ChatGPT Work launches an agent for your business data

ChatGPT Work launches an agent for your business data

OpenAI expands ChatGPT’s role in organizations

OpenAI wants to take ChatGPT a step further in the enterprise. In an announcement published under the title “Now everyone can put data to work”, the company introduces a data agent in ChatGPT Work. The stated goal is to enable teams to query professional data using natural language, then derive analyses and interactive dashboards from it without having to write complex queries themselves.

The move is significant because it shifts ChatGPT’s center of gravity. The tool is no longer viewed solely as an interface for writing, summarization, research or general assistance. It is also becoming a conversational gateway to the data that structures a company’s operations: sales, operational, financial and marketing data, as well as information from business applications. OpenAI is therefore seeking to position ChatGPT Work in a role closer to that of a daily decision-making interface.

The very name ChatGPT Work reflects this ambition. Since ChatGPT’s public launch at the end of 2022, OpenAI has gradually expanded its product to professional uses, with offerings and capabilities aimed at organizations. The initial logic was largely focused on generating and transforming text: drafting documents, coding assistance, meeting summaries, content production, searching files, or help preparing presentations. Structured data was a natural, but particularly strategic, extension of this use.

In companies, information is generally not lacking. On the contrary, it is spread across numerous systems: customer relationship management tools, accounting software, e-commerce platforms, data warehouses, support tools, human resources applications, or audience measurement solutions. The challenge is not only to store this data, but to connect it, understand it and make it usable by people who master neither SQL nor internal data models.

This is precisely the problem targeted by the data agent announced by OpenAI. According to the company’s presentation, it can connect professional data sources in order to extract actionable information from them. Users can therefore phrase a question in their own words rather than go through a succession of exports, filters, queries and spreadsheet manipulations. The product is then expected to be able to generate analyses as well as interactive dashboards.

The promise is part of a broader trend: conversational business intelligence. For years, access to an organization’s figures has been organized around dashboards designed by specialized teams. Business users consumed the indicators made available to them, requested new reports when they needed a different view, and then sometimes waited for an analyst or data team to turn their question into a query. Assistants based on language models aim to shorten this chain.

OpenAI does not claim merely to simplify visualization. The announcement suggests a more direct interaction between the business question and the data that can answer it. A sales manager might seek to understand a change in results; a marketing team might want to examine a campaign’s performance; an operations department might compare periods or identify a trend. The product’s value will then depend less on its ability to display a chart than on its ability to accurately translate human intent into reliable analysis.

OpenAI’s original publication summarizes this direction in its title: “Now everyone can put data to work”.

This wording is revealing. The word “everyone” does not necessarily refer to the disappearance of data teams, but to the idea that querying data should no longer be reserved for technical profiles. For OpenAI, ChatGPT Work could become the access layer that allows more employees to quickly obtain answers from their organization’s information.

An agent that connects sources and translates business questions

The central point of the announcement is the arrival of an agent specialized in data. OpenAI states that this data agent can connect professional data sources in order to derive usable information from them. This connectivity capability is essential: without it, a conversational assistant remains limited to files or content provided on a case-by-case basis. With connected professional sources, the issue becomes continuous access to information that is closer to the actual state of operations.

In principle, the agent is intended to allow requests to be phrased in natural language. This approach breaks with traditional interfaces in which users must know a database’s structure, the exact name of a field, the logic of a data model or the location of an indicator. Instead of asking an analyst to prepare a report, a person can explain what they are looking for: a trend, a comparison, a breakdown, an anomaly or how an indicator has changed over time.

The nuance is important. An apparently simple question can encompass several interpretations. Asking for revenue for a period notably requires knowing which internal definition is used, which scopes are included, which currency is used, how refunds are handled, and whether revenue is recognized at the order, invoice or payment stage. In business intelligence tools, these conventions are generally set in the data model, documentation or existing reports. An agent makes interaction more natural, but it does not eliminate the need for shared definitions.

OpenAI also highlights the production of analyses and interactive dashboards. The distinction between these two elements matters. A one-off answer to a question can meet an immediate need. A dashboard, on the other hand, is more useful for monitoring a set of indicators, comparing segments and regularly returning to the same view. By combining the two, ChatGPT Work seeks not to stop at an isolated conversation: the tool is intended to help turn exploration into a management aid.

The interactive nature of dashboards is also a key element of the proposition. In traditional analytics tools, consultation does not merely involve reading a static visualization. Users filter, change periods, explore categories or compare groups. OpenAI presents its agent as capable of generating this type of artifact, which brings the product closer to the world of BI platforms rather than that of a simple text assistant.

