every price tier, not a premium feature.

The competitive context matters here. OpenAI’s GPT-5.6 Sol launched in preview the previous week, also pitched as an agentic model. Google’s Gemini 3.5 Flash, launched in May, was similarly framed as a shift from conversational chatbot to agentic tool. The differentiator is no longer which model can do agentic work best, but how cheaply organisations can run those tasks. Sonnet 5 is Anthropic’s direct answer to that question.

Why it matters
For any organisation that has started running AI agents in production, this changes the cost conversation immediately. If your current agent stack runs on Claude 3.5 Sonnet or Opus for tasks that do not require the most powerful model available, Sonnet 5 is worth evaluating for a direct swap. The key test is whether the model maintains output quality on your specific workflows, not on benchmark scores. Run it on a representative sample of your actual agent tasks before committing to a full migration. The broader signal is that the model providers are competing hard on inference cost, which means the economics of running AI automation are improving faster than most organisations have planned for.

Anthropic brings Claude Fable 5 back globally after US lifts export controls

Sources: Anthropic | VentureBeat | 1 July 2026

Anthropic redeployed Claude Fable 5 globally on 1 July, restoring general access to its most powerful publicly available model. Fable 5 first launched on 9 June but was suspended on 12 June when the US government applied export controls to it and to the more capable Claude Mythos 5. With the controls now lifted, Fable 5 is available again across the Claude Platform, Claude.ai, Claude Code and Claude Cowork, with the major cloud providers to follow. Anthropic added an improved safety classifier that it says blocks the specific jailbreak technique behind the suspension in over 99% of cases.

For paid plans, Fable 5 was included for up to 50% of weekly usage limits through 7 July, after which it is available through paid usage credits. Anthropic described Fable 5 as a Mythos-class model made safe for general use, with performance in software engineering, knowledge work, vision and scientific research that exceeds anything the company had previously released to the public.

Why it matters
For any organisation already building on Claude, the return of Fable 5 restores access to the strongest general-purpose model Anthropic offers, which matters for the most demanding work such as complex coding, research and multi-step analysis. The practical point is to understand the pricing model before you rely on it: Fable 5 moves to usage-credit billing after the introductory period, so it is best reserved for tasks that genuinely need frontier capability rather than routine work that a mid-tier model such as Claude Sonnet 5 can handle at a fraction of the cost. Match the model to the task and the economics stay sensible.

Google’s AI updates from June 2026, including Gemini 3.5 Live Translate

Sources: Google Blog | 2 July 2026

Google published its monthly AI round-up for June, covering updates across Gemini, Android 17 and Google Workspace. Among the headline additions is Gemini 3.5 Live Translate, which provides real-time spoken translation, and a series of Android 17 features built around on-device AI. The recap also references expanded Gemini integrations across Google Workspace products and continued development of AI Overviews in Google Search.

The monthly format Google has adopted for these announcements reflects the pace of releases. Rather than individual launch events for each update, Google is bundling incremental changes into a single communique, which makes it easier for marketers to track the cumulative shift in Google’s product range rather than chasing individual news items.

Why it matters
The Gemini 3.5 Live Translate feature is the detail most relevant to international marketers and brands with multilingual audiences. If real-time spoken translation is accurate enough to be genuinely useful in customer-facing contexts, it opens new possibilities for video content and live events. The more immediate consideration for search and paid media teams is monitoring how each Workspace AI update changes the workflows of the people they are trying to reach. When the tools people use to work change, the intent patterns behind their searches tend to change shortly after.

OpenAI expands personal finance features to ChatGPT Plus customers

Sources: 9to5Mac | 30 June 2026

OpenAI has expanded its personal finance features from the $100-per-month ChatGPT Pro plan down to the $20-per-month ChatGPT Plus tier in the US. The features allow users to connect financial accounts and receive AI-driven analysis of their spending, savings and financial situation directly inside ChatGPT. The expansion is currently limited to the US market.

Why it matters
This is primarily a signal about the direction of travel for AI in financial services, but it matters for marketers in fintech, banking and retail financial products. As ChatGPT becomes the place where consumers analyse their own financial behaviour, the advertising and content strategies that worked on a separate financial app or a bank’s own website will need to adapt. The question for financial services marketers is whether they are building presence in the AI interfaces where consumers will increasingly do their financial thinking, not just in the search results where they used to look things up.

Amazon launches new $1 billion field deployment engineering organisation

Sources: TechCrunch | 30 June 2026

Amazon has created a new $1 billion Field Deployment Engineering (FDE) organisation, following similar moves by OpenAI and Anthropic. The FDE model embeds engineers directly within client companies to deploy purpose-built AI agents, with a focus on rapid deployment timelines and leaving customers self-sufficient once the engagement ends rather than creating ongoing dependency.

The pattern of the three largest AI labs all creating embedded deployment teams in the same period is notable. It signals that selling model access alone is not sufficient to capture enterprise value, and that the real competition is now for the implementation layer inside large organisations.

