Profitability & margin
Spread measured by customer, product, line of business and region, using one certified set of transfer-pricing rules (see how “net interest margin” is used below).
Industry · Financial services
A margin calculation you can explain to an examiner.
AtScale defines each financial services metric once, so AI assistants, agents and BI tools give treasury and finance the same number, every time.
When an analyst asks for margin, AtScale checks which version they mean, then calculates it the same way for every request for Claude, Grok, and Power BI. Below are six ways financial services teams use AtScale and six common metrics, from prompt to the automated agents that act on them.
Use cases
Six domains where a bank’s numbers must agree.
FSI runs on numbers. AtScale defines them in one place, with the precise SQL required to compute them, every time.
Portfolio & research
The AtScale ACE engine ensures “Position” and “Performance” are calculated the same way, every time, on every desk and in every region.
Risk & regulatory
A “semantic calculator” like AtScale ensures risk figures are always accurate, never drift, hold up under examination, and are traceable from source to filing.
Customer 360 & marketing
Prospect, client and counterparty data get resolved in one way across systems that describe your customer.
Fraud & financial crime
Analysts work from set of transaction data, one set of cases, and one view of actors so cases are triaged quickly and consistently.
Cost & cloud governance
Which systems are using which compute? Who spent what on warehousing? Teams exploring these questions get the same answers so quarter-over-quarter computations all foot.
Net interest margin, from question to agent.
Regulators don’t accept “the AI said so.”
A bank has to explain every number it reports to examiners who can fine it. If an AI agent produces a number the bank can’t trace back to its calculation, the bank can’t use it.
What analysts ask and what they want can be different.
Analysts ask vague questions, and LLMs are happy to hallucinate what they meant. Before AtScale ACE computes anything, we work out which metric and rules apply.
What a Analyst prompts | What they meant |
|---|---|
What a Analyst prompts “What’s our margin by region?” | What they meant Net interest margin by region, each one from its own net interest income over its own average earning assets. |
What a Analyst prompts “What’s our exposure to this counterparty?” | What they meant Net exposure to this counterparty across every desk, on today’s positions. |
What a Analyst prompts “Where did the loss provision move?” | What they meant Which segments moved the loss provision this quarter, and by how much, against last quarter’s close. |
Three jobs AI agents can automate with an AtScale semantic layer
Consider three simple AI agents at work in a typical bank. An AtScale semantic layer applies computations analysts already tested, so automated systems can be build and relied upon not to hallucinate answers. Below, fully automated agent decisions are low-stakes; person approves means exceptions are raised, and the human in the loop signs off before the action happens. With AtScale, you decide which automation style is right, and AtScale provides the right numbers.
AI agent | What Atscale calculates | how action is taken |
|---|---|---|
AI agent Every morning, “Flag any region whose net interest margin slipped more than 10 basis points.” | What Atscale calculates Calculates NIM by region, each from its own net interest income over its own average earning assets. | how action is taken Fully automated. Agent sends an alert to flag slipped region. Nothing happens until it’s reviewed. |
AI agent Every hour: “Hold new trades with any counterparty over its exposure limit.” | What Atscale calculates Net exposure to each counterparty across every desk, on current positions. | how action is taken Circuit breaker fired and flagged for approval. The agent puts the trade on hold and pages the risk officer, who releases or blocks it. |
AI agent At quarter close: “Draft the loss-provision commentary for the CFO.” | What Atscale calculates Which segments moved the loss provision, and by how much, against last quarter’s close. | how action is taken Analyst workflow. Drafts are used as suggestions. Finance checks every number before sending. |
Proof points from this sector
White Papers & ReportsSemantic Layers and the Economics of AI on the Warehouse
Case StudiesTrumid
White Papers & Reports