The real value of such a system nevertheless depends on the quality of the connected data and the governance applied to its use. An agent cannot on its own correct incomplete data, contradictory definitions, duplicates or sources that are not up to date. Likewise, connecting several systems does not automatically guarantee that they refer to the same customer, product or transaction. In organizations that have already invested in data warehouses and semantic models, an assistant can accelerate access. In those where data remains fragmented, it can also make existing inconsistencies more apparent.

OpenAI’s announcement does not detail, in the elements presented, all supported sources, configuration methods, or precise administration arrangements. This lack of detail means the operational scope of the launch should not be extrapolated. OpenAI’s message is primarily about the function: enabling questions to be asked of company data, connecting professional sources, generating analyses and building interactive dashboards.

For business teams, this promise addresses a familiar frustration. Data tools are often powerful, but advanced use requires training. Conversely, spreadsheets are ubiquitous because they give an impression of immediate control, even when manipulations are difficult to reproduce or audit. A conversational interface could serve as an intermediary: more accessible than a manual query, but potentially more connected to authoritative sources than a static export.

Accessibility must nevertheless be distinguished from complete autonomy. Employees may be able to obtain initial answers more quickly, explore more avenues and reduce the volume of simple requests sent to analysts. But analyses that inform a financial, regulatory or strategic decision will continue to require human validation, an understanding of the definitions used and sufficient traceability. Natural language makes the request easier; it does not replace control of the answer.

ChatGPT Work moves directly toward business intelligence

With this launch, OpenAI is entering territory already occupied by major business intelligence vendors. Microsoft Power BI, Salesforce Tableau, Qlik and Google Looker are among the established players in organizations for preparing, visualizing and distributing analyses. These platforms are not limited to displaying charts: they generally rely on connectors, data models, sharing mechanisms, access rules and governance tools.

The difference OpenAI is seeking to establish lies in the interface. Traditional BI tools were long designed around reports, dashboards, fields, filters and menus. They then added search or natural-language query functions, followed by generative assistants. ChatGPT Work starts with conversation. In this model, dialogue is not a secondary feature added to an analytics platform; it becomes the main entry point to data.

Microsoft has notably introduced Copilot features in its analytics environment, particularly around Power BI and Microsoft Fabric. Tableau launched Tableau Pulse, an approach focused on delivering personalized information and indicators. Google has also highlighted AI-powered conversation and assistance capabilities in Looker. These developments show that the market already views natural language as an important interface for decision-making data.

OpenAI is therefore entering a category that is not new, but with a potential advantage: familiarity with ChatGPT. For many users, conversing with an assistant has become a habit in writing, research, programming or summarization tasks. If that same environment enables access to business data, adoption may seem more natural than that of a new specialized tool. The proposition is to bring analysis into the same place as other knowledge tasks.

This strategy aligns with the broader evolution of workplace suites. Microsoft integrates AI into Microsoft 365 and its data ecosystem; Google does the same around Workspace and Google Cloud; Salesforce organizes its AI functions around customer data and its business applications. In each of these cases, data becomes the foundation on which the assistant is expected to provide useful context. OpenAI, for its part, is seeking to make ChatGPT Work a sufficiently general interface to support multiple functions and multiple work environments.

Competition is not played out solely on the quality of a chart or the fluency of an answer. It concerns the ability to bring together several conditions: understanding the request, accessing the right data, respecting each user’s permissions, generating verifiable results and integrating into established processes. Legacy tools often have deep roots in information systems. OpenAI is betting on a conversational experience whose simplicity is already recognized by the general public and many professionals.

Traditional BI also has the strength of formalizing indicators. In a large company, the same term can have several meanings depending on the department. A “sale,” an “active customer,” a “margin” or a “conversion” are not necessarily calculated in the same way in finance, sales and marketing. Decision-making platforms serve precisely to stabilize these definitions. To compete with these solutions for large-scale uses, a conversational agent will have to fit into this discipline rather than avoid it.

The opposite risk is a multiplication of informal analyses. When users easily obtain an answer, they may be tempted to reuse it without checking the scope, filters or freshness of the source. Yet a persuasive chart is not necessarily an accurate chart. Companies deploying this type of tool will therefore need to distinguish between rapid explorations, useful for guiding thinking, and certified figures intended for official reporting.