Why it matters
For organisations considering large-scale AI deployment, the arrival of embedded engineering teams from Amazon, OpenAI and Anthropic changes the procurement conversation. These are not consultancy engagements in the traditional sense. They are designed to move fast, build specific agents, and leave the client able to run and extend what was built. Organisations that already have internal AI capability will get more from these engagements. Those without it risk becoming permanently dependent on a single vendor’s approach. The practical advice is to use any embedded engagement as an opportunity to build internal understanding, not to outsource the thinking entirely.

UK firms prefer pre-built AI agents over custom code as token costs rise

Sources: IT Brief UK | 4 July 2026

Research from IT Brief UK finds that rising token costs are pushing UK businesses towards ready-made AI agents rather than custom-built solutions. The survey finds that organisations are prioritising speed of deployment and predictable cost structures over the flexibility that comes with bespoke development. The pattern holds across sectors and company sizes, with even organisations that have internal development capability choosing to start with pre-built agents and customise from there rather than building from scratch.

Why it matters
This reflects a pragmatic shift that is already visible in how UK businesses talk about AI investment. The early adopter phase, when bespoke development was a point of competitive differentiation, is giving way to a more operational mindset in which time-to-value and cost per task are the primary metrics. For marketing technology teams, this means the conversation around AI tooling is converging with the conversation around any other SaaS procurement decision. Pre-built tools with clear pricing models will win more deployments than custom-built systems, even when the custom approach would theoretically produce better outputs.

AI governance needs to scale at the pace of AI

Sources: Forbes Tech Council | 2 July 2026

A Forbes Tech Council piece argues that AI governance policies which are not embedded directly into workflows leave organisations in a permanently reactive posture. The author distinguishes between governance frameworks that exist as documents and governance that is built into the actual systems through which AI decisions are made. The former, they argue, is almost useless in practice because the pace of AI deployment outstrips any policy review cycle.

Why it matters
For marketing leaders, this is a practical challenge that is already arriving. As AI tools become embedded in campaign management, content production and customer communication, the governance question is not whether to have a policy but whether that policy is enforceable at the point where the AI actually operates. A policy document that nobody reads before using a new AI tool is not governance. Governance is a review step in the workflow, a human sign-off on AI-generated outputs before they reach customers, or a clear escalation path when an AI recommendation is uncertain. The organisations that build this into their processes now will avoid the harder conversation later when something goes wrong.

Microsoft merges enterprise and consumer Copilot apps

Sources: PYMNTS | Microsoft Blog | 6 July 2026

Microsoft is planning to combine its Copilot consumer and enterprise AI applications into a single product, eliminating features that have low adoption. The move comes alongside Microsoft’s positioning of itself as a “Frontier Company” – a new term it is using to describe organisations that have moved beyond AI experimentation into full operational deployment. Microsoft’s own blog post notes that “customers have moved well beyond experimentation and understand the importance of adopting AI.”

Why it matters
The merger of consumer and enterprise Copilot is significant for organisations that use Microsoft 365. A single product surface means that the AI behaviours employees encounter in their personal use of Microsoft tools will increasingly match what they encounter at work, which should reduce the learning curve but also means Microsoft has more influence over how AI assistance is experienced across both contexts. For IT and marketing leaders, the immediate implication is to review which Copilot features your teams are actually using before the product rationalisation removes options you depend on.

2. AI in Marketing

Three studies land on AI Overviews in one week, and they do not agree

Sources: Marketing Dive | MediaCat UK | Martech Cube | 1 July 2026

Three pieces of research on Google AI Overviews landed in the same week and reached almost opposite conclusions. Research by LQ Digital, shared with Marketing Dive, found that 42% of brand citations in organic search results do not appear in AI Overviews for the same query, and that 28% of brands cited by AI Overviews do not appear in the organic results at all. YouTube videos are 4.3 times more likely to appear in AI Overviews than in standard search, and how a query is worded significantly affects which brands appear. Separately, a study by professors Saharsh Agarwal and Ananya Sen, randomly assigning 1,000 real users to Google Search with or without AI Overviews, found that AI Overviews reduced outbound organic clicks by 39.8% and increased zero-click searches by 34.5%, with no measurable improvement in user satisfaction.

Against those two findings, incrementality research reported by Martech Cube found that AI Overviews actually boosted incremental revenue and orders for the brands studied, arguing that the users who do click through from an AI Overview are further along in their decision and more likely to convert. Both sets of findings are measuring something real: the LQ Digital and university work measures aggregate click volume, while the incrementality work measures revenue per click.

