RetailCRM has announced Release 12, titled 'Retail with AI,' with a presentation scheduled for October 29, 2026, in Moscow at 'Soglasie Hall.' The key innovation is the 'three-atom model,' which unifies sales, marketing, and service in a single CRM around a shared customer history, along with AI agents for routine inquiries whose responses are sent by a specialist. New features will only become available after the release, so this is currently a vendor statement about development direction rather than a finished product.

What happened

RetailCRM announced Release 12, titled 'Retail with AI,' which will be presented on October 29, 2026, at 'Soglasie Hall' in Moscow, located at 36s1 Profsoyuznaya Street. The core of the update is the 'three-atom model,' unifying sales, marketing, and service in a single CRM around a shared customer history. Sales can see a buyer's open inquiries before offering them a product; marketing automatically suppresses promotional campaigns for customers with unresolved complaints; and service receives full context in a single ticket. The release will also include AI agents for routine inquiries and a new customer service module that aggregates inquiries with purchase and campaign history. The program features a panel with representatives from Hoff, Askona, Familia, HENDERSON, Lemana PRO, and Geltek, as well as talks by Dmitry Borodin (CEO of RetailCRM), Ilya Salikhov (CTO), and Evgeny Butman (ECS Group).

Context

The underlying problem around which the release is built is formulated by the vendor itself: for an AI agent to work with customers, it needs to know the buyer's history and business rules. For agentic systems, this means that an agent's usefulness is limited by the completeness and accuracy of the context at least as much as by the quality of the model. Therefore, the 'three-atom model' essentially describes a unified context layer where a shared customer history serves as the common memory for all departments. The second element of the framework is a scheme where the agent only prepares a response, while the specialist makes the decision and sees what the agent is doing: this is a classic Human-in-the-loop with observability, reducing the risk of errors when deploying automation to production. This design explains why the announcement focuses on specialist support rather than replacing people: the agent is embedded in processes as a drafting tool with observable behavior.

Why this matters for the industry

For the industry, the signal is more important than the features themselves: a CRM vendor is normalizing Human-in-the-loop as the default architecture for agentic automation in retail, rather than full autonomy, and embedding agents into familiar processes with a division of roles between humans and machines. This has implications for founders and builders of agentic systems: it is becoming increasingly difficult to sell 'autonomous agents' directly to retail, as platforms are turning agents into a feature of their ecosystem and capturing the most valuable resource of the agentic era—the unified customer context and business rules. The combination of 'shared customer history plus human control' is already being used as a ready-made reference for in-house pilots. The forecast is as follows: if the release reaches production and public case studies with metrics appear, agentic support with a human having the final say will likely become the norm for the mid-market retail segment. Competitors will have to respond with their own versions of a 'shared history,' and competition will shift from model selection to the quality of customer data and the formalization of business rules. This forecast is an interpretation, not a confirmed fact.

Why this matters for users

If you work in retail, e-commerce, or marketing, the announcement is useful right now as a framework for your own decisions, rather than as a finished product. Stores on RetailCRM should consider watching the October 29 presentation to see the agents in action and understand where the agent's zone of responsibility ends and the specialist's begins. Those choosing a CRM or preparing their own agent pilots should compare the three mentioned patterns with their own funnels and backlogs: do all departments have a unified customer history, does the system operate in a draft mode, and does promotional communication stop when an inquiry is open. These points also become a list of questions for any CRM agent vendor—starting with latency and cost. It is also useful to study RetailCRM's approach to unifying data from different departments and to human control.

What is currently unknown / limitations

The event on October 29, 2026, has not yet taken place, so new features are unavailable before the release, and their performance in real-world operation is unconfirmed: the announcement declares capabilities but does not prove them. The materials contain no measurable metrics—no deflection rate, no percentage of agent responses accepted by operators, no processing time, no response quality, and no comparison with a scenario without an agent. Therefore, claims about agent capabilities should first be viewed as marketing and only then as technical evidence. The automatic stopping of promotional campaigns is technically a data-based rule, not a model, and its value depends on the accuracy of complaint tagging and their statuses. Finally, conclusions that the proposed scheme will become an industry default are interpretations that can only be verified by the first user data after the release.

Sources

Author

Look at AI, editorial team