An Excel reconciliation can begin as a simple question and become a recurring finance process. Prophecy Inc.’s newly launched DataGPT aims to support that transition within ChatGPT, Claude and Copilot, pairing answers with a visual account of how they were produced. For business analysts, the practical test is whether a useful chat interaction can become an approved, repeatable workflow.
From Spreadsheet Reconciliation to a Monthly Process
Prophecy’s walkthrough starts with the kind of mixed-input task that makes conversational analytics attractive. An analyst brings in an untidy Excel workbook, checks it against an audit list, and combines it with information from SharePoint or Snowflake Inc.’s platform. The user can then introduce a discount rule for product bundles and prepare the process for monthly execution.
The business value sits in that final step. Reconciling information once may save time, but repeating the work reliably requires the organization to preserve its definitions and exceptions. A pricing adjustment must mean the same thing next month unless someone deliberately changes it.
DataGPT exposes the intermediate stages alongside their outputs, Prophecy says. Analysts can inspect a stage, correct it and approve the resulting workflow. This gives reviewers something more useful than a final total: a way to locate where a business rule entered the calculation and assess its effect.
A visual representation still needs careful review. It can make logic accessible to someone who does not read code, but it cannot establish whether an audit list is complete or a discount policy is appropriate. Those remain business responsibilities.
Why an Answer Is Insufficient for Governed Analytics
Prophecy describes an existing pattern in finance and operations: employees send spreadsheets to chatbots, sometimes without IT approval, and accept results produced by scripts they cannot interpret. The company argues that closing the conversation can also discard the logic and business context supplied during the session.
That creates a gap between personal productivity and organizational accountability. A plausible reconciliation result offers little assurance if another employee cannot establish which revenue definition or exception shaped it. For recurring reporting, the question becomes whether the calculation can be inspected and reproduced.
Raj Bains, Prophecy’s founder and chief executive, frames DataGPT as an attempt to give business users the production-oriented assistance already available to software developers. His argument is that conversational answers alone have not fundamentally changed how those users deliver operational work.
Prophecy also points to a January Gartner Inc. forecast that half of organizations will take a zero-trust approach to data governance by 2028 as unverified AI-generated information proliferates. The forecast provides context for the product’s emphasis on verification; it does not establish that any particular workflow is trustworthy.
How an Approved Conversation Becomes Scheduled Work
Prophecy says DataGPT requires human review before a workflow can be scheduled. Each process has an underlying implementation stored as open code in Git. Subsequent executions use that code rather than returning to the model, which the company says avoids additional AI charges for reruns.
That separation gives organizations a clearer review boundary. AI helps construct the process; approved code handles repeat execution. Teams should still check the business logic and source inputs, since consistently executing an incorrect rule would simply repeat the error.
| Workflow step | DataGPT’s stated approach | Practical review question |
|---|---|---|
| Build and inspect | Displays stages with their outputs | Do the transformations apply the intended rules? |
| Approve and schedule | Requires a person to review the process | Who owns approval and later changes? |
| Execute again | Runs the code without another model call | Will changing inputs still produce valid results? |
| Retain business context | Adds approved workflows to a customer knowledge graph | Which rules are suitable for reuse? |
Prophecy says the customer-specific knowledge graph draws on business rules and data catalogs as well as signed-off workflows. This could help preserve context beyond an individual conversation. Its usefulness will depend on the quality of what gets approved and how the organization maintains those definitions.
Adoption Depends on the Existing Data Environment
DataGPT’s deployment model addresses where the work happens as well as how users request it. Prophecy says uploaded files pass through its service into the customer’s data platform rather than remaining in the chat. IT and platform administrators can inspect and govern the workflows there.
Scheduled execution uses the customer’s existing Snowflake, Databricks Inc. or BigQuery environment. For adoption, this puts platform ownership near the center of the decision. Business teams may initiate the work conversationally, but operational responsibility still needs an owner within the organization.
The service is available through connector marketplaces for ChatGPT, Claude and Copilot. Organizations that have not approved those assistants can use Prophecy’s application instead. A free individual account provides an entry point, while administrators can deploy the product centrally.
Those options support different starting points, but individual access should not be mistaken for production readiness. A sensible evaluation would begin with a bounded reconciliation task, compare its outputs against accepted results, and review how corrections move into the scheduled version.
Measuring Value Beyond the First Successful Answer
Prophecy brings an established business backdrop to the launch. Founded in 2017, it has raised more than $100 million, including a $47 million Series B extension led by Smith Point Capital in January 2025. Its named customers include SAP SE, HSBC Holdings plc and Pfizer Inc.; those relationships should not be read as evidence of DataGPT adoption.
For analytics leaders, the useful measures are operational: time spent preparing recurring work, effort required to review it, and the ability to reproduce approved results. DataGPT’s proposition is strongest where a conversational task can become a maintained business process. An initial rollout should therefore test the full path from messy input to reviewed code and scheduled execution, with clear responsibility for keeping the rules current.