Why it matters
The honest reading is that all three are true at once. AI Overviews reduce the number of clicks a page receives, and the clicks that remain tend to be higher quality. What that means for your brand depends on your margins, your current traffic volume, and how much of your funnel relies on volume rather than quality. Brands with high-volume, low-margin traffic will feel the 40% click reduction most acutely; brands with high-margin, considered-purchase products may find the quality improvement compensates. The two most actionable data points are the YouTube multiplier, so if you have no substantive YouTube content you are underrepresented in AI Overviews, and the query-wording sensitivity, which means optimising for a single canonical query is not enough. The practical step is to run your own incrementality analysis rather than assuming any one study applies to you.

Why Generative Engine Optimisation is the future of AI search visibility

Sources: Entrepreneur | 4 July 2026

Entrepreneur.com describes Generative Engine Optimisation (GEO) as one of the biggest shifts in how people find businesses online. The article explains the core principle: where traditional SEO focuses on signals that cause a page to rank in a list of links, GEO focuses on signals that cause a brand or piece of content to be cited or referenced in an AI-generated response. The two disciplines overlap significantly but are not identical, and organisations that treat GEO as simply a new name for existing SEO practice will underinvest in the areas where they differ.

Why it matters
GEO is not a future consideration. It is a present one for any brand in a competitive category. The practical differences from traditional SEO include a greater emphasis on being cited by third-party sources, structured data that helps AI systems understand what your business does and who it serves, and content that answers specific questions rather than targeting broad keyword clusters. If your organisation does not have a GEO strategy separate from your SEO strategy, this week is a reasonable moment to start that conversation.

How to get cited in Google AI Overviews – Backlinko’s analysis

Sources: Backlinko | 3 July 2026

Backlinko’s analysis of what drives citations in Google AI Overviews proposes the concept of “owning a criterion”: being the definitive source on a specific, well-defined aspect of a topic rather than attempting to cover the whole topic. The analysis argues that AI systems are more likely to cite sources that are clearly authoritative on a narrow, well-defined question than sources that cover broad topic areas at a high level. Supporting signals include strong internal linking, structured data markup, and consistent citation by other authoritative sources in the same domain.

Why it matters
The “own a criterion” framing is a useful planning tool for content teams. Instead of asking “what topics do we cover?”, the more productive question is “what specific questions are we the most credible source to answer?” For most organisations, that will produce a much smaller and more focused content agenda than the broad topic map that drives most SEO content strategies. That is a feature, not a limitation. Focused, authoritative content on a narrow criterion is more likely to generate AI citations than a large volume of broadly competent content on many topics.

Google Ads testing AI-generated summaries under sponsored descriptions

Sources: Search Engine Roundtable | 2 July 2026

Google is testing a new format in paid search results where AI-generated summaries appear directly below the advertiser’s written description. The summaries are placed in the sponsored results section and are generated by Google’s AI from the landing page and ad content, rather than being written by the advertiser. The test is currently limited and has been spotted by a small number of users.

Why it matters
This is an early signal of where Google Ads creative is heading. If AI-generated summaries become a standard element of the paid search result, advertisers will have less control over what users see before clicking, but the AI may also surface more relevant information than a manually written description that has been constrained by character limits. The practical response is to ensure your landing page content and ad copy are consistent and clearly answer the questions your target audience is most likely to have. AI summaries generated from ambiguous or contradictory content will underperform relative to well-structured, question-answering landing pages.

Google Ads API v24.2 adds AI transparency and PMax segmentation

Sources: PPC Land | 29 June 2026

Google Ads API version 24.2 introduces synthetic content labelling, multi-party approvals for AI-generated creative, Performance Max ad network segmentation, and campaign mix experiments. The PMax segmentation update is particularly notable: advertisers can now see how their Performance Max budget is distributed across ad networks within the campaign, which has been a significant gap in PMax reporting since the campaign type launched.

Why it matters
The PMax segmentation update is the most practically significant item here for paid media teams. The inability to see how PMax budget was distributed across Search, Shopping, Display, YouTube and other placements has been a persistent criticism of the format and has made it very difficult to attribute performance accurately. This change does not restore the full control that Standard Shopping or Search campaigns provide, but it does give teams the data to have a more informed conversation about whether PMax is allocating budget in a way that matches their strategic priorities. If you run PMax campaigns, reviewing this new segmentation data should be a priority task this week.

AI is making creative the new targeting across Google, Meta and TikTok

Sources: Martech.org | 30 June 2026

A Martech.org analysis argues that as Google Ads, Meta and TikTok all push advertisers toward broader, AI-driven audience targeting, creative quality becomes the primary lever for campaign performance. When the algorithm handles audience selection, the differentiation between a campaign that works and one that does not comes down almost entirely to the creative assets. The article argues that the traditional media-buying skill of audience definition is being absorbed by the platforms’ AI systems, leaving human expertise to focus on the brief, the message, and the creative execution.

Why it matters
This shift has significant implications for how paid media teams are structured and where investment is allocated. If targeting is increasingly handled by platform AI, the return on investment in media buying expertise is falling relative to the return on investment in creative capability. Organisations that have historically allocated most of their paid media budget to media buying and platform management and relatively little to creative production are likely to see performance gains from rebalancing that allocation. The question to ask in your next campaign review is: what percentage of our paid media spend goes on creative, and does that reflect where performance is actually being determined?

