AI may replace the CRM interface before the platform

AI may replace the CRM interface before it replaces what the CRM knows

AI assistants such as ChatGPT and Claude could displace the CRM marketing platform from one of its most visible roles: the place where marketers work. A marketer can describe an objective in plain language and let an assistant start the task. That changes where work happens. It does not establish that the assistant can reproduce the customer data, decisioning, and accumulated marketing judgment previously reached through the platform UI.

This distinction matters for executives deciding what to build, integrate, or renew. Pini Yakuel, founder and CEO of Optimove, argues that assistants will work with CRM marketing platforms rather than make them unnecessary. Optimove sells such a platform and benefits commercially if companies continue to need the category, so its claims require that context. The broader executive question is whether AI has replaced a capability or simply created a new way to reach it.

The answer affects architecture and procurement. A company can remove an interface while retaining the data, rules, models, and execution systems behind it. It can also reproduce those functions elsewhere and reduce its dependence on the incumbent platform. AI makes those decisions easier to separate.

The new division of labor: reasoning on top, marketing substance underneath

One possible architecture has three layers with different jobs. The assistant interprets a marketer’s request, reasons about the task, and produces an answer or action. An integration layer connects that assistant to external systems. Marketing infrastructure then supplies the customer information and decisioning needed to turn the request into campaign choices.

One mechanism for the middle layer is the Model Context Protocol, or MCP, a standardized way for assistants to access external tools and data. MCP can make that access more consistent, but the information and rules still live somewhere outside the assistant. For example, an assistant can ask an integrated system which customers appear likely to churn when that system can provide the relevant customer histories and logic.

Optimove’s position is that its platform supplies customer data, decisioning, and an accumulated record of what works, while the assistant supplies reasoning and conversation. Under that model, a marketer could work through ChatGPT, Claude, or another assistant while marketing-specific functions remain in the systems underneath. Because Optimove sells the underlying platform, this description also supports its commercial case for remaining part of the stack.

This architecture gives executives a more precise way to ask whether AI “knows” the customer. A model can interpret a request and reason over retrieved information while customer history and campaign rules remain in external systems. The result may also depend on prior organizational decisions about eligibility, offers, testing, channels, and timing. The visible AI response can therefore reflect work performed across several systems.

Direct-to-data AI makes build versus buy harder

Technical teams can connect an assistant directly to company data and build the required tools around it. Yakuel concedes that in-house systems can work for some use cases and describes teams pursuing this path as early rather than mistaken. That makes the decision a build-versus-buy assessment. Each company needs to determine which functions it can reproduce and govern itself.

Audience creation shows why that assessment extends beyond database access. A query can identify people who meet specified conditions, while a production workflow may also include exclusions, treatment choices, an optimization objective, and test and control groups. Those choices determine who receives a campaign and what the company can learn from the result. Any replacement architecture has to represent and govern the choices the business requires.

Yakuel argues that these decisions reflect years of marketing practice that a general-purpose model has yet to absorb well enough to execute unsupervised. He predicts that sufficient autonomous marketing judgment is “years from now, not months.” This is Optimove’s view, and Optimove benefits if customers continue to rely on specialized marketing infrastructure. Executives can address the architecture question without accepting that timeline: identify the decisions their own system must make and decide where those decisions will reside.

“Connect the model to our data” is only part of an architecture specification. In the audience example, customer records may be insufficient if eligibility rules, offer suppression, campaign priorities, or experiment design sit in other systems or processes. That knowledge can live in platform configuration, company code, operational processes, or employee judgment. Removing a platform requires a company to know which of those dependencies move with it.

The same test applies to the incumbent platform. Long use does not establish that valuable judgment has been captured there in reusable form. A business can inventory the data, rules, models, campaign histories, and controls that depend on the platform, then determine which can be moved or reproduced. That inventory gives the build-versus-buy decision a concrete basis.

Customer data can live outside a CRM marketing platform, and organizations can build decisioning and encode domain knowledge in systems they control. Future models may also become capable of making more marketing decisions autonomously. The near-term architecture question is narrower: if removing a platform removes audience logic, customer histories, experimentation rules, or campaign decisioning, the replacement design must provide those functions somewhere else. Their replacement cost belongs in the build decision.

