How It Works

A client's rules, entered once. A clear answer, every time the situation is covered.

Agent First is built from four pieces: a structured Client Profile for each account, a decision engine that checks the live situation against it, a small vocabulary of plain outcomes, and a rule that the agent, not the system, always has the final word.

1. The Client Profile

What gets configured, once, per client.

A Client Profile is a structured operating model built from the client's approved rules and the real decision moments the work requires: handbook language, internal wikis, call scripts, compliance documents, exceptions, authority, and next actions. For pilot partners, Agent First first observes a bounded slice of the workflow so the Profile reflects how work actually happens, not only how documentation says it happens. The result is not a document the agent has to search; it is what the decision engine checks against on every call.

1

Services and eligibility

What this client offers, and what makes a specific situation eligible or not.

2

Authorization levels

What an agent can decide alone, what needs verification, what's off-limits entirely.

3

Required language

Exact disclosures and wording that have to be said, and what an agent should never promise.

4

Escalation and transfer rules

Specifically who a situation goes to, not a generic “escalate.”

5

Hours, contacts, and routing

Who's reachable when, and what happens outside those hours.

6

What “not specified” should mean

The one decision that protects every gap the rest of the Profile doesn't catch.

2. The decision engine

Configure once. Ask on the call. Get a verdict and the words.

1

Configure once

A client's rules get entered once into a structured Client Profile, checked on every call, not searched on any of them.

2

Ask, on the live call

The agent tells Agent First who they're talking to (the Active Profile) and what the customer wants. No searching, no second system to check.

3

Get a verdict and the words

Agent First hands back a clear decision, plus approved or configured wording. The agent still decides what to say.

3. The outcome vocabulary

Every answer resolves to one of six plain outcomes, not a paragraph to interpret.

The same six outcomes are designed to apply whatever the industry, whatever the client. Inside one industry, learning the vocabulary once means adding a client is genuinely just configuration: real rules entered once, running on the same engine every Profile uses. Reaching that same depth in a brand new industry is the direction the architecture is built toward, not a claim already proven.

ALLOW / DO NOT ALLOW

Whether the agent can proceed on their own, one way or the other. Most calls resolve here, and resolving here is the whole point: a clear yes or no instead of an agent's best guess.

VERIFY

A detail to confirm before deciding anything else, like whether a policyholder purchased a specific coverage add-on, or whether a patient's symptoms fall inside a defined list.

DISCLOSE

A specific line this client requires the agent to say out loud, word for word, before or during the rest of the interaction.

DOCUMENT

Something that has to be logged or recorded before the call ends, so the record exists even when nothing else about the call was unusual.

TRANSFER / ESCALATE

When the call, or just the decision, belongs with someone else, a supervisor, an adjuster, a triage nurse, a property manager, named specifically, not just “escalate.”

NOT SPECIFIED

This client hasn't set a rule for this situation. Agent First says so instead of guessing, and routes it to a person. That gap then becomes a clear signal for what the Client Profile needs next.

4. Provenance: source shown

Every answer says which client's rule it came from.

Agent First doesn't just hand back a verdict. It names the specific rule behind it, for example “Source: Meridian Auto Insurance's approved rental-reimbursement rule.” That's what makes the answer checkable instead of a black box, and it's what makes NOT SPECIFIED useful instead of just an error: the gap is visible, attributable to a specific client and situation, and fixable in the Profile.

5. Human control

The agent decides. Agent First never talks to the customer.

Agent First works from the client's approved operating rules. It can help interpret the request. It doesn't get to invent the policy, and it doesn't get to speak for the agent. The agent reads the verdict, reads the suggested wording, and decides what to actually say. On every call, every time.

The trust model, in one line

Approved rules → Active Profile → Explicit outcome → Source shown → Human decides

See it work on your own rules.

Try the interactive demo, or apply for a pilot that begins with your real workflow and builds the Client Profile from what the work actually requires.

Try the live demo Apply for the pilot program