Roger Dunn

The third option matters most, they argue, because it is the one most retailers took without choosing: a platform-level change quietly enrolled every store trading on it. They also argue that blocking is harder to sustain than it appears, because the legal question of who is 'accessing' a site when an agent shops may not fall the way a retailer assumes.

Amazon shut Meta's Muse out on 20 September; Walmart, Best Buy, Sephora, Ulta Beauty and Wayfair signed on as Muse partners within days; and on 28 September Shopify let any browser agent complete a checkout on its stores once the shopper approves, with no set-up needed from the merchant.

• Dunn notes an August US appeals court finding that when an agent shops, the shopper is the one accessing the site, leaving a block to rest on terms of use and on spotting the agent at all.
• The post points to a comparison Dunn published of seven agents that can already buy — Muse, Instinct, OpenAI's Dots, Gemini Spark, Manus AI's Cue, xAI's GrokBot and Microsoft Copilot — and what actually lands in a retailer's order system.

Read original

If the sale happens somewhere the ad is not, the network is closer to advertising on a screen in a petrol station — reach without a direct link to the transaction.

• They apply this to McDonald's without dismissing it, conceding that a large loyalty base could make such a network effective anyway.
• Buyers should demand more than impressions as evidence that the placement worked.

McDonald's confirmed a media network at its 23 September Chicago investor day, live since August in about 450 company-owned US restaurants — roughly 3% of its ~14,000 US sites — with other brands' ads on digital order boards shown after the customer orders, and a stated ambition of a $1 billion business across app, kiosks, menu boards and restaurants.

• Dunn notes that unless your product is on the menu, the sale happens elsewhere, putting this closer to the petrol-pump screen than Sponsored Products.
• Nearly 220 million 90-day active loyalty members across 70 markets could change the maths.
• They also flag that franchisees running most of the US estate are not yet in the pilot, and cite executive Dario Baroni saying the rebuilt digital platform was built to plug into conversational commerce channels without reengineering the core.

Read original · Original 2

Once an assistant answers 'what should I buy', the product data a retailer or brand holds stops being purely an operations concern and starts behaving like advertising inventory, because it is what the answering system reads.

• The argument is as much about where influence moves as about who sells it.
• Dunn says plainly that this is the logic Thrad builds around, making the thesis company positioning as well as observation.

In a post about being named to the RTIH Top 100 Retail Technology Influencers List, Dunn writes that retail technology spent a decade optimising the shelf and the next decade is about the shortlist.

• When an assistant answers 'what should I buy', product data stops being an operations problem and becomes media inventory.
• The post also describes Thrad as building ways for brands to appear inside those answers and for retailers to monetise conversational surfaces they run themselves, and notes a weekly LinkedIn newsletter, The AI Commerce Brief.
• A TikTok post makes the parallel point: once discovery becomes a conversation, the commercial question for any platform is what a sponsored placement looks like inside the answer rather than around it.

Read original · Original 2

They keep the vendor's own wording and figures attached to the item they came from, then distinguish confirmed capability from direction of travel. Where they speculate, the post labels it as anticipation rather than fact.

On TikTok's 5 October newsroom post, Dunn lists Buy Direct in-app checkout from the For You feed, a conversational Shopping Assistant handling product details, shipping, sizing, availability and purchase, and Agentic Leads, plus the partner set behind them — Salesforce, Shopify, Shoplazza, Stripe and others.

• They note that no launch dates or markets were given, making this a direction of travel rather than something bookable today.
• TikTok for Business MCP, built on the open Model Context Protocol, is now a connector inside Claude, Kimi, Manus AI, Perplexity, Replit, Snowflake, Tencent WorkBuddy and the IAB Tech Lab Agent Registry.
• The figure showing advertisers using it for campaign activation and management up more than 200% since launch is flagged as internal data for July to September 2026, with the base not disclosed.
• On the McDonald's network, they separate what is live from the $1 billion ambition; on OpenAI's dots, they label the expectation that a cloud browser will be blocked and fall back to the owner's browser as anticipation rather than fact.

Read original · Original 2 · Original 3

If a sale can begin in a chat and finish without a click, click-based attribution does not describe what happened, and the platform hosting the conversation may sit between the retailer and its customer. They raise these as unresolved questions the announcements leave unaddressed, not as settled outcomes.

On TikTok's Shopping Assistant, Dunn argues that an on-platform checkout raises the question of who holds the customer relationship, and notes the announcement does not say how order data or returns are handled.

