Skip to main content

Outcome-Based Agents

An Outcome-Based Agent is an AI agent FunnelStory builds and ships to reliably produce one specific outcome for one specific workflow — investigate a signal and draft the right response to it, for example. Where AI Agents are graphs you configure to automate whatever your team needs, an Outcome-Based Agent is pre-built around a single job, tuned specifically for it, and available the moment the underlying feature is on. There's nothing to author — at most, a setting to turn on.

The shape: read evidence, reason, act​

Every Outcome-Based Agent follows the same shape. It's grounded in real evidence pulled live from your workspace — conversations, relationship history, commercial terms, prediction data — not a generic prompt. It reasons through a tool-calling loop rather than a single LLM call, so it can pull more evidence mid-investigation if the first pass isn't enough. And it ends by either taking an action directly or producing something a person reviews before acting on it.

That shape is what turns a two-step manual habit — notice a problem, then go dig into it yourself — into one: detect → investigate → act, with the middle step no longer resting entirely on a person's time.

Two modes of the same agent​

An Outcome-Based Agent can run two ways, and both draw on the exact same underlying agent — same tools, same reasoning, same bar for what a good output looks like. They differ only in what triggers them and what happens with the result:

  • Assist mode — a person asks for it, on demand, inside the relevant workflow. The agent's reasoning streams live, and a person is present to review what it decides to do.
  • Auto mode — the agent runs unattended, triggered automatically as the relevant event occurs, with no one asking for it and no one watching it run.

Both modes matter for different reasons. Assist mode gives a person a faster way to do work they were already going to do. Auto mode means the work sometimes gets done before anyone thought to ask — the loop can close itself.

The first example: investigating a Needle Mover​

The first Outcome-Based Agent FunnelStory shipped investigates Needle Movers — the leading indicators of churn or expansion risk FunnelStory surfaces from your conversations, product usage, and support data.

  • In Assist mode, it's the Investigation and Triage chat on a needle mover's detail view — ask it to look into the signal, and it can draft a follow-up Action Card, outreach email included, directly in the conversation.
  • In Auto mode, it's autosuggested action cards — turned on per workspace, it investigates open needle movers on a schedule and drafts a card before anyone opens them, with no agent configuration required.

Both modes read the same account evidence, apply the same judgment for what makes a good recommendation, and land in the same place: an Action Card ready for a CSM to act on.

How this differs from AI Agents​

FunnelStory uses "agent" for two genuinely different things, and it's worth keeping them apart:

AI Agents (flow builder)Outcome-Based Agents
Who builds itYou — on a canvas, in JSON, or via Vibe CodingFunnelStory's engineering team
ScopeWhatever your workspace configures it to doOne specific job, built for that job
Where it shows upWherever you wire a trigger to point itBuilt into the specific feature it serves
SetupAuthor a trigger, steps, and functionsNothing to author — turn a setting on, if there is one

They're related, not competing: an Outcome-Based Agent is what a specific, high-value workflow looks like once FunnelStory has built and hardened it for you, rather than left for you to assemble yourself in the flow builder.