If you have ever watched a business analyst paste a spreadsheet into ChatGPT, you know the feeling. The answer comes back fast, the numbers look plausible, and nobody in the room can actually explain how they got there. If finance asks "show me the steps," there are no steps to show — just a chat transcript and a script nobody read.
That is the exact problem Prophecy is aiming at with DataGPT, launched today, October 28, 2026. Instead of giving analysts another chatbot that answers questions about data, it turns the chat itself into a place where real data work happens — and, crucially, where every result arrives with a workflow you can look at, check, and change.
What DataGPT actually does
DataGPT lives where analysts already work: inside ChatGPT, Claude, and Microsoft Copilot, plus Prophecy's own app. An analyst describes the outcome they need in plain language, and DataGPT builds the whole workflow inside the conversation: take in messy files, prepare and clean the data, run the analysis, and produce the finished output.
The example Prophecy leans on is a familiar one. Imagine reconciling a quarter of transactions: upload a messy Excel file, check it against an audit list and data sitting in SharePoint, Snowflake, and Databricks, apply a pricing rule for discounted product bundles, review the results, and schedule the whole thing to run again every month. That is not one question and one answer. It is a pipeline — and DataGPT treats it like one.
It connects to files, SharePoint, data warehouses, and ERP and CRM systems. And when a workflow is approved, it does not stay trapped in a chat session. It ships as repeatable code that runs on the data platform the organisation already owns — Snowflake, Databricks, or BigQuery — on a schedule. The logic is open and editable, stored in Git, so it stays transparent and auditable instead of vanishing when the conversation ends.
Why "inspectable" is the whole point
Here is the part that matters most for developers and data teams, and it is worth slowing down for.
Most AI data tools today are black boxes with a friendly face. You ask, it computes, it answers. The script it wrote runs somewhere you cannot see, and when the session closes, the path from messy input to finished output disappears with it — along with the domain knowledge you fed in along the way: your business rules, your revenue definitions, your approved exceptions.
DataGPT's bet is that the workflow is the product. Every answer comes with a visual workflow behind it, so a business data team can inspect each step, approve it, and refine it. Analysts stop building every workflow from scratch and start reviewing instead: check each step and its results on screen, make small corrections, approve the final version. Once approved, the workflow reruns exactly the same way every time — no fresh AI generation, and no added AI cost on each rerun.
That last detail is easy to miss and genuinely important. A lot of AI tooling charges you every time it thinks. A workflow that is generated once, verified by a human, and then executed as ordinary code is a different cost model — and a different trust model. You are not re-rolling the dice on every monthly report.
Prophecy backs this with a knowledge graph that connects business context to data workflows, so definitions and business rules get applied consistently across a team's work rather than being re-explained in every new chat.
The shadow AI angle
There is also a governance story here that data platform leaders will recognise immediately. Analysts are already pasting company spreadsheets into AI chat tools without IT knowing — one of the most common forms of shadow AI. The data leaves the building, and nobody can audit what happened to it.
DataGPT routes uploaded files through Prophecy to the organisation's own data platform, not into an AI chat. Data platform leaders and IT can see every workflow and govern it on the platform they already own, without adding another silo. Whether that satisfies a given company's security review is, as always, something to verify in practice — but the architecture is pointed in the right direction: the chat is the interface, not the destination for the data.
Who it is for, and what it costs to try
DataGPT is aimed squarely at finance, operations, and line-of-business analysts — the people who live in spreadsheets and desktop tools, not the people who write Spark jobs for a living. If you are a data engineer, the interesting part is probably the other side of the same coin: analysts producing governed, repeatable workflows on your platform, in open code you can read, instead of mystery spreadsheets landing in your inbox.
Availability, as announced today: DataGPT is available now, inside ChatGPT, Claude, and Microsoft Copilot, or in the Prophecy app. Organisations can start with a free account, with centralised deployment options for administrators. Prophecy has not published detailed paid pricing in this announcement, so treat the free tier as the starting point and expect the real enterprise conversation to happen with their sales team.
Some context on the company behind it: Prophecy was founded in 2017 and is headquartered in Palo Alto, with more than 130 employees across the US, India, the UK, and Australia, and more than $100 million raised. Its customer list includes SAP, HSBC, JP Morgan Chase, and Pfizer. That does not make DataGPT good — but it does mean this is not a weekend project. It is a funded, enterprise-focused company betting that the next phase of enterprise AI is not better answers, but verifiable, repeatable work.
The honest caveats
A launch announcement is a claim, not a test. A few things to keep in mind before you get excited:
- We have not tried it. This is based on Prophecy's announcement, not hands-on testing. The workflow inspection experience, the quality of generated pipelines on genuinely messy enterprise data, and the Git-backed code quality all need real-world verification.
- "Works in ChatGPT, Claude, and Copilot" needs unpacking. How deep the integration goes in each assistant, what permissions and connectors each one supports, and how the experience differs between them is not fully spelled out yet. Try it in the assistant your team actually uses before assuming parity.
- Governance claims need your security team's sign-off. Routing data through Prophecy to your own platform is a better story than pasting spreadsheets into a public chatbot, but it is still a new vendor in your data path. Run it through your normal review.
- The free account is a starting point, not the enterprise product. Centralised deployment, admin controls, and presumably the serious connector and scale story will sit behind a commercial conversation.
None of that takes away from the idea. The shift from "AI that answers questions about your data" to "AI that builds a workflow you can inspect, approve, and rerun as code" is exactly the direction enterprise data work needs to go. The question is execution — and that is something the market will answer over the next few months.
What to do with this
- If you are a business analyst: try the free account on a real, small, recurring report — the one you redo by hand every month. The test is simple: can you inspect every step, fix the one that is wrong, and schedule it to rerun?
- If you are a data engineer or platform lead: look at the code it produces. Is it code you would accept into your repo? Is it readable, editable, and governed on your platform? That is the real quality bar.
- If you lead a data team: this is a useful conversation starter about shadow AI in your org. If analysts are already pasting spreadsheets into chat tools, a governed path that meets them where they work beats a policy nobody follows.
- If you build AI tools: note the pattern. The winning enterprise AI products of the next few years may not be the ones with the smartest answers, but the ones whose work a human can check, sign off, and rerun without paying the model again.
The age of "trust the chatbot" is ending. The age of "show me the workflow" is just starting — and DataGPT is one of the first serious attempts to build for it.
Source: Prophecy announcement via GlobeNewswire, October 28, 2026 (09:00 UTC) — https://www.globenewswire.com/news-release/2026/10/28/3389911/0/en/prophecy-launches-datagpt-bringing-enterprise-data-workflows-to-claude-chatgpt-and-copilot.html — and prophecy.ai. Facts verified against the announcement before writing; no press-release text copied verbatim.
Top comments (0)