This is where the notion of an agent is more ambitious than that of a chatbot. A chatbot responds based on provided context. A data agent is expected to orchestrate an approach: understand the question, access relevant information, produce an analysis and format it in a dashboard. This chain brings ChatGPT Work closer to the analytics professions. It also raises higher requirements for accuracy, transparency and control.

The decisive challenge of trust, governance and data professions

In the professional world, data access is never only a technical question. Company information may contain sensitive commercial data, financial elements, employee information, customer data or trade secrets. Any interface that makes querying this information easier will have to fit into existing confidentiality and control rules.

OpenAI’s announcement describes the analytics function, but it does not on its own make it possible to draw conclusions about the precise security, compliance or administration parameters that will accompany each deployment. It would therefore be premature to attribute technical properties to the product that are not detailed in the source. For organizations, the practical questions will nevertheless remain unavoidable: which users can query which sources, which data can be exported, how are results shared, and how can usage be controlled?

In Europe, and particularly in France, this dimension is reinforced by the regulatory framework surrounding data protection. The General Data Protection Regulation already imposes obligations on organizations when they process personal data. The arrival of a conversational layer above databases and business applications does not alter the fundamental fact: the company’s responsibility will depend on its own configuration, processing activities and governance framework.

The issue is also sensitive for French companies that must reconcile innovation, sovereignty, sector-specific requirements and control over their information systems. Banks, insurers, healthcare institutions, public-sector entities and industrial companies do not deploy a new analytics tool as an isolated application. They assess it against their access policies, contracts, security procedures, constraints on data location or movement, and ability to audit decisions.

Data teams will therefore play a decisive role, even if OpenAI’s ambition is precisely to open analysis to non-specialists. Democratization does not mean that analysts, data engineers, governance managers or security teams become unnecessary. On the contrary, it may increase the importance of their preparation work: documenting sources, defining metrics, identifying reference datasets, organizing permissions and flagging interpretive limitations.

In the best-case scenarios, a data agent reduces repetitive tasks. Analysts can spend less time responding to basic aggregation or filtering requests and more time examining complex questions, improving data or supporting decision-making. But this redistribution is not automatic. If the tool produces ambiguous queries, analyses that are difficult to verify or numerous requests on poorly prepared sources, specialized teams could instead be called upon more often to correct and explain.

The issue of traceability is particularly important. A conversational answer appears immediate, even though it potentially results from a series of choices: source selection, interpretation of terms, aggregations, filters, periods compared and visualization method. To be used in a professional setting, it must be possible to place it back in this context. Users must in particular know what the result measures and what it does not measure. This requirement exists in traditional BI; it does not disappear with a natural-language interface.

The risks associated with generative models must also be considered carefully. An assistant can formulate a convincing explanation that gives the impression of greater certainty than the data actually makes possible. Correlations may be mistaken for causes, figures may be misinterpreted, or a poorly phrased question may lead to a misleading comparison. The challenge is therefore not only to obtain results faster, but to preserve the quality of the reasoning that accompanies them.

For managers, this entails a change in practice. They may be encouraged to query data more directly, but they will also need to learn to specify their questions, compare results and request explanations of the methods used. For executive management, the benefit may be faster circulation of information. For finance departments and control teams, the priority will remain consistency of definitions and trust in the figures used.

• For business functions: the expected benefit is more direct access to information and reduced dependence on manual queries.
• For data teams: the challenge is to provide coherent, documented and governed sources.
• For security and compliance managers: attention will focus on access rights, sensitive data and sharing conditions.
• For executives: the question will be which uses relate to exploration and which can feed into reference reporting.

This distribution of responsibilities shows why the battle for conversational BI will not be won solely through a spectacular demonstration. Companies will expect evidence of the product’s robustness in their own environments. An agent’s value will not be measured only by the number of questions it answers, but by teams’ ability to trust it on matters that count.

An opportunity for French companies, but a deployment that will need to be selective

For the French-speaking market, OpenAI’s announcement may resonate particularly with organizations seeking to broaden access to data without multiplying technical training. Many companies already have reporting tools, data warehouses, spreadsheets and business platforms, but the use of these resources often remains concentrated among a few specialists. A conversational layer could enable sales, marketing, finance or operations functions to express their needs more easily.

Small and medium-sized enterprises may see a different prospect from that of large groups. They sometimes have fewer resources dedicated to data and BI, while still needing to monitor their operations responsively. A tool capable of connecting professional sources and answering common questions can reduce certain obstacles. But these companies are also often dependent on data spread across management software, sales tools and spreadsheets. The quality of integration will therefore remain decisive.