• The next things to watch are whether Shopping Assistant carries ads and how TikTok attributes a sale that starts in a chat and finishes without a click.
• On OpenAI's dots, purchases run through retailers' own sites, where logged-in accounts and saved cards do the work; a dot's cloud browser may be blocked and fall back to the owner's browser with permission.
• The agent comparison post examines what actually lands in a retailer's order system across seven buying agents, including three that pay through the same Stripe wallet.

Read original · Original 2 · Original 3

If a personal agent acts for someone who has an account and a saved card, a blanket bot-block turns that person away at the door. The advice is concrete and checkable: make product data legible to agents, decide whether an agent would get through checkout, and treat the block-or-permit question as a policy choice with a customer consequence.

Describing OpenAI's DevDay 2026 launch of 'dots' — personal agents with their own cloud computer and browser that keep working after ChatGPT is closed — Dunn notes more than 4,000 app connections that are read-only when researching in the background, sign-ins to supported sites using saved passwords the model never sees, and custom rules to allow, require approval for or block specific actions.

• Dots can buy on a retailer's website with a saved card once the owner approves.
• OpenAI expects a dot's cloud browser to be blocked by some sites and lets it fall back to the owner's browser with permission.
• Dunn's conclusion is that retailers treating all agent traffic as bot traffic are making a decision about dot owners.
• The post ends with two questions: can a dot read your product data, and would one get through your checkout?

Read original · Original 2

It answers the newcomer question 'why does agentic commerce matter to a retailer this week?' in three concrete moves rather than a trend summary.

Dunn sets out three responses retail made in two weeks: Amazon blocked Meta's Muse on 20 September; Walmart, Best Buy, Sephora, Ulta Beauty and Wayfair signed on as Muse partners within days; and on 28 September Shopify let any browser agent complete a checkout on its stores once the shopper approves, with no merchant set-up.

• They argue the third matters most because most retailers joined it without deciding to — if you sell on Shopify, you are already in that camp.
• The piece also notes that blocking is harder to hold than it looks: a US appeals court found in August that when an agent shops, the shopper is the one accessing the site, so a block rests on terms of use and on spotting the agent at all.
• It closes by pointing to a comparison of seven agents that can already buy and what lands in a retailer's order system.

For brands planning a campaign

Useful for a first conversation with a retailer or brand that has not yet chosen a position on shopping agents. The three-way framing gives a simple diagnostic, and the legal point about terms of use is a practical risk to raise with whoever owns the site terms.

A reader deciding whether a headline affects them gets that discipline modelled on a real example.

Covering TikTok's 5 October newsroom post, Dunn lists Buy Direct (in-app checkout from the For You feed), Shopping Assistant (a conversational agent handling product details, shipping, sizing, availability and purchase in one conversation) and Agentic Leads, plus the TikTok for Business MCP connector now available in Claude, Kimi, Manus AI, Perplexity, Replit, Snowflake, Tencent WorkBuddy and the IAB Tech Lab Agent Registry.

• No launch dates or markets were given, so ask when these features will be available in your market before planning a booking.
• Dunn also flags that the announcement does not say how order data or returns are handled.
• The post reports that advertisers using the MCP connector for campaign activation and management are up more than 200% since launch, described as internal data for July to September 2026 with the base not disclosed.

For brands planning a campaign

A model for press-announcement diligence: which capabilities exist, which are promised, and which measurement questions remain open. Brands weighing TikTok commerce investment can reuse the same three questions before committing budget.

It is the clearest illustration of why retailer data and checkout readiness now matter, and it ends with two questions a reader can act on.

On OpenAI's DevDay 2026 launch of 'dots' — personal agents with a cloud computer and browser that keep working after ChatGPT is closed — Dunn describes the capabilities that touch commerce: more than 4,000 app connections, read-only in the background; shared context across ChatGPT, Slack and Teams; saved-password sign-ins; custom rules to allow, require approval for, or block actions; and the ability to buy on a retailer's site with a saved card once the owner approves.

• Dots sit on ad-free plans, so influence currently runs through connected apps and readable product data.
• OpenAI expects an agent's cloud browser to be blocked and lets it fall back to the owner's browser with permission — so, Dunn argues, treating all agent traffic as bot traffic is a decision about your own customers.
• The post closes with two questions for brands and retailers: can a dot read your product data, and would one get through your checkout?

