EMINENCE VSP Talk to the architect

The Salesforce partner built for banking — lending journeys live this quarter, not eighteen-month programmes.

AI · OPTIMISATION· 6 min read · by Srinivasa, Founder & Architect · published Sep 2025, updated Jun 2026

Before Agentforce: the data foundation banks need for honest AI agents

Every bank wants the AI agent. Few want to hear that the agent is the last 10% of the work. Here's the unglamorous 90% that decides whether your AI tells the truth.

Why AI agents fail in banks

An agent answers from the data it can reach. Point one at a bank where the customer exists in five systems under three spellings, and it will answer confidently from whichever fragment it found — which is worse than not answering at all. The first regulator-facing AI incident at most banks won't be a hallucination; it will be a correct answer drawn from the wrong record.

That's why "deploy Agentforce" is the second project. The first is the foundation that makes an honest agent possible — and it's smaller than the data-transformation programme you're imagining, if it's scoped to what the agent actually needs.

The foundation, in four layers

One client record. The FSC data model — person accounts, households, financial accounts — gives the agent a single place where the customer is true. Without it, every answer is a guess about which fragment wins.
Permissions the agent inherits. An agent must see exactly what the requesting user may see — no more. On Salesforce that's the sharing model and Compliant Data Sharing doing the work: the guardrail is the same architecture that governs humans.
Data without copying. Core banking balances and warehouse history reach the agent through Data 360's zero-copy connections — grounded answers without a second copy of sensitive data to govern.
Grounding and audit. Every answer traceable to the records it drew on; every action logged. If you can't replay it, don't ship it.

2026 raised the stakes

The platform side of this is moving fast: Agentforce now takes voice calls inside FSC’s security model, Agent Script lets builders make critical flows deterministic, and Multi-Agent Orchestration lets agents hand work to each other with shared context. All of it makes agents more capable — none of it makes a wrong customer record less wrong. The better the agent, the more confidently it amplifies whatever foundation you give it.

Readiness, scoped to the first use case

AuditFor the one use case you'll pilot: which facts does the agent need, and where does each live today? Usually a one-page answer — and more revealing than a data strategy deck.
ModelMap those facts to FSC objects. What fits the standard model is free; what doesn't is your real scope.
PermissionDecide what the agent may see, expressed as sharing rules — then test that it cannot see anything else.
PilotOne agent, one journey, measured against the human baseline. Then expand from evidence.
The banks winning with AI didn't buy better agents. They built a foundation where an honest agent was possible.

The honest caveat

The opposite failure exists too: the bank that spends eighteen months "fixing data" before allowing any AI near a customer. The foundation isn't a lake-boiling exercise — it's the narrow slice the first use case needs, done properly. Scope it to weeks, prove it, and let each new use case pay for the next slice of foundation.

Three questions before any agent pilot

Which three questions should the agent answer on day one? If you can't name them, you're buying technology, not capability.
Can you trace each answer to a record? Grounding is a property of your data, not the model.
Who signs off the agent's scope — by name? Someone must own what it may do, and what it may never do.

How Eminence VSP helps

We run the Agentforce readiness exercise as a scoped engagement: one use case, a one-page data audit, the FSC mapping, sharing rules the agent inherits, and a measured pilot. Weeks, not a transformation programme. See AI & Agents or talk to the architect.

S.
Srinivasa
Founder & Architect, Eminence VSP — the person who scopes and delivers these builds.
Share on LinkedIn →

Let’s talk.

One conversation with the architect — and a clear view of what your bank could ship next quarter.

Talk to the architect