The biggest change in business AI in 2026 is not simply that models have become smarter. ChatGPT and Claude are increasingly moving from assistants that answer questions to agents that can access business information, work across applications and complete multi-step tasks.
For companies evaluating ChatGPT for business, Claude for business, AI agents or enterprise AI automation, this shift is important. The highest returns are increasingly coming from automating complete workflows rather than asking employees to use AI simply for writing emails or summarizing documents.
What's New in ChatGPT for Business?
One of OpenAI's biggest developments is ChatGPT Work, an agent designed to take a business objective and work across applications, files and workflows to produce finished outputs such as spreadsheets, presentations, documents and web applications. It can break larger projects into individual tasks and continue working through them rather than requiring the employee to prompt it at every stage.
ChatGPT Business has also introduced Workspace Agents. Companies can create reusable agents connected to systems such as Google Drive, Google Calendar, Slack and SharePoint, add custom MCP integrations, publish them internally and even schedule recurring workflows. This effectively allows businesses to turn successful employee prompts and processes into repeatable AI-powered operations.
Another particularly important development for management teams is the new Data capability in ChatGPT Work. Employees can ask questions such as "Why did sales decline last month?" or "Which customers are showing renewal risk?" and have the agent investigate connected company data and build interactive reports and dashboards without requiring the user to write SQL queries.
OpenAI has also moved AI closer to everyday business applications. ChatGPT for Excel and Google Sheets can build, clean, update and explain spreadsheets, while ChatGPT for PowerPoint can create and revise editable presentations. Company Knowledge can search information distributed across systems such as Slack, SharePoint, Google Drive and GitHub while returning citations to the original company sources.
The result is an important change in how businesses should think about ChatGPT: less as a separate application employees visit and more as an intelligence layer operating across the systems where employees already work.
What's New in Claude for Business?
Anthropic is moving in a similar direction with Claude Cowork, Claude Tag, Microsoft 365 integrations and increasingly capable long-running agents.
Its newest Claude Opus 5.5 model, released on September 22, 2026, is positioned specifically for professional and enterprise work. Anthropic says it can produce professional spreadsheets, presentations and documents while handling multi-stage projects and financial analysis. Anthropic also estimates typical token-billed workloads cost around 40% less than Opus 5, making model economics increasingly relevant when deploying agents at scale.
Claude Fable 5.1 is designed for even more ambitious long-running knowledge work and agentic workflows. Anthropic reports approximately 25% lower costs for typical workloads and potentially up to 45% lower costs for highly agentic workloads compared with Fable 5.
For ordinary business users, however, some of the most interesting improvements are happening outside the model itself.
Claude can work across Excel, PowerPoint, Word and Outlook through Microsoft 365 add-ins. An analyst can begin financial analysis in Excel and continue building the resulting executive presentation without having to re-explain the entire project when moving to PowerPoint.
Claude Tag takes another approach by putting Claude directly inside Slack. Teams can tag Claude with a task, give it access to selected company tools and data, and let it work through the task before returning the result to the Slack thread. Anthropic says its own product organization is already using this approach extensively for software development, analytics, support work and investigating technical problems.
Anthropic has even packaged this concept specifically for smaller organizations. Claude for Small Business connects with platforms including QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace and Microsoft 365 and includes pre-built workflows covering finance, operations, sales, marketing, HR and customer service.
3 Real-World Examples Showing Where the ROI Is
- BBVA: Giving AI to Approximately 100,000 Employees
BBVA has deployed ChatGPT Enterprise to approximately 100,000 employees globally.
According to OpenAI's customer case study, employees report saving around three hours per week, while selected workflows have achieved efficiency improvements of as much as 80%.
The important lesson is not simply the size of BBVA's deployment. AI is being integrated across customer experience, operations, software development and employees' everyday knowledge work.
For organizations considering enterprise AI, this illustrates where scale creates ROI: saving even relatively small amounts of time on activities performed thousands of times every week can generate enormous organizational capacity.
- IG Group: Claude Reaches Full ROI Within Three Months
Financial services company IG Group provides a particularly useful example of measurable Claude ROI.
According to Anthropic, IG Group achieved full ROI within its first three months through productivity improvements, faster delivery and reductions in external agency costs. Its analytics team reportedly saves approximately 70 hours every week, while some individual workflows experienced productivity improvements of around 100%.
Marketing teams also accelerated campaign delivery while reducing dependence on external agencies.
This demonstrates another major AI opportunity: use AI first where expensive employees or external suppliers repeatedly perform research, analysis, content production or data manipulation.
Reducing outsourcing expenditure while simultaneously increasing internal capacity can create a much clearer financial return than simply measuring the number of employees using an AI tool.
- NVIDIA: Moving From AI Assistance to Delegated Work
NVIDIA provides an example of the next stage of adoption: giving complete pieces of work to an AI agent.
OpenAI reports that NVIDIA teams using ChatGPT Work saved around 16 hours per week during its GTC planning cycle. Teams also created working prototypes in approximately three to five days instead of two to three weeks.
ChatGPT Work is also being used to monitor large volumes of external information and convert those updates into actionable signals for internal teams.
This is potentially much more valuable than asking AI to write individual documents. The employee defines the objective while the agent handles much of the research, organization, analysis and production required to reach the finished output.
Where Businesses Should Look for Maximum AI ROI
The companies generating the greatest returns are increasingly moving beyond isolated prompts.
OpenAI's 2026 enterprise research found that its highest-usage organizations were consuming substantially more AI intelligence per worker, with the biggest differences appearing in advanced tools and delegated agentic workflows rather than simply sending more chat messages.
For most businesses, the highest-potential opportunities therefore share several characteristics:
- High volume: the task happens hundreds or thousands of times.
- High labour cost: skilled employees spend significant time completing it.
- Repeatable process: the workflow follows reasonably consistent steps.
- Accessible data: the AI can securely access the information required to perform the task.
- Measurable outcome: management can measure hours saved, turnaround time, outsourcing costs, error rates, revenue generated or customer satisfaction.
Instead of asking, "Where can employees use AI?", businesses should increasingly ask "Which complete workflows can we redesign around AI?"
ChatGPT vs Claude for Business: The Bigger Picture
There is no longer a simple distinction where ChatGPT is primarily a chatbot and Claude is primarily a document assistant.
Both are evolving into enterprise AI platforms capable of connecting company information, applications, agents and workflows.
ChatGPT is building strongly around ChatGPT Work, Workspace Agents, company knowledge, business data analysis and integrations across workplace applications.
Claude is building around Cowork, Claude Tag, long-running agents, Microsoft 365 integration, MCP connectivity and increasingly sophisticated knowledge-work automation.
For business leaders, the competitive question is therefore becoming less about which model produces the better individual response and more about which platform can be integrated most effectively into the company's actual workflows, data and governance model.
The Real Opportunity: From Copilot to Digital Workforce
The next phase of enterprise AI will not be defined by employees occasionally opening an AI chatbot.
It will be defined by employees delegating increasingly complete pieces of work to AI systems that understand company information, operate approved tools and return finished outputs for human review.
The organizations seeing the strongest returns are already moving in this direction.
For companies beginning their AI transformation, the best place to start is usually not with dozens of experimental use cases. Identify three to five repetitive, expensive and measurable workflows, connect AI securely to the systems required to perform them, maintain human approval for important decisions and measure the financial impact.
That is where tools such as ChatGPT Work, ChatGPT Workspace Agents, Claude Cowork and Claude Tag can move AI from an interesting productivity tool to a measurable business investment.