For brands planning a campaign

A ready-made readiness checklist for an ecommerce or retail client: product data legibility to agents, checkout handling of agent traffic, and the policy choice to block or permit. Also a useful framing for content aimed at commerce teams assessing agent traffic.

A reader learns a reusable way to judge any new retail media network.

Dunn reports that McDonald's confirmed a media network at its 23 September Chicago investor day, with advertising on digital order boards running since August in about 450 company-owned US restaurants, roughly 3% of its ~14,000 US sites.

• The stated ambition is a $1bn business spanning app, kiosks, menu boards and restaurants.
• McDonald's has nearly 220 million 90-day active loyalty members across 70 markets.
• Dunn's own test is the load-bearing part: Amazon's ads work because ad and purchase sit in the same place, but unless a product is on the menu the sale happens elsewhere, so McDonald's media sits closer to screens at a petrol pump than to sponsored product listings — though the loyalty base could make it a very smart petrol pump.
• They also point to a McDonald's executive's comment about plugging into emerging conversational commerce channels without reengineering the core, and say buyers should ask how outcomes beyond impressions will be proved.

For brands planning a campaign

Gives a retail media buyer or brand a comparative frame for judging non-endemic retail media: where the loop closes, where it does not, and what proof of performance should be demanded. Also shows the franchise-versus-company-owned caveat that limits scale today.

Useful for anyone judging whether this is a sustained voice or a one-off commentator.

Included in the second edition of the 2026 RTIH Top 100 Retail Technology Influencers List, Dunn uses the post to state the core argument: retail technology spent a decade optimising the shelf and the next decade is about the shortlist, so when an assistant answers 'what should I buy', product data stops being an operations problem and becomes media inventory.

• They say Thrad is building ways for brands to appear inside those answers and for retailers to monetise conversational surfaces they run themselves, and mention writing The AI Commerce Brief on LinkedIn every week.
• The post also notes the list's spread — finance, social media, autonomous stores, RFID, last-mile, loss prevention and data platforms — and that the next edition lands in April 2027.

For brands planning a campaign

The single best quote to anchor a partnership pitch: one sentence connects AI answer surfaces, product data and retail monetisation. The weekly newsletter also gives a predictable channel and cadence for a content collaboration.

A reader weighing a partnership should know the voice is not only long analysis.

A short 'Friday ≈ Memeday' post built on a joke about no longer needing 47 open tabs to find the best deal, tagged with retail media, RMN, shopper marketing, grocery, FMCG, adtech, martech and agentic commerce hashtags, and naming a long list of other retail media and commerce practitioners.

• The substance is the joke and the community tagging rather than an argument.
• A companion post in the same format jokes about Muse's mascot.

For brands planning a campaign

Relevant if a partnership needs a lighter, more shareable touchpoint or a practitioner-network amplification post. Treat the meme format as reach and community, not as evidence of Dunn's analytical positions.

They argue the third matters most because most retailers took that path without deciding: any store on Shopify can now let a browser agent complete checkout once the shopper approves, with no merchant setup.

• Blocking is also harder to sustain than it looks: an August US appeals court found that when an agent shops, the shopper is the one accessing the site.
• Dunn compares seven agents that can already buy and what actually lands in an order system.

For brands planning a campaign

A brand or commerce vendor can use this three-way frame to brief internal teams on where agent traffic will come from and what it means for checkout. A sponsored edition or co-written comparison using the same structure is the most direct way to work with it.

Examples: Source 1 · Source 2 · Source 3

On OpenAI's dots, they cover the 4,000-plus app connections, read-only background research, sign-ins using saved passwords the model never sees, and opt-in access to the owner's own computer.

Working from primary releases — TikTok's newsroom post, OpenAI's DevDay materials, McDonald's investor day — keeps the vendor's own wording and numbers, then separates what's live now from what's announced direction.

For brands planning a campaign

A platform or vendor launch can be translated for a brand-side audience this way: expect a detailed, sourced teardown rather than a reaction to the announcement headline.

Examples: Source 1 · Source 2 · Source 3

At its investor day, McDonald's confirmed a media network live in 450 company-owned US restaurants, with other brands' ads on digital order boards after the customer orders, and an eventual $1 billion ambition across app, kiosks and menu boards.

• Dunn's read: the ad and the purchase do not sit in the same place, putting it closer to screens at the petrol pump than to Sponsored Products, unless the roughly 220 million 90-day active loyalty members change the maths.
• On TikTok, they cover the US ad network with nearly 400,000 apps, Search Ads powered by Smart+, and GMV Max now factoring in affiliate commissions, coupons and referral fees.