Microsoft introduces new Performance Max experiment types

Sources: Performance Marketing World | 2 July 2026

Microsoft Advertising has introduced two new Performance Max experiment types: Uplift experiments, which measure the incremental impact of adding PMax to an existing campaign mix, and Upgrade experiments, which test the performance change when migrating existing campaigns to PMax. The additions give Microsoft Advertising a more structured testing framework than has previously been available for PMax on the platform.

Why it matters
For advertisers running both Google and Microsoft paid campaigns, this makes it easier to run comparable PMax experiments across both platforms and draw meaningful conclusions about whether PMax is the right format for specific campaign objectives. The Upgrade experiment type is particularly useful for teams that have been hesitant to migrate existing campaigns: it provides a controlled way to test the transition without committing the full campaign budget to a new format.

Cost of clicks rises 15% year-over-year and ROAS drops 46%

Sources: ChannelX | 2 July 2026

Independent research by Channable, the multichannel e-commerce and feed management platform used by more than 17,000 brands and agencies, reports that cost-per-click across paid channels has risen 15% year-over-year while ROAS has dropped 46%. The research draws on data from Channable’s own platform and covers a broad range of paid channels. The combination of rising acquisition costs and falling returns is being described as a structural shift rather than a cyclical one.

Why it matters
A 15% CPC increase alongside a 46% ROAS decline is a severe deterioration in paid media efficiency by any measure, and it is happening at the same time as organic clicks are being reduced by AI Overviews. The two trends together create a genuinely difficult position for brands that have built their customer acquisition model around either paid search or organic search. The response is not to spend more on the same channels in the hope that efficiency recovers. It is to diversify acquisition channels, invest in direct audience relationships, and treat paid media as one component of a broader mix rather than the primary driver of growth. Brands that have already invested in owned audiences, email lists, and community will be significantly better positioned than those that have not.

UK advertisers now pay a 2% Meta location fee as Digital Services Tax is passed on

Sources: Webtopia | Bloomberg | 1 July 2026

From 1 July 2026, Meta has started adding location fees to advertising delivered to UK and EU audiences, passing on the Digital Services Tax costs it previously absorbed. For the UK the fee is 2% of spend, matching the UK DST rate, with higher rates elsewhere (3% for France, Italy and Spain, 5% for Austria and Turkey). The fee is calculated on where the ad is shown, not where the advertiser is based, and it is added on top of the campaign budget rather than taken out of it. That means the total invoice will exceed the figure shown in Ads Manager, and VAT applies on top where relevant. Google and Amazon already charge similar fees.

Why it matters
This is a direct, immediate cost increase for any UK business advertising on Facebook or Instagram, and it is easy to miss because it does not show up inside the Ads Manager spend figure. If you budget and report from Ads Manager alone, your actual Meta bill from July onward is 2% higher than the number your reports show, before VAT. The practical steps are to update your budget models and client billing to account for the fee, check that your return-on-ad-spend calculations use the true invoiced cost rather than the Ads Manager figure, and make sure clients are told about the change before they see it on an invoice. For UK advertisers running large Meta budgets, 2% is a meaningful sum that should be planned for, not absorbed by surprise.

Sources: Social Media Today | 2 July 2026

LinkedIn has officially launched a brand kit feature for marketers, allowing teams to set colour palettes, fonts and other brand guidelines within LinkedIn’s Campaign Manager. The update also includes new AI-powered tools for promotional content creation that use the brand kit settings to generate on-brand ad creative. The brand kit feature addresses a longstanding gap in LinkedIn’s advertising tools relative to Meta and Google’s more developed brand governance capabilities.

Why it matters
The brand kit launch is a practical improvement for any team running LinkedIn campaigns who has struggled with inconsistent creative output from LinkedIn’s AI ad tools. Setting brand parameters before generating creative should reduce the manual editing time required to bring AI-generated LinkedIn ads up to brand standard. If your team uses LinkedIn Campaign Manager and has not yet explored the brand kit feature, this is worth setting up before the next campaign build.

Building AI image and video workflows for marketers

Sources: Social Media Examiner | 30 June 2026

Social Media Examiner published a structured guide to building AI image and video workflows that maintain brand consistency while generating high-end assets at scale. The approach outlined emphasises starting with a documented brand foundation (visual identity, tone, audience) before introducing AI generation tools, and using AI to fill in within defined parameters rather than to generate from scratch without constraints.

Why it matters
The brand-foundation-first approach described here is the difference between AI creative tools that save time and AI creative tools that create a review and correction burden. Teams that have tried to adopt AI image generation without a documented brief for the AI tend to produce a high volume of assets that need significant manual correction, which often eliminates the time saving. The practical recommendation is to document your visual brand parameters in a format that can be used as a prompt prefix or system instruction for any AI image tool your team uses. Thirty minutes spent on that document will save hours of revision across every subsequent campaign.