For large organizations, the potential lies more in scaling. A company with several thousand employees may already have sophisticated dashboards, but discovering and using them can still be difficult. Users do not always know which report to consult, who to ask for a modification or which metric matches their need. ChatGPT Work could, according to the direction described by OpenAI, serve as a more intuitive interface between users and this analytics estate.

French represents a practical issue here, not merely a linguistic one. Employees must be able to phrase questions in the language they use every day, with their business expressions and internal conventions. In an organization present in France, Belgium, Luxembourg, Quebec, Switzerland or French-speaking Africa, data may also be described using local vocabularies and acronyms specific to each function. A conversational interface has value only if it understands these uses without erasing the formal definitions established by the company.

Deployment could begin with scopes where indicators are well identified and data is relatively homogeneous. Sales teams, for example, often use regularly monitored metrics. Marketing functions may have campaign, acquisition or performance data. Operations may monitor volumes, lead times or incidents. Conversely, areas that combine highly sensitive information or particularly complex calculation rules will likely require greater caution.

The question of adoption must also be distinguished from that of availability. A tool can be technically accessible without being immediately adopted at scale. Employees will need to know which questions to ask, which sources are reliable and how to interpret results. The companies that gain the most value will probably be those that support use with documentation, clear rules and appropriate training, rather than those that present the agent as a universal substitute for all analytics expertise.

Integrators, consulting firms, data specialists and French or European management software vendors could also be affected. The shift toward conversational interfaces creates new needs: preparing sources, building reference frameworks, organizing permissions, defining indicators and supporting teams. OpenAI’s agent can simplify the final stage of access to information, but it does not remove the need to work on data architecture and quality upstream.

For BI providers, the message is clear: visualization alone is no longer sufficient to differentiate a platform. Users increasingly expect to be able to converse with their data, request an explanation and move from a question to an action without constantly changing tools. OpenAI is not the first to pursue this ambition, but integrating this function into ChatGPT Work may accelerate pressure on the entire market.

OpenAI’s strength lies in the potential centrality of its interface. If ChatGPT is already used to prepare a note, analyze a document, help structure a presentation or summarize exchanges, data access can appear as a coherent extension. The limitation is equally obvious: the company will have to demonstrate that this ease of use can be reconciled with the security, accuracy and governance requirements specific to professional data.

Toward a new conversational layer for the information system

The prospect opened by ChatGPT Work’s data agent goes beyond dashboard generation alone. Through this announcement, OpenAI is testing a broader idea: making the conversational assistant a common interaction layer above enterprise work tools and data. Users would no longer necessarily start by choosing an application, report or database. They would start by stating their objective, then let the interface guide them toward the necessary information.

This evolution could change how enterprise software is used. Specialized applications will retain their own functions: entering an order, managing a customer file, tracking a budget, administering a project or controlling an operational chain. But the search for cross-functional information could gradually shift toward assistants capable of navigating among several sources. ChatGPT Work clearly aims to take its place in this space.

The change is also cultural. For a long time, data access relied on learning tools. Users had to know a spreadsheet, understand a reporting interface or master query syntax. The conversational model reverses this logic: technology must adapt more closely to human phrasing. It is an attractive promise, but it shifts the difficulty toward the quality of interpretation, the reliability of sources and the ability to verify results.

In the long term, competition could be structured around two visions. The first extends existing BI platforms by adding a generative assistant. The second, embodied by OpenAI’s direction, makes the assistant the main entry point, with dashboards and analyses becoming products of the conversation. In practice, companies could adopt hybrid models: certified reports for official management, and agents for exploration, preparing questions and everyday access to information.

The success of this approach will depend on several elements that remain more important than the novelty effect: the ability to connect the right data, understand business terms, ensure clarity of generated results, respect access permissions and distinguish an exploratory hypothesis from a reference figure. With its data agent, OpenAI states that it wants to give everyone the means to put data to work. The market must now assess whether this promise can be durably integrated into the very concrete requirements of organizations.

For French and European companies, the question will therefore not simply be whether conversational analytics is appealing. It will be to determine in which processes it can create value without weakening trust in the figures. If ChatGPT Work succeeds in becoming the interface between business teams and genuinely governed data, OpenAI could strengthen its status as a work tool. If answers remain difficult to verify or integrate into existing control frameworks, traditional BI platforms will retain a structural advantage. It is in this tension between accessibility and rigor that the next stage of AI applied to business data will be decided.