What is function calling in AI Agents? A Guide for developers in wealth management

Function calling is the mechanism that lets an AI agent decide when it needs to use a specific tool or API, then pass structured arguments to that tool instead of guessing an answer. In practice, it turns a language model from a text generator into a system that can query data, trigger workflows, and return grounded results.

How It Works

Think of function calling like a private banker handing a task to operations.

The banker does not personally check every account balance, tax lot, or policy document. They identify the right internal team, provide the required details in a standard format, and wait for the result. Function calling works the same way: the model sees a user request, selects an available function, fills in the parameters, and your application executes the call.

A simple flow looks like this:

• •User asks: “What is my client’s exposure to tech stocks?”
• •The model decides it needs portfolio data.
• •The model emits a structured function call like get_portfolio_holdings(client_id)
• •Your backend runs the function against your portfolio system.
• •The result comes back to the model.
• •The model summarizes it in plain English for the user.

The important part is that the model does not directly invent the answer. It requests data from systems you trust.

Here’s what that looks like in practice:

{
  "name": "get_portfolio_holdings",
  "arguments": {
    "client_id": "C12345",
    "asset_class": "technology"
  }
}

For developers in wealth management, this matters because most useful agent actions are not pure chat. They are controlled operations against internal systems:

• •retrieving holdings
• •checking suitability rules
• •generating performance summaries
• •creating service tickets
• •drafting client communications from approved data

Function calling gives you a clean contract between the model and your business logic. The model handles intent recognition; your code handles execution, permissions, validation, and auditability.

Why It Matters

• •

  Reduces hallucinations

  The agent can fetch actual portfolio or policy data instead of making up numbers. That is critical when users ask about AUM, allocation drift, risk scores, or claims status.

• •

  Fits regulated workflows

  Wealth management systems need controls around who can see what, which actions are allowed, and how outputs are logged. Function calling keeps those controls in your application layer.

• •

  Improves user experience

  Users can ask natural questions like “Show me clients with concentrated exposure” without learning system-specific commands or report names.

• •

  Makes agents actionable

  A good agent should do more than summarize text. It should be able to open cases, pull records, calculate metrics, and route work to downstream systems.

Real Example

Say you are building an assistant for relationship managers at a wealth firm. A common request is:

“Which clients have more than 20% exposure to one sector and need rebalancing review?”

That sounds simple to a human analyst but requires multiple steps:

1. •Identify eligible clients.
2. •Pull their holdings.
3. •Calculate sector concentration.
4. •Compare against policy thresholds.
5. •Return a ranked list with supporting evidence.

With function calling, you expose tools such as:

{
  "name": "find_clients_by_risk_rule",
  "arguments": {
    "rule_name": "sector_concentration_gt_20"
  }
}

Then another function might fetch details for each client:

{
  "name": "get_client_portfolio_summary",
  "arguments": {
    "client_id": "C12345"
  }
}

The agent can then produce something like:

Client C12345 has 28% exposure to financials and 24% exposure to technology. Both exceed the firm’s concentration review threshold. Recommend advisor follow-up and rebalance review.

In an insurance context, the same pattern applies to policy servicing:

• •get_policy_status(policy_id)
• •check_claim_eligibility(claim_id)
• •create_service_case(customer_id, reason)

The model is not replacing business logic. It is orchestrating it.

Related Concepts

• •

  Tool use

  Broader term for letting models interact with external systems such as databases, search engines, calculators, or internal APIs.

• •

  Structured outputs

  A way to force model responses into JSON or another schema so downstream systems can parse them reliably.

• •

  Agent orchestration

  The control layer that decides which tools run, in what order, and under what constraints.

• •

  RAG (Retrieval-Augmented Generation)

  Used when the agent needs documents or knowledge base content before answering questions about products, policies, or investment guidelines.

• •

  Guardrails and policy enforcement

  Rules that block unsafe actions, enforce entitlements, and keep regulated workflows auditable.

If you are building AI agents for wealth management, treat function calling as the bridge between conversation and execution. The model understands intent; your functions make it safe, deterministic code that actually does the work.

Keep learning

• •The complete AI Agents Roadmap — my full 8-step breakdown
• •Free: The AI Agent Starter Kit — PDF checklist + starter code
• •Work with me — I build AI for banks and insurance companies

By Cyprian Aarons, AI Consultant at Topiax.