AI Agents for SAP HANA
HANA changed how fast you can ask. It did not change who opens the documents, works the exceptions and writes the sentence that goes in the pack.
- HANA moved the bottleneck from the database to the people. Faster reports and the same monthly workload is a frustrating place to be after a significant investment.
- HANA support means at least four different read paths. S/4HANA has the universal journal, ECC on HANA does not, and a data mart is a model rather than the ledger.
- Running ECC on HANA makes it faster. It does not give you ACDOCA, and an agent assuming that table is reading something that is not there.
- On SAP the control is structural. An RFC enabled BAPI does not auto commit, so a document an agent creates is parked and waits for release through your existing workflow.
- Reconciling to a modelled data mart view rather than to the ledger produces answers that are internally consistent and wrong.
ChatFin runs finance AI agents on SAP HANA. It reads through CDS views and OData services on S/4HANA, or through RFC enabled BAPIs and the ledger tables where you run ECC on HANA, answers questions at transaction grain, works exception queues by cause and prepares close entries. Everything it creates is parked for a named person to release through SAP workflow.
The honest starting point is that HANA already solved the problem it was bought to solve. Queries that were overnight jobs return in seconds. What did not change is who opens the documents behind a movement, decides what an exception means and writes the commentary. That is a people bottleneck sitting behind a very fast database, and it is the part ChatFin was pointed at.
Fast data is not automated work
SAP HANA gives finance a genuinely fast in memory foundation. Queries that used to be overnight jobs return in seconds. What that changes is how quickly you can ask. It does not change who has to open the documents, decide what an exception means and write the sentence that goes in the pack.
| What HANA changed | What it did not change |
|---|---|
| Query response time, dramatically | Who works the exception queue |
| The practicality of transaction level reporting | Whether anyone has time to read it |
| Real time balances | The need to explain why a balance moved |
| The cost of a new analysis | The queue to get one built |
Most finance teams on HANA end up with faster reports and the same monthly workload, which is a frustrating place to be after a significant investment. The bottleneck moved from the database to the people, and no database upgrade fixes that.
ChatFin is additive to your HANA investment. It uses the speed rather than duplicating it, and where a CDS view or an existing report already answers a question cleanly there is no reason to put an agent in front of it.
How ChatFin connects, and it depends what runs on HANA
HANA is the database. What matters for finance work is the application on top of it, and the read path differs.
| What you run | Where the finance data sits | How ChatFin reads it |
|---|---|---|
| S/4HANA | The universal journal, ACDOCA | CDS views and OData services, plus the underlying tables where a view does not exist |
| SAP ECC on HANA | BKPF, BSEG and FAGLFLEXA, as on any ECC | RFC enabled BAPIs and the ledger tables. Not ACDOCA, which is not there |
| BW or a data mart on HANA | Modelled views over extracted data | Calculation views, with the caveat that a model is not the ledger |
| A side by side scenario | Both | Read from the system of record, not the copy |
That second row is the one that catches tools. Running ECC on HANA makes it faster. It does not give you the universal journal, and an agent assuming ACDOCA is reading a table that is not there.
What ChatFin does once it is connected
Four jobs, and the speed of HANA makes the first one better than it is elsewhere.
Answer at transaction grain, interactively
On a slower platform, transaction level questions get batched because each one costs an overnight run. On HANA they can be conversational, which changes what is worth asking. ChatFin returns the filter, the period, the posting status and the row count with every figure.
Explain a movement rather than report it
Split into components that reconcile to the reported total, with the documents behind each. The retrieval that makes this possible is exactly what HANA is good at.
Work exception queues by cause
Payables, receivables and clearing populations grouped into repeating patterns and routed to whoever can resolve them.
Prepare close entries with their support
Accruals, reconciliations and corrections, each with the rows behind it, for a named person to review and post.
Governed write back on an SAP platform
On SAP the control is unusually clean, because the interface enforces it rather than a setting.
ChatFin uses the parked document as the control point, so no second approval mechanism has to be built or audited alongside your existing SAP workflow.
Where ChatFin underperforms on SAP HANA
Three, and all three predate any tool.
More on ChatFin and SAP
Deeper reading, all live.
In build for SAP HANA
SAP HANA FP&A AI · SAP HANA month end close AI · SAP HANA accounts payable AI · SAP HANA accounts receivable AI
Questions SAP finance teams ask first
Does ChatFin work on S/4HANA or on ECC running on HANA?
Both, with different read paths, and the distinction matters more than vendors usually admit. On S/4HANA the finance data sits in the universal journal ACDOCA and ChatFin reads through CDS views and OData services. On ECC running on HANA the documents are still in BKPF and BSEG with the general ledger in FAGLFLEXA, and ACDOCA is not there at all. Establish which you run before evaluating any tool that claims HANA support.
Can an agent post directly into SAP?
It should not, and on SAP the interface makes that easy to enforce. RFC enabled BAPIs do not auto commit, so a document created through one is parked rather than posted and waits for release through SAP workflow. ChatFin uses that parked document as the control point, which means the approval is structural rather than a setting, the audit trail is a standard SAP one, and if the agent stops nothing is half posted.
We already have a HANA data mart. Can ChatFin just read that?
It can, and usually it should not. A modelled view is somebody's interpretation of the ledger, built for a purpose, and reconciling to it rather than to the system of record produces answers that are internally consistent and wrong in ways nobody notices. ChatFin reads the system of record and will say so when it is being pointed at a copy.
Use the speed. Remove the retrieval.
ChatFin reads your SAP platform through CDS views or the ledger tables depending on what you run, answers at transaction grain with the filter and row count attached, and groups exception queues by cause.
Everything it creates is a parked SAP document released by a named person through the workflow you already maintain.
Bring one report your team already signs. On HANA the tie out takes minutes.
Book a demoAll images are served from the Pexels content network. No third party product screenshots, vendor logos or licensed media are used on this page.
See it on your own SAP data.
Bring one report your team signs and one exception population. We will tie out the first and group the second by cause, and tell you which read path applies to you.