3. AI in Management

Four reasons AI projects fail that have nothing to do with technology

Sources: CIO | 2 July 2026

A CIO analysis by Richard Mendis, drawing on work with dozens of companies across stages of AI adoption, identifies four non-technical reasons AI projects fail: employee fear of job replacement, a lack of AI-first culture, competing leadership priorities, and governance structures that diffuse accountability. On the fear point, Mendis cites a Writer study finding that 29% of employees, and 44% of Gen Z employees, admit to having sabotaged their employer’s AI strategy. He also notes that the Microsoft Work Trend Index found “in many cases, people are ready, the systems around them are not,” suggesting the bottleneck is often structural rather than human.

Why it matters
The sabotage statistic is striking and worth taking seriously as a genuine risk factor in AI adoption planning. If nearly a third of employees actively resist AI initiatives, the ROI projections in any AI business case that assume smooth user adoption are likely to be too optimistic. The practical implication is that communication, change management and genuine workforce involvement in how AI tools are introduced need to be treated as project deliverables, not as afterthoughts. Leaders who announce AI tools rather than introduce them with explanation and training are creating the conditions for the failure pattern this article describes.

KPMG: enterprises are scaling AI but 42% still cannot see where the money goes

Sources: UC Today / KPMG Global AI Pulse Q2 2026 | 2 July 2026

KPMG’s Global AI Pulse Q2 2026 report, based on 2,000+ senior leaders across 20 countries at organisations with revenues above $50 million, finds that 22% of organisations now describe AI as part of everyday work (up from 13% in Q1), AI retained its status as a top investment priority for 79% of leaders, and average AI spending is holding steady at $188 million. However, only 7% of leaders report having established ROI from AI. Forty-two percent report only partial visibility into AI spending, 33% cite limited understanding of token-based pricing as a major challenge when deploying AI agents, and 49% say they have already scaled back AI agent deployments because costs outweighed the benefits.

The governance finding is equally pointed: organisations where the CEO is explicitly accountable for AI outcomes are nearly four times more likely to report established ROI (14% versus 4%) and report significantly higher confidence in their AI strategy (60% versus 22%).

Why it matters
The gap between AI investment and established ROI is the central management challenge in AI right now, and these numbers quantify it precisely. The token-based pricing complexity point is particularly relevant: usage-based AI costs behave differently from traditional software licensing, and organisations that are accustomed to predictable per-seat SaaS costs are finding that AI agent deployments can generate highly variable spend that is difficult to forecast. Building cost monitoring infrastructure before scaling AI deployments, not after, is the practical recommendation from this data. The CEO accountability finding is also actionable: diffuse accountability produces diffuse results, and naming a single executive who owns AI ROI is associated with materially better outcomes.

HR in 2030: how will HR embrace agentic AI in the next four years?

Sources: HR Magazine | 4 July 2026

HR Magazine sets out the choices facing HR functions as agentic AI becomes capable of handling tasks that have historically required human judgement: dynamic workforce planning, skills matching, and elements of performance management. The article frames HR as standing at a fork in the road between being an early adopter of agentic AI to handle administrative and analytical functions, and being an organisation that resists the change and finds itself managing the consequences rather than the process.

Why it matters
The HR function is both a user of AI and a manager of the human response to AI across the wider organisation, which puts it in an unusual position. HR teams that adopt agentic AI for their own workflows will have direct experience of the capability and limitations of the technology, which makes them better equipped to support the rest of the organisation through the same transition. HR teams that remain outside the AI adoption curve will find themselves advising on a transformation they have not participated in, which is a credibility challenge in practice.

Why AI adoption is becoming the deciding factor in digital transformation projects

Sources: Enterprise Times | 30 June 2026

Enterprise Times argues that AI capability has moved from being a component of digital transformation programmes to being the primary criterion against which those programmes are judged. Organisations managing growing data volumes, ageing systems and competitive pressure are finding that transformation projects which do not have a credible AI integration story are increasingly difficult to justify to boards. The article describes a shift in how investment decisions are made: the question is no longer whether a transformation project incorporates AI, but whether the AI integration plan is substantive or cosmetic.

Why it matters
For marketing leaders involved in technology investment decisions, this shift in how boards evaluate transformation projects has practical consequences. Marketing technology proposals that do not include a clear AI capability roadmap are increasingly at a disadvantage in the budget allocation process, regardless of their other merits. The more substantive challenge is that “we will add AI later” is no longer a credible position for a new platform or tool selection. AI integration needs to be part of the initial requirement, not a subsequent upgrade.

