What an Agentic Commerce Agency Actually Does (And What You Should Be Paying For)

By Stephen Honight, Founder of Lmo7

Most brands hearing the phrase “agentic commerce” for the first time in 2026 are doing one of two things. They are either getting pitched a tool wearing a strategy hat, or they are reading a thought-leadership piece that defines the category without ever explaining what the work looks like in practice.

This piece is the opposite. It is what an agentic commerce agency actually does, what you should expect to pay for it and how to work out which shape of engagement fits your business. Written for founders, ecommerce directors, marketplace managers and transformation leads who are pricing this category for the first time.

A definition in plain English

Agentic commerce is the shift from search-led shopping to AI-mediated shopping.

People used to type a query, scan ten links and pick a product. They are increasingly asking an AI assistant a question and acting on its answer. ChatGPT, Gemini, Claude, Perplexity, Google AI Overviews and Amazon Rufus are now the discovery surface for a growing slice of consumer decisions. Behind the scenes, AI agents will start booking, comparing, restocking and buying on behalf of users. That is the agentic part.

An agentic commerce agency helps brands win inside this layer. The work is part visibility, part content engineering, part marketplace optimisation and part capability transfer. The output is commercial: more brand mentions in AI answers, more recommendations from AI shopping assistants, better PDP performance on AI-mediated marketplaces and a team that knows how to keep this running.

It is not generic SEO. It is not generic ecommerce strategy. It overlaps with both but the mechanism, the metrics and the timeline are different.

The four real workstreams

When an agentic commerce agency is doing its job, the work sits across four pillars. Any agency that only covers one or two of these is doing a slice of the problem.

The first pillar is AI search visibility. Share-of-model analysis, tracking how a brand appears across ChatGPT, Gemini, Claude, Perplexity and Google AI Overviews, citation source optimisation, answer engine optimisation and the entity work that makes a brand legible to AI engines. Output: a brand goes from absent or inconsistent in AI answers to consistently mentioned with the right framing.

The second pillar is marketplace AI. Rufus is the obvious one, now connected to Alexa for Shopping. The work is SPN-mapped product detail page content, structured attributes, Q&A seeding, A+ Content tuned for AI extraction, parent-child hygiene, Buy Box recovery and the operational backbone underneath. Output: products surfaced in Rufus answers and recommended in the AI shopping flow on Amazon.

The third pillar is content for AI commerce. Blog content, product descriptions, brand narratives and authority-building assets, all written to do two jobs at once. They have to land with a human reader. They also have to be clean for AI retrieval. Most agencies still write for one audience or the other. The work is to do both without making the content sound like SEO bait.

The fourth pillar is agentic enablement. Workshops, training, capability transfer and the structural changes a client team needs to keep this running internally. This matters because the platforms move quickly and a brand that depends entirely on external delivery will lag the curve. Output: the client team understands what is changing, why it matters commercially and what they own.

Cutting any of these out simplifies the proposal but undersells the work.

Pricing as a buyer signal

The most useful question to ask early is what a credible engagement actually costs. Two reasons. It tells you whether the agency is honest about scope, and it lets you map your budget to the right shape of engagement.

At Lmo7 we structure the offer in three commercial tiers. The shape is deliberate, because the right kind of work for a sub-£1m founder is not the right kind of work for a multi-brand portfolio.

The lightest tier is DaaS, Data as a Service, from £250 plus VAT per month per marketplace. Direct API access across Amazon, Meta, Google and Shopify, a 60 to 90 minute onboarding session, a quarterly upskilling refresh and light async interpretation support. The client runs the day-to-day work. It is the tier we recommend when Challenger will not fund or when a founder wants to start somewhere serious before committing to hands-on engagement.

The default sell is Challenger. Hands-on management of Amazon and Agentic channels, modular across two stacks. The Amazon Stack covers Retail Ops, Content Optimisation for Search Visibility and Ads, available individually from £750 a month or as a Full Amazon Stack at £2,500. The Agentic Stack covers Agentic Ops, Tracking and Content Optimisation for AI Search Visibility, and Ads, with a Full Agentic Stack at £2,500. The combined Amazon and Agentic Full Stack is £5,000. This is the tier for founder-led brands in the £1m to £50m band that need execution and commercial momentum.

The top tier is Enterprise. Workshops, insight sessions and multi-brand tracking for organisations with portfolios and stakeholder weight. Workshops start at £5,000. Tracking pilots sit at £5,000 to £10,000. Ongoing multi-brand tracking and recommendations retainers sit at £20,000 to £80,000 a year. The client executes through internal teams. Lmo7’s role is to diagnose, educate and prioritise.

If an agency is quoting Enterprise scope at Challenger prices, the work will arrive thin. If they are quoting Challenger pricing for what is really a DaaS need, you will be paying for hands-on delivery you do not have capacity to absorb. Pricing is not a vanity number. It tells you whether the agency understands their own offer.

What an agentic commerce agency actually delivers in the first 90 days

This is where most pitches go quiet. Worth being concrete.

