$11.2 million is what investors now think it is worth charging brands to be recommended by a machine. Azoma, a four-year-old startup with 40 employees in London and Toronto, closed a Series A round of that size led by Puma Growth Partners, it told Business Insider’s CMO Insider on Wednesday. The product is a service Azoma calls agentic commerce optimization, or ACO: adjusting a brand’s listings, content and data so that AI shopping assistants — ChatGPT, Google’s Gemini, Amazon’s Alexa for Shopping, Walmart’s Sparky, Meta’s Muse, TikTok’s new Shopping Assistant — surface that brand’s soap or dog food when a user, or the user’s agent, asks what to buy.
The customers are the companies whose margins depend on shelf position, except the shelf is now an answer: Unilever, L’Oréal, Reckitt, Mars and Heinz. Azoma charges per brand, per geography and per shopping agent, so its revenue scales with the number of bots a brand worries about.
From SEO to GEO to ACO
The lineage is search engine optimization. When consumers typed queries into Google, an industry existed to game the ranking. When they began asking chatbots, that industry rebranded as generative engine optimization, or GEO — Azoma’s original business when it was founded in 2022. ACO is the next step in the same line: the buyer is no longer the person asking the question but the software that asks it, gathers prices and reviews from forums, editorial sites and the brand’s own pages, and in some cases completes the purchase itself. Personal agents like Meta’s Muse, OpenAI’s Dots and Instinct can now shop and check out on a user’s behalf, and at least a dozen retailers have launched their own assistants in the past year, per Azoma’s count. Each assistant is one more surface on which a brand can be invisible.
CEO Max Sinclair’s pitch for why ACO is a separate company rather than a feature of a GEO platform is that generic optimization tools “don’t understand commerce at all”: “They don’t understand the integrations, they don’t understand the workflows, they don’t understand the existing tech stack, they don’t understand how the teams work, how they’re operated, how they’re motivated.” That is a sales argument, not a technical one — nothing in the deck describes a mechanism a generalist could not build — but it is the argument that justifies the per-agent fee.
The mechanics Azoma actually describes are unglamorous: honing marketplace listings, optimizing a brand’s presence on YouTube and Reddit — two sources large language models lean on heavily — and a “digital twin” system that simulates consumer demographics so a brand can check how an agent presents its product to the shopper segment it wants. The planned Dunnhumby integration is the most concrete piece: Dunnhumby, the Tesco-owned retail analytics firm (and an investor in the round, through Dunnhumby Ventures, alongside Twinpath, Mars IAF, Rank Ventures and Ignite Ventures), holds brick-and-mortar sales data, which would let Azoma test whether share of voice inside ChatGPT correlates with units moved in physical stores. That correlation is the load-bearing claim of the whole business, and the integration to measure it does not exist yet.
What the deck claims
A pitch deck is marketing, and Azoma’s should be priced accordingly. Still, the numbers it chose are worth laying out, because they are the numbers brands will be quoted in sales meetings for the next year.
The market-size figure is Morgan Stanley’s December estimate that AI-powered shopping assistants could handle up to $385 billion in US e-commerce sales by 2030 — a projection, with “up to” doing arithmetic work. The consumer-behavior figure is 2025 Capgemini research saying 58% of consumers have replaced traditional search with generative AI tools for product recommendations; survey data, self-reported.
The platform figures are from earnings calls, which makes them auditable. Amazon CEO Andy Jassy said in July that more than 350 million customers used Alexa for Shopping over the past 12 months, and that US users of it spend 40% more per order than non-users. Walmart CEO John Furner said in August that Sparky’s user count grew 70% year-on-year and that its users also spend 40% more per order. Both 40% figures are comparisons between self-selected groups — people who adopt a shopping assistant may simply be bigger spenders — so they describe correlation, not lift caused by the assistant. Sparky’s 70% growth has no disclosed base; 70% of a small number is a small number. Shopify president Harley Finkelstein said on an August earnings call that traffic and orders from AI channels to Shopify stores tripled in a year, and a May Shopify blog post claimed AI-referred traffic converted at a 50% higher rate with a 14% higher average order value than traditional search. Shopify, note, sells the storefronts that benefit if you believe this.
The deck also cites Cloudflare data showing automated bots now exceed humans in web traffic — a statement about the web as a whole, not about shopping — and notes that in September, Meta’s Muse was the top app in the US Apple and Android app stores, a ranking that measures downloads, not purchases made.
One case study, no methodology
The deck’s headline proof point is a 43% revenue uplift for Mars versus a control group. That is worth more scrutiny than the platform statistics, because it is the only number that describes what Azoma itself did rather than the weather around it. It comes without a stated time period, category, baseline or method for constructing the control, and it is a single client which, not coincidentally, has its investment arm in the round. It could be real; a well-chosen control group is standard practice in retail analytics, and Dunnhumby’s participation suggests Mars at least has access to people who know how to build one. But a vendor-reported uplift for a client-investor should be treated as an advertisement until the methodology ships.
Azoma plans to spend the round on tracking new agents and on headcount — Sinclair estimates 80 to 100 employees next year, up from 40, across sales, customer success, AI research, software development and operations. That is roughly $112,000 to $140,000 of new funding per current employee, a normal Series A ratio for a company whose asset is a sales relationship with packaged-goods buyers rather than a model.
The open question is whether the fee holds. The agents Azoma monitors are owned by companies that could sell placement directly, and Amazon and Walmart already do, in the form of sponsored listings inside their assistants’ answers. If agent-makers decide an answer engine and a checkout are the same funnel, charging both sides becomes their business, not Azoma’s. Two agents launched by TikTok or Meta next quarter would double Azoma’s addressable complexity; two closed recommendation APIs would halve it. Watch which arrives first.