4. AI in E-commerce, Retail and Agentic Commerce

Visa completes live AI agent-powered transactions across Europe with 30+ banks

Sources: FStech | 4 July 2026

Visa has completed live AI agent-powered payment transactions across Europe, with more than 30 banks now enabling autonomous purchases at participating merchant websites. The announcement was made at the Visa Payments Forum in Paris and marks a shift from controlled demonstrations to real-world transactions with independent merchants. Participating banks include Barclays, HSBC UK, Lloyds Banking Group, Nationwide Building Society, NatWest, Revolut and Klarna. Merchants involved span travel, retail and e-commerce and include lastminute.com, Frasers Group, Cleverbridge and BrickDepot.

Visa’s Trusted Agent Protocol and Agent Directory allow merchant websites to identify verified AI agents and distinguish them from unverified automated traffic. Transactions are authenticated using Visa Payment Passkeys, which the company says ensure every payment is linked to a verified cardholder and explicit authorisation while meeting Europe’s Strong Customer Authentication requirements. Visa said the model will be extended to commercial and B2B transactions where trusted AI agents could automate purchasing processes while maintaining visibility for businesses and financial institutions.

Why it matters
This is not a pilot. Live transactions with named major UK banks and named merchants confirm that agentic commerce has crossed from demonstration into production infrastructure. For e-commerce operators and retailers, the question is now when to engage with agentic payment standards rather than whether they will ever be relevant. The Trusted Agent Protocol is the detail that matters most: merchants will need to implement this to be discoverable by AI shopping agents, in the same way that structured data implementation made products discoverable in Google Shopping a decade ago. Retailers that build this capability early will have a material advantage over those who wait for it to become a formal requirement.

Stripe launches Agentic Commerce Suite in Germany with reach into 195 markets

Sources: FF News | 1 July 2026

Stripe has launched its Agentic Commerce Suite in Germany, enabling businesses to sell products through AI interfaces and reach 195 markets via new Managed Payments and localised pricing tools. The suite includes Adaptive Pricing, which Stripe’s own data shows increases revenue by 17.8% by presenting localised prices to buyers in their own currency. Stripe also blocked €2 billion in fraud in Germany in the past year, and the new suite includes enhanced fraud prevention tools integrated with the agentic commerce functionality.

Why it matters
The 17.8% revenue uplift from Adaptive Pricing is a concrete number that makes the business case for agentic commerce infrastructure straightforward for any business with international customers. Stripe’s global reach means this is not limited to large enterprises: the suite is designed to be accessible to businesses of all sizes selling internationally. For any UK e-commerce business that has not yet localised its pricing for international markets, the combination of agentic commerce capability and currency-localised pricing represents a meaningful revenue opportunity that does not require a large development investment.

Prime Day 2026: how AI impacted Amazon and wider online retail

Sources: Retail Week | 29 June 2026

Retail Week’s analysis of Prime Day 2026 examines how AI tools and assistants are now actively driving sales decisions on Amazon and across wider online retail. The piece covers the role of Amazon’s AI-powered search and recommendation systems in directing buyer attention during the high-volume sale period, and the impact of third-party AI shopping tools that are increasingly being used by consumers to identify and compare deals across multiple retailers simultaneously.

Why it matters
Prime Day is the clearest annual stress test of where AI sits in the consumer purchase journey, because the volume of deals and the time pressure force buyers to rely on shortcuts and recommendations rather than researching every option manually. The growth of third-party AI shopping agents that compare across multiple retailers simultaneously is the trend that matters most for brands: it means product visibility in AI-driven discovery tools is becoming as important as product visibility in Amazon search itself. For brands that sell across multiple channels, ensuring consistent, accurate product data feeds to all AI-accessible shopping tools is a priority that should be addressed before the next major retail event.

5. AI for Other Sectors and Industries

FINANCE: Bank of England warns autonomous AI agents could trigger a market meltdown

Sources: Let’s Data Science | 30 June 2026

Speaking at the European Central Bank’s Sintra forum on 30 June, Bank of England Deputy Governor Sarah Breeden warned that autonomous AI agents operating in financial markets could amplify volatility under stress and potentially trigger a market meltdown, and said existing financial regulation was not built for agentic AI. The Bank is exploring enhanced recovery arrangements that would let one bank take over another’s core functions during a disruption, alongside market-wide circuit breakers or kill switches to halt trading if faulty AI models cause correlated failures.

Why it matters
For any business in financial services, this signals that the regulatory conversation about agentic AI has moved from theory to active supervisory planning. Firms deploying AI in trading, risk or core decision-making should expect closer scrutiny and should be able to explain how their systems behave under stress. For marketers in the sector, the message is that trust and demonstrable safety are becoming central to how financial brands are judged, so content and positioning that address AI governance and safeguards will matter more, not less, as regulation catches up.

LEGAL: Legal services named the first sector for the UK’s AI Growth Labs sandbox

Sources: Legal Futures | 2 July 2026

Justice Secretary David Lammy announced that legal services will be the first sector to join the UK government’s new AI Growth Labs, a regulatory sandbox that lets AI systems be tested within a structured, compliant environment. Applications open later this summer for technology innovators including LawTech firms, legal service providers and conveyancing companies. The announcement came alongside research showing AI use across UK and Ireland law firms is now near-universal, with nearly 9 in 10 legal professionals using it in some capacity and UK practitioners recording the highest daily use of integrated legal AI tools globally.