In month one the work is diagnostic. A visibility baseline across the AI engines that matter to the category. A bots and reachability audit (OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot, Google-Extended, Applebot-Extended). A content audit graded against AI extraction, not just SEO. If Amazon is in scope, a catalogue health check including parent-child integrity, suppressed listings, Buy Box posture, SPN coverage and review density. The deliverable at the end of month one is a prioritised action plan with owners. Without that, nothing in month two has direction.

In month two the work shifts to execution. PDP rewrites grounded in real search query data and AI shopping language. Entity schema (Organization, Service, Product, Article) deployed where it matters. The first round of content engineering on product, category and brand pages. A first version of share-of-model tracking that becomes the baseline for everything that follows. If Ads is in scope, the first weekly optimisation loop on Sponsored Products, Brands and Display.

In month three the work compounds. Version-over-version tracking shows what has moved. The action plan refreshes. Content additions roll in. Citation source work begins (the slower track that pays off over months six to twelve). The agency should be running monthly review calls with a clear two or three priorities for the next month, not 30-slide decks of generic findings.

This is what real delivery looks like. If the pitch is heavier on tooling than on this kind of rhythm, you are paying for a platform tour, not an agency.

Who we do this with

Worth saying who this work is actually for.

Lmo7 works across the full ladder. At the Enterprise end that means multi-brand portfolio businesses like Brown-Forman and Haleon, where the work is strategic, layered for different stakeholders and executed through the client’s internal teams. At the Challenger end it means founder-led consumer brands like Trip Drinks and Veloforte, where we run the work hands-on and the job is commercial momentum on Amazon and in AI search.

The shapes differ. The underlying logic is the same at every tier. Diagnose, prioritise, execute, track, repeat.

What an agentic commerce agency cannot fix in 90 days

This is the section most agency pitches skip. Worth being upfront.

Domain authority does not move in a quarter. Backlinks, third-party citations, expert quotes, review density and category authority signals are the highest-impact lever for AI citation, and they are also the slowest. If your brand is starting from a low domain rating, the visibility curve in months one to three will be driven mostly by content and structure fixes. The authority track runs in parallel, paid by a different budget, often outside the agency’s main scope. We are honest about this because clients who believe content fixes will solve the visibility problem feel let down by month four.

Category positioning does not move in a quarter either. If your brand has been confused with a competitor by AI engines because the entity disambiguation is weak, the schema and content work shifts the dial, but the underlying authority signals take longer to clean up. We have seen this with brands whose name overlaps with a generic category term. The work is patient.

And the underlying product story still has to be commercially compelling. If the brand promise is unclear to a human reader, AI engines will reflect that confusion in their answers. Agentic commerce work amplifies what is already there. It does not invent the brand.

A decision tree for picking the right tier

A quick guide for figuring out which tier of engagement fits.

If you are a founder running a brand under £1m in revenue, or you have a lean team with capacity to action work but no data, DaaS is the right tier. Direct API access. Onboarding session. Light interpretation support. The pricing is built for this. Anything heavier will be overscoped for the size of the operation.

If you are a single-brand consumer business in the £1m to £50m band, founder-led or with a lean ecommerce team, and you want execution rather than strategy decks, Challenger is the default. Pick the modules that match the bottleneck. Most brands start with the Full Amazon Stack or the Full Agentic Stack and add the other one as the engagement matures.

If you run a multi-brand portfolio, or you are a transformation lead inside a £100m-plus business, Enterprise is the right entry. Workshop or insight session first. Tracking pilot after that. Then a recurring tracking and recommendations retainer. The work is consultative, not delivery, because the client team is the one executing through their existing structure.

If none of these fit cleanly, the honest answer is usually Challenger to feel out the scope, with the option to step up or down once the picture is clearer.

So what should you do next

If you are pricing an agentic commerce agency for the first time, the first useful step is to baseline your brand. Run a share-of-model on your top category prompts. Audit your bots layer. Check what Rufus says when asked about your product against your category. That is two or three days of work and it tells you where you actually sit. Most brands are surprised by what they find.

If you are choosing between agencies, ask each one how they would structure month one, month two and month three for your brand. Ask what they would and would not change. Ask what they would expect you to fund outside their scope. Vague answers signal a vague engagement.

If you want a short version of all this, the agency that is worth paying knows the mechanism of how AI engines build answers, has a clear view on what the work moves and what it doesn’t, prices honestly against the scope, and ends every deliverable with two or three priorities and an owner. The agency that is not worth paying will give you a tool tour and a deck.

If you would like to see what this looks like on your brand, we can run a baseline and an opening action plan inside two weeks. The follow-on path is built in, so by the end of it you know whether you want Lmo7 running the work, or whether you want the data and the training and to run it yourselves.

Either way, the worst version of this is doing nothing. AI-mediated discovery is already shaping consumer decisions in your category. The brands that show up first will set the recommendation default for the next cycle.

Stephen Honight is the founder of Lmo7, the AI-native agency helping challenger consumer brands win in AI-powered discovery and agentic commerce. Lmo7 works across Amazon, D2C and AI search surfaces with brands including Trip Drinks, Veloforte, Brown-Forman and Haleon.