Part of marketers’ work still consists of moving between screens, building queries and manually searching data for answers. New tools make it possible to ask an analytical or operational question in natural language and let the system perform the necessary steps. This shortens the path from question to answer, but at the same time increases the importance of permissions, correct interpretation of data and verification of the result.
Marketing technology was long built around screens, menus, forms and query languages. Users had to learn where in the system to create an audience, where to modify a customer journey and how to obtain an answer to an analytical question. Salesforce describes a shift toward controlling these functions through natural language.
The underlying problem is operational overhead. A marketer may arrive at work intending to launch or evaluate a campaign and spend several hours copying settings, moving through screens and waiting for an analyst to prepare the necessary query. The activity itself may add no marketing value.
An AI-based interface can shorten this intermediate layer. A marketer can, for example, ask how many people completed a particular customer journey, at which step they most often drop out or how a particular segment behaves. The system itself converts the question into the operations needed on the data.
From a practical perspective, this is a bigger change than merely a simpler user interface. The person no longer sees every step through which the system arrived at the answer. They therefore have to place more trust in data definitions, permissions and the AI’s correct understanding of the meaning of the question.
Salesforce recommends starting with a journey or campaign the marketer already knows well so that the correctness of the answer can be verified. This principle is broadly applicable: the first deployment of an automated analytical assistant should not be on a problem whose correct result nobody can assess independently.
Before broader deployment, prepare a test set of real marketing questions together with manually verified answers. Monitor not only whether the numbers are correct but also whether the system correctly understands internal terminology. “Active customer,” “qualified lead” or “completed journey” can have a different definition in every company.
The second area is authority. If natural-language instructions can be used not only to analyze but also to change campaigns, segments or automated scenarios, the company must distinguish among a query, a proposed change and the actual execution of that change. Higher-risk operations should retain a separate human confirmation step.
Marketing can thereby shift from knowledge of a particular system interface toward more precise formulation of the problem. This does not mean subject-matter expertise loses value. On the contrary: the easier it is technically to ask a question, the more important it becomes to recognize whether the answer makes business and data sense.
KEY TERMS
- Natural-language control: Assigning analytical or operational tasks to a system through naturally phrased requests.
- Data semantics: The precise definition of the meaning of individual metrics, segments and internal terms.
- Validation set: A set of known questions and correct answers used to test an AI system.
- Separation of proposal and execution: A rule under which AI may prepare a change, but a more significant intervention must be confirmed separately by a person.