Why it matters
The sandbox model is worth watching well beyond the legal sector, because it is the template the UK government is likely to apply to other regulated industries. For legal firms specifically, near-universal adoption means AI is no longer a point of differentiation but a baseline expectation, and the firms pulling ahead are those integrating it into pricing and service models rather than treating it as an add-on. The parallel shift to fixed fees, as AI compresses billable hours, is a reminder that AI adoption eventually reshapes the commercial model of a professional service, not just its back office.

HEALTH: NHS gives more than 500,000 staff Microsoft Copilot after admin time falls

Sources: NHS England | 3 July 2026

NHS England is giving more than 500,000 staff access to Microsoft Copilot, following a trial in which workers cut the time spent on administration by an average of two days a month. The AI assistant helps staff draft documents and analyse data, with the stated aim of freeing up time for patient care. Alongside the rollout, patients will be able to use the NHS App to request follow-up appointments, and NHS-approved digital tools will help them manage rehabilitation for common lung and heart conditions.

Why it matters
This is one of the largest single deployments of a general-purpose AI assistant in any UK organisation, and the two-days-a-month figure is a concrete productivity result rather than a projection. For any organisation weighing an AI rollout, the NHS example is a useful reference point: the value came from applying an existing tool to routine administrative work at scale, not from a bespoke system. It also signals that AI assistance is becoming standard in public-facing services, which raises the baseline expectation that private-sector organisations will be measured against.

PHARMA: Anthropic launches Claude Science, an AI workbench for researchers and pharma

Sources: Anthropic | TechCrunch | 30 June 2026

Anthropic launched Claude Science on 30 June, an application that adapts its existing Claude models for scientific research and, in particular, for the research operations of pharmaceutical companies. Notably, it is not a new or more capable model. It runs the same Claude models already available to everyone, including Claude Opus 4.8, with no special access or gating. What is new is the workflow around the model: Claude Science combines databases, coding tools, compute and research workflows in a single workspace, pre-configured with more than 60 scientific databases and connectors for genomics, proteomics, structural biology and cheminformatics. Anthropic also said it will use the product to pursue its own research into drugs for rare and neglected diseases, and is funding up to 50 AI for Science projects with up to $30,000 in credits each, with applications open until 15 July.

Why it matters
The strategic signal here is more interesting than the product itself. Anthropic is betting that the way to win a professional audience is not a bigger model but a better workflow built around the model the audience already has. That is a lesson well beyond science: for most organisations, the value of AI now comes from wrapping existing models in the right tools, data connections and processes for a specific job, rather than waiting for the next frontier release. If your business has been holding back on AI until the models get better, Claude Science is a reminder that the constraint is usually the workflow around the model, not the model itself. The same models that power a drug-discovery workbench are available to build a workbench for your own work.

Key Takeaways

Claude Sonnet 5 confirms that inference cost is now the primary AI model competition axis. Capability parity has largely been achieved at the mid-tier. The question your organisation should be asking is not which model is most powerful but which model delivers acceptable quality at the lowest cost per task.
AI Overviews reduce organic clicks by approximately 40% but the clicks that remain convert at a higher rate. Both findings are accurate for different measurement approaches. Your response should depend on whether your business model is optimised for traffic volume or conversion quality.
42% of brands cited in organic search do not appear in AI Overviews for the same query. Your organic SEO presence and your AI search presence are different populations. You need a GEO strategy that is distinct from, though informed by, your existing SEO strategy.
YouTube content is 4.3 times more likely to appear in Google AI Overviews than standard search results. If your brand does not have substantial YouTube content, you are systematically underrepresented in AI-generated search responses.
CPC costs are up 15% year-over-year and ROAS has dropped 46%. This is a structural deterioration, not a cyclical one. Diversifying acquisition channels and investing in owned audiences is a financial imperative, not just a strategic preference.
Visa has completed live agentic payments with 30+ European banks including Barclays, HSBC, Lloyds, NatWest and Revolut. Agentic commerce is production infrastructure now. Retailers that begin implementing Trusted Agent Protocol compatibility today will be in a significantly better position than those who wait.
49% of enterprises have already scaled back AI agent deployments because costs outweighed benefits. Token-based pricing is catching organisations by surprise. Build cost monitoring infrastructure before scaling agents, not after.
Only 7% of leaders have established ROI from AI despite 79% naming it a top investment priority. The gap is primarily an accountability and measurement problem. Organisations where the CEO explicitly owns AI outcomes are nearly four times more likely to report established ROI.
29% of employees admit to sabotaging their employer’s AI strategy. Change management and communication are project deliverables in AI adoption, not optional extras. Business cases that assume smooth user adoption are overstating expected returns.
AI governance that lives in policy documents rather than embedded in workflows is effectively non-functional. Governance needs to exist at the point where the AI actually operates, not in a document that nobody reads before using a new tool.