For brands planning a campaign

Useful for evaluating a retail media network or a brand weighing a non-grocery placement. The McDonald's analysis model — ask what actually closes the loop — applies to any network pitching new surfaces.

Examples: Source 1 · Source 2

Dunn writes that retail technology spent a decade optimising the shelf and the next decade is about the shortlist, and that product data stops being an operations problem and becomes media inventory. They state plainly that this is the logic Thrad builds around, making the thesis positioning for their own company too.

For brands planning a campaign

A brand weighing conversational commerce or an AI-answer placement gets a clear thesis to agree or argue with. A partnership should treat the Thrad link as commercial context, not neutral commentary.

Examples: Source 1 · Source 2

On TikTok's Shopping Assistant, they ask who holds the customer relationship and note that the announcement did not say how order data or returns are handled.

• On OpenAI's dots, purchases run through retailers' own sites, where logged-in accounts and saved cards do the work; a dot's cloud browser may be blocked and fall back to the owner's browser.
• The closing question: how does a platform attribute a sale that starts in a chat and finishes without a click?

For brands planning a campaign

A measurement, attribution or customer-data vendor is the strongest fit for co-produced work here; the questions are specific enough for a written exchange or panel rather than a general sponsorship.

Examples: Source 1 · Source 2 · Source 3

Posts follow a recognisable shape: a short thesis, emoji-led bullets on what was announced, a plain statement of what it means for brands and retailers, a "what I'll be looking for next" section, and a call to action.

Announcement summary, buyer consequence, open question, newsletter prompt — with figures and links kept attached to the news item they came from.

For brands planning a campaign

A recurring sponsorship or co-branded edition fits because the format and cadence already exist, while a comment-gated asset gives readers a measurable response step rather than impressions alone.

Examples: Source 1 · Source 2

One is about no longer needing 47 tabs to find the best deal; the other jokes about Muse's mascot, Jolly. Hashtags span retail media, RMN, trade and shopper marketing, grocery, FMCG and CPG, adtech, martech and agentic commerce.

These posts carry little analysis and heavy tagging, unlike posts about specific topics. They read as a community-building habit rather than an argument about a topic.

For brands planning a campaign

Not the right slot for a substantive sponsorship. Treat these posts as evidence of an engaged retail-marketing network around the same audience: useful for reach, but not for a considered brand message.

Examples: Source 1 · Source 2

TikTok gave no launch dates or markets, so this is a direction of travel rather than something a brand can book today.

• The 200% advertiser figure is internal data with no base disclosed; McDonald's franchisees, who run most of the US estate, are not yet in the pilot.
• Where they predict, such as dot traffic being blocked and falling back to the owner's browser, they label it as anticipation rather than fact.

For brands planning a campaign

A brand gets a cautious read that flags the open questions, which suits decisions that depend on what is genuinely available rather than announced intent.

Examples: Source 1 · Source 2 · Source 3

The posts explain what the business does but do not state a job title, founding role or equity stake. Ask Dunn to clarify their role and ownership.

View example

Dunn analyses the announcement as an industry development, not as a TikTok partner or sponsor.

View example

Dunn frames the launch around what it means for brands and retailers, asking whether an agent can read their product data and complete their checkout. No commercial tie to OpenAI is stated.

View example

Dunn covers the 450 company-owned US restaurants live since August, the $1bn ambition, the digital order-board format and the limits of the comparison with Amazon and Walmart. They analyse the strategy rather than working with the company.

View example

A second post jokes about Muse's mascot, Jolly. Both treat Meta's agent as a subject of commentary.

View example

They thank the publication and the list's sponsor, 3D Cloud.

View example

Both treat retail media as a platform problem rather than a pure advertising one, and both sit in the same retail-technology circle: Zollan's inaugural summit post tags Dunn alongside the Constructor team.

Zollan's posts are company-forward: a Magic Quadrant placement, advertiser and GMV growth numbers, and an event invitation for Constructor's Melbourne summit.

• Dunn's focus is news analysis — agent checkout, media-network economics and attribution — with open questions rather than vendor milestones.
• Zollan is the better read for what one platform claims; Dunn, for how an unfamiliar development changes a buyer's decision.

Read original · Original 2

Both argue that agent-driven discovery shifts the ground under brands, and urge readers to make product data legible to machines.