Optimove’s examples show how assistants can coordinate existing systems

Optimove describes a campaign-drafting example in which a team gives an AI assistant a weekly brief and receives a full week of campaigns: eight campaigns across two brands in one 90-minute session. The 90-minute figure is an Optimove example, so it should be read as a vendor-reported result rather than a general productivity benchmark. The architectural point is clearer: the assistant turns a brief into campaign work while drawing on marketing infrastructure underneath it.

A second Optimove example is always-on QA. Optimove says a scheduled agent can check each morning for journeys, templates, and campaigns expiring during the coming week, flag missing or broken items, and open tasks before employees arrive. Routine inspection moves to an automated process in this workflow. The process still relies on campaign state exposed by underlying systems and on rules defining what counts as missing, broken, or ready for escalation.

The third Optimove example expands from individual tasks to end-to-end retention work. Optimove describes a framework that researches retention topics, combines that research with company platform data, recommends approaches, helps construct campaigns, performs post-campaign analysis, and feeds what it learns into subsequent work. The assistant coordinates several stages. Customer information and marketing decisioning remain inputs throughout.

Yakuel says campaign drafting, daily automated QA, and end-to-end retention are “in use right now.” That statement represents Optimove’s account of current use and supports its commercial argument for an AI-accessible CRM platform. The examples share a technical pattern: an assistant operates over information, rules, and execution capabilities supplied through other systems. Executives can test that pattern against their own workflows without treating the examples as proof of market-wide results.

Optimove also argues that AI can let small teams execute work associated with much larger teams because assistants can continuously handle planning, checking, and coordination over the platform. This is another vendor claim with a direct commercial incentive behind it. For an individual company, the test is operational: identify the tasks that can be delegated safely, then compare the resulting workflow with current staffing and controls.

CRM platforms face a visibility and defensibility test

AI integration can create a commercial tension for platform vendors. A user may invoke a platform’s data and decisioning through an assistant while spending less time in the platform interface. If a marketer asks an assistant to build an audience, check campaigns, or analyze retention performance, the assistant becomes the visible place where the task occurs. The platform’s value then depends more on the capabilities the assistant invokes than on direct interaction with its screens.

That can make UI engagement a weaker proxy for customer value in workflows initiated through integrations. Vendors may instead need to demonstrate the value of data access, decisioning, execution, permissions, and machine-readable interfaces. Buyers can apply the same test during procurement: determine which underlying capabilities they actually use, which business processes depend on them, and what reproducing them elsewhere would require.

For Optimove and other CRM platform companies, this is also a defensibility test. If conversational interfaces become a common access layer, interface design may account for a smaller share of differentiation in those workflows. The remaining case depends on the assets behind that access, including customer context, decisioning, operational logic, and reliable execution. Optimove has a commercial reason to argue that these functions will remain valuable, while buyers have a reason to test which are genuinely specific to the platform.

That division of labor can change as models and internal systems improve. Yakuel acknowledges that in-house approaches can work for some cases while maintaining that general-purpose models are not yet ready to exercise sufficient marketing judgment unsupervised. Companies can progressively move customer data, decision rules, experimentation logic, and marketing knowledge into architectures they control more directly. Each move changes what the specialized platform must contribute to remain defensible.

Key highlights

• Separate the interface from the capability: AI assistants may become the place where marketers work while CRM platforms continue supplying customer data, decisioning, rules, and execution. Technology buyers should evaluate each underlying capability independently when deciding what to build, integrate, or renew.
• Map the marketing architecture: AI assistants can handle conversation and reasoning while integrated systems provide customer context and marketing logic. Architecture owners should document where customer history, eligibility rules, campaign decisions, and domain knowledge reside.
• Price the full build-versus-buy decision: Direct AI access to company data does not reproduce audience logic, exclusions, experimentation, or campaign governance by itself. Internal platform teams should inventory these dependencies and include their replacement and operating costs when evaluating an in-house CRM architecture.
• Test AI workflows against real operations: Optimove reports AI use cases spanning campaign drafting, automated QA, and retention workflows, with the assistant coordinating capabilities supplied by underlying systems. Marketing operations teams should identify similarly bounded workflows, define required controls, and measure results in their own environment.
• Evaluate CRM platforms beyond their interfaces: As AI assistants absorb more user interaction, CRM value may shift toward customer context, decisioning, execution, permissions, and machine-readable access. Procurement and architecture teams should test whether those capabilities remain differentiated and costly to reproduce elsewhere.