What Delivery Hero shipped
Delivery Hero has launched an agentic AI assistant aimed at small restaurant and shop owners across its delivery platform, giving them a way to automate business operations and growth tasks without a dashboard, an app or a training session. Owners interact with the assistant directly through WhatsApp, using text, voice notes or images, the same channel many small merchants already use for daily communication with staff, suppliers and customers. The assistant identifies operational issues such as underperforming dishes, optimal timing for promotions, and customer reviews that require a response.
The system operates on a spectrum of autonomy rather than a single mode: it can surface a suggestion for the owner to approve, or it can act independently on categories of changes the owner has pre-authorized. Othman, who runs Bahraini Kabab in Dubai, described the practical value in terms of attention rather than automation for its own sake, saying the tool 'spots things I'd miss' while he retains control over what actually changes in his business.
Why the channel matters more than the model
The most consequential design decision here is the choice of WhatsApp as the interface, ahead of anything about the underlying AI model itself. Small restaurant and shop owners running single-location businesses typically do not have a back office team, do not log into a separate analytics dashboard daily, and often manage their entire operation from a phone between service rushes. A tool that requires a new app, a new login and a new habit competes against zero adoption for owners with no spare attention to give it.
Meeting merchants inside a messaging app they already use dozens of times a day removes the single biggest adoption barrier for SMB software: behavior change. Delivery Hero's own scale numbers, more than 40,000 restaurants onboarded already against a base of roughly 1.5 million partners, are only meaningful if usage translates into owners actually acting on the assistant's suggestions rather than ignoring another notification. Channel choice is what determines whether that translation happens, far more than any model capability improvement would.
The autonomy design worth studying
Delivery Hero's decision to let owners pre-authorize categories of autonomous action, rather than requiring approval for every single change, is the harder and more interesting design choice than the assistant's underlying capabilities. Full autonomy with no approval step risks a merchant discovering a pricing change or promotion they never intended. Approval required for every action defeats the purpose of automation for an owner with no time to review suggestions constantly. Delivery Hero's middle path, owner-defined autonomy boundaries by category, is the pattern most enterprise agentic deployments will eventually converge on as well.
That graduated autonomy model is directly transferable to enterprise contexts even though this specific product targets small merchants. Any organization deploying agentic AI internally faces the identical design question: which categories of action can an agent take unsupervised, which require human sign-off, and how does a user adjust those boundaries as trust in the system builds over time. Delivery Hero's SMB product is effectively running that experiment at a scale, 1.5 million potential users, that most enterprise deployments will never reach, making its usage data worth watching regardless of company size.
What this means for agentic commerce beyond delivery
Delivery Hero's assistant sits squarely inside the broader agentic commerce trend of 2026, but its target user, a single-location restaurant or shop owner with no technical staff, is a meaningfully different buyer than the enterprise CIOs most agentic AI vendors are pitching. That gap matters because SMB-focused agentic tools have to prove value inside days or weeks of unsupervised use, with an owner who will abandon the tool immediately if it produces a bad suggestion, unlike an enterprise deployment with a change management process and a dedicated implementation team absorbing early friction.
For platform operators, marketplaces and vertical SaaS vendors serving small merchants generally, grocery, retail, services, Delivery Hero's approach is a template worth studying regardless of industry. An agentic layer that lives inside an existing communication channel, offers graduated autonomy, and demonstrates value through specific, checkable suggestions rather than abstract insights is a far more realistic adoption path for time-poor small business owners than another standalone analytics product competing for their limited attention.
The scale test still ahead
Forty thousand restaurants is a meaningful early cohort, but it represents under three percent of Delivery Hero's roughly 1.5 million total partners, meaning the real test of this product is still ahead: whether the assistant's suggestion quality holds up as it scales to businesses with far more varied menus, languages, regulatory environments and customer bases than the initial rollout cohort. Agentic systems tend to perform well in curated early deployments and reveal edge cases only once usage broadens past the initial reference group.
The metric worth tracking over the next two quarters is the ratio of suggestions accepted versus ignored or reversed, and whether Delivery Hero publishes that data, well ahead of the vanity metric of additional merchant sign-ups. A platform genuinely improving merchant outcomes should be willing to share acceptance and reversal rates as proof; a platform focused primarily on adoption headlines may not. That distinction will tell operators evaluating similar agentic tools for their own merchant or partner networks whether this category is delivering real operational value or simply generating usage statistics that look impressive without checking the deeper number.