Sinclair writes tactical marketplace instructions for sellers, addressing Amazon and Walmart catalogue changes and closing with comment-gated breakdowns.

• Dunn writes news teardowns and strategic framing for retailers and brands.
• Read Sinclair for how to restructure listings; Dunn for what a launch changes about checkout, measurement and who holds the customer.

Read original · Original 2

Both write about AI moving from buzzword to real application in advertising and operations, grounding general arguments in what a brand can act on.

Toscano's material covers Amazon advertising craft — Q4 spend, DSP and AMC audiences, EU expansion, FBA and AWD — plus conference recaps for the seller community.

• Dunn's covers wider-platform analysis of agentic commerce, retail media economics and conversational selling.
• Toscano suits a brand asking how to operate on Amazon; Dunn suits one asking where agent-mediated shopping leaves it.

Read original · Original 2

The posts give an operator's account of building retail media on a commerce search and product discovery platform, the Gartner quadrant recognition for that platform, and specific figures — more than $120bn in ecommerce GMV flowing through it, 122% year-on-year growth in active advertisers, nearly 3,000 brands advertising, and 96% customer retention — plus the claim that the network beats legacy retail media on relevance, fill rate and ROAS.

• Zollan also speaks on retail media at Constructor's own retail summit, addressing retailers as a practitioner rather than a commentator.
• For a vendor selling the infrastructure beneath retail media, the framing — a network is only as good as its discovery engine — makes the same argument more directly than Dunn does.

The tradeoff is independence and currency.

• Both posts cover the employer's own platform and results, published in February and August 2026, giving readers a supplier's scoreboard rather than an outside read on how the market is changing.
• They say nothing about how to attribute an agent-initiated sale.
• There is no visible publishing cadence or comment-gated resource comparable to the comparison asset Dunn offers.
• A sponsor funds a platform narrative rather than the weekly analytical slot and repeatable reader-response mechanic they provide.
• If the campaign needs an external, critical voice on measurement, Zollan is the weaker choice.

Examples: Source 1 · Source 2

Both posts offer practical instruction.

• One ties the arrival of a high-profile open-source agent developer at OpenAI to what it means for consumer brands, arguing that when agents handle routine purchases, visibility moves from shelf placement to algorithmic preference, and brands must optimise how their products appear in AI recommendations and connect their commerce systems to agentic protocols.
• The other reads Walmart's tighter product-title rules and the Sparky shopping agent as a signal that clean, structured product data is the raw material agents need, and tells brands to get their highest-volume features into the title formula first.
• This is closer to campaign instruction than commentary, making it a fit for a vendor selling catalogue, feed or product-data tooling.

The tradeoff is urgency and a gated next step.

• Each post asks readers to comment a keyword for a full breakdown, making the public post a teaser while the substance sits behind the response.
• Dunn's posts carry the full argument and underlying questions in the post itself.
• Two posts, one from February and one from July 2026, offer less of a track record for judging a longer content programme.
• A vendor buying Sinclair gets tactical product-discovery advice and a lead-capture mechanic, but less independent scrutiny of platform announcements and less of the attribution economics a measurement vendor would want its partner to own.
• If the goal is to be seen as the company that understands where agent-driven sales are counted, ask whether Sinclair can address that question.

Examples: Source 1 · Source 2

Both posts come from that world: a recap of a London Amazon EU Accelerate meetup and a keynote at the Amazon Ads EU Business Accelerator.

• The advice is operational — treat Q4 advertising as audience-building rather than extra spend, use Amazon Marketing Cloud and DSP to see who actually buys, protect margin before cost-per-click spikes, and retarget into January — alongside cross-border expansion, AI in advertising and operations, and fulfilment changes such as FBA Select and AWD.
• A vendor selling to Amazon sellers and brands, or into the EU seller community, gets a speaker with agency leadership, a client base and an event platform where its brand could appear.

The tradeoff is subject fit and format.

• Both posts, from early October 2026, are event and sponsorship recaps rather than standalone analysis.
• A sponsor is buying presence in a room and a community, not a run of independent written pieces.
• The coverage is narrow to Amazon's ecosystem and EU expansion, a poor match if the campaign needs to reach retail media and commerce infrastructure buyers more broadly.
• Ask whether the content can address how agent-initiated purchases should be measured and who holds the customer relationship — questions a measurement vendor wants its name attached to.

Examples: Source 1 · Source 2