Frequently Asked Questions

What is Claude Sonnet 5 and why does it matter?
Claude Sonnet 5 is Anthropic’s latest mid-tier AI model, released on 30 June 2026. It delivers agentic capabilities, including the ability to use tools, browse the web and run autonomously on multi-step tasks, at a lower cost than the frontier models. It matters because it signals that the competition between AI labs has shifted from raw capability to cost efficiency, which changes the economics of running AI automation for organisations of all sizes.

What is the difference between AI Overviews and organic search rankings?
AI Overviews are the AI-generated summaries that appear at the top of Google Search results for many queries. They draw from a different set of sources than the traditional blue-link organic results. Research this week confirmed that 42% of brands that appear in organic results for a query do not appear in the AI Overview for the same query, and vice versa. This means optimising for organic rankings and optimising for AI Overview citations require different but overlapping strategies.

What is Generative Engine Optimisation (GEO)?
GEO is the practice of optimising content and brand presence to increase the likelihood of being cited or referenced in AI-generated search responses, including Google AI Overviews, ChatGPT Search, Perplexity and other AI search tools. It differs from traditional SEO in its emphasis on third-party citation, structured data, and content that answers specific questions with authority rather than targeting broad keyword clusters.

What is agentic commerce and why is the Visa announcement significant?
Agentic commerce refers to commercial transactions initiated and completed by AI agents acting on behalf of human users, without requiring the user to manually navigate the purchase process. The Visa announcement is significant because it confirms that live, real-world agentic transactions are now taking place across a major European payment infrastructure, with mainstream UK banks including Barclays, HSBC, Lloyds, NatWest and Revolut already participating.

What is the KPMG finding about AI ROI?
KPMG’s Q2 2026 Global AI Pulse, covering 2,000+ senior leaders, found that while 79% of organisations name AI a top investment priority and average AI spending is $188 million, only 7% report having established ROI from AI. The organisations most likely to demonstrate ROI are those where the CEO is explicitly accountable for AI outcomes, and those that have built cost monitoring infrastructure to track token-based AI spending.

Conclusion

The pattern across this week’s news is a consistent move from experimental to operational, accompanied by a growing accountability gap between what organisations are spending on AI and what they can demonstrate in return.

On the technology side, Claude Sonnet 5 and the broader competitive pressure from Google and OpenAI are driving down the cost of running AI agents, which will accelerate deployment timelines for organisations that have been waiting for acceptable unit economics. On the commerce side, Visa’s live transactions across 30+ European banks and Stripe’s 195-market Agentic Commerce Suite confirm that the infrastructure for AI-driven purchasing is production-ready, not theoretical.

For marketers, the AI search picture is genuinely complicated this week, with two credible studies reaching different conclusions about what AI Overviews do to performance. The honest answer is that both are right for different metrics, and your response needs to be based on your own measurement of what matters for your business, not on a headline from either study.

The management findings from KPMG and the CIO analysis together paint a picture of organisations that are scaling AI spend faster than they are building the governance, accountability and measurement infrastructure to manage it. The organisations that close that gap, by building cost visibility, naming clear ownership, and treating user adoption as a deliverable rather than an assumption, will be significantly better positioned in 12 months than those that continue to treat AI as a technology project rather than an operational and financial one.

Ann Stanley is Founder & CTO of Anicca Digital and Anicca AI & Insights. If you want to discuss any of the themes in this week’s round-up, you can reach the team at [email protected] or visit anicca.co.uk.

Free download

New Edition (3) of The Complete
Claude Code Implementation Guide

Ann StanleyFounder & CTO, Anicca Digital

Updated May 2026 – 88 pages for marketers, managers and business leaders, now with a brand new chapter on Company AI & AI operating systems. No coding required.

Why it matters
Understanding Claude Code
Getting set up
Working with Claude
Skills
Business use cases
Rollout & governance
Company AI & AI operating systems

Download → anicca.co.uk/claude-code-guide

The Complete Claude Code Implementation Guide ebook cover - Edition 3

Join the Thursday AI Club

A hands-on AI club for marketers and managers. Fortnightly 3-hour sessions: Hour 1 is an open Q&A on any AI question; Hours 2-3 split into a workshop track (for newbies) and an advanced track (live demos and deeper questions).

Led by Ann Stanley, Darren Wynn, James Allen.

Workshop recordings library

Skills and prompts to copy

Private community channel

Secret Agents community access

Full-day hackathon every quarter

Cancel any time

Membership: £40/month or £400/year

Join the Thursday AI Club →

This weekly round-up is curated and written by the team at Anicca Digital and Anicca AI & Insights. Sources are linked throughout. For help applying any of this to your own marketing, contact us at [email protected].