The short answer
An agent action is a verified outcome that an AI agent completes for someone, such as delivering a quote, submitting a lead or booking a meeting. It is the unit of work that matters, rather than how many messages the agent sends or how long a conversation lasts. Agent actions do two jobs at once: they tell the agent what to work toward, and they define what counts as completed work when you measure or pay for it.
As AI agents move from answering questions to getting things done, the question every team asks changes from “How many chats did it have?” to “What did it actually do?” Agent actions are how you answer it.
Agent actions and conversations aren’t the same thing. A conversation can end with zero actions, when a visitor just browses or asks a quick question, or with several actions, when the agent delivers a quote and then books a meeting. Counting actions shows what the agent achieved; counting conversations only shows that people showed up.
Other names for agent actions
The industry hasn’t settled on one term yet. Depending on the vendor or the context, you’ll see agent actions called:
| Term | Where you’ll see it |
|---|---|
| Outcomes, or agreed outcomes | Intercom’s Fin bills per “outcome”; Sierra prices on outcomes agreed with each customer |
| Resolutions, or automated resolutions | HubSpot’s Breeze customer agent bills per resolution; Zendesk charges for automated resolutions |
| Actions | Salesforce Agentforce’s Flex Credits are consumed per action the agent takes |
| Goal completions or conversions | Web analytics and marketing teams, for the same idea measured on a website |
| Billable outcomes, or pay-per-outcome | Finance and procurement teams describing how agent work is charged |
The pricing model behind all of these is usually called outcome-based pricing or results-based pricing. Whatever the name, the idea is the same: measure, and pay for, the work an agent completes rather than the activity around it.
Why conversations are the wrong way to measure AI agents
Most AI chat tools are measured, and priced, by activity: conversations, messages, tokens or seats. Those numbers are easy to count, but they say little about value. A 40-message conversation that ends with a frustrated visitor counts the same as a three-message conversation that books a qualified demo.
Activity metrics also pull in the wrong direction:
- They reward length, not results. An agent that takes longer to help looks busier.
- They hide what the agent is for. “10,000 conversations” tells a CMO nothing about pipeline, support deflection or revenue.
- They make budgets unpredictable. Usage spikes from curious visitors or bots cost the same as real buyers.
- They blur accountability. When nobody has agreed what “done” looks like, nobody can say whether the agent is working.
Agents are different from chatbots because they take actions: they submit forms, call tools, update systems and hand work to people. It makes sense to measure them by the actions they complete.
What makes up an agent action
Each agent action has two parts: a name that states the result, and a rule for when it counts. Together they make a short, written agreement that the agent, your team and your finance lead all read the same way.
- Name: the business result, in plain words. For example, “Meeting booked”, “Quote delivered” or “Lead submitted”.
- Counts when: the proof that it happened. For example, “a confirmed calendar event exists for the visitor” or “the quote was shown to the visitor in the conversation”.
A good “counts when” rule describes the finished result, not the steps on the way. Agents often do smaller things to get there, like collecting an email address so they can book a meeting. Those steps are part of the outcome they serve, so they don’t count on their own. That keeps the numbers honest and stops the agent being rewarded for busywork.
How agent actions are counted: four examples
The simplest way to understand agent actions is to look at real conversations. Each row below is one visitor conversation with a website agent.
| What the agent did | Agent actions | Why |
|---|---|---|
| Collected the visitor’s contact details, then booked a meeting | 1 | Collecting details was a step toward the booking, so it is included |
| Delivered a quote, submitted a separate sales lead, then booked a meeting | 3 | Three distinct outcomes, each with its own evidence |
| Delivered a quote, collected an email only to book, then booked a meeting | 2 | The quote and the meeting count; the email was part of the booking |
| Retried a failed submission, or repeated the same quote | 0 extra | Retries and repeats of the same outcome don’t count again |
Two principles sit behind these examples. First, only verified completion counts: a meeting counts when the booking is confirmed, not when the agent says it booked one. Second, one outcome is counted once, however many tries or messages it took to get there.
Why agent actions matter
Clear goals make the agent work toward results
This is the most important reason to define agent actions, and it is easy to miss. An agent with clear goals works toward them. An agent without them just answers questions and hopes something good happens.
When an agent knows that “Quote delivered” and “Meeting booked” are its goals, it recognizes the moments in a conversation when those outcomes fit, and it moves the visitor toward them: offering a quote when someone asks about pricing, or suggesting a meeting when a buyer is clearly evaluating. Defining actions is how you turn a helpful chat into an agent with a job.
Clear goals also keep the agent focused. It pursues the outcomes that match the visitor’s intent, without forcing every conversation toward every outcome. A visitor who only wants a spec sheet gets the spec sheet.
They connect agents to business results
Agent actions map directly to the numbers leaders already track: meetings, qualified leads, resolved support cases, completed applications. Instead of reporting “8,200 conversations”, a team can report “412 meetings booked and 1,150 quotes delivered”. That is a number a CMO or CFO can act on.
They make cost predictable and fair
When you pay for completed outcomes rather than activity, cost follows value. Long conversations, curious browsers, retries and bot traffic don’t drive your bill, which is why Webless includes unlimited conversations on every plan. You can forecast spend from the outcomes you expect, the same way you forecast pipeline.
They build trust through evidence
Every counted action has a record behind it: what happened, when, and the proof, such as a booking or submission ID. Your team can click into any count and see exactly why it counted. Nothing is billed on the agent’s word alone.
They make agents easier to govern
Writing outcomes down forces a useful conversation: what should this agent do, and what should it never do? That clarity makes agents easier to review, approve and change safely as your business evolves.
How agent actions work in Webless
In Webless, agent actions are built into how every website agent is created, approved and measured. Each agent has a short list of intended agent outcomes, and those outcomes are both its goals and the definition of what counts.
Every Webless plan includes unlimited conversations. Whichever plan you choose, visitors can talk to your agents as much as they like, and you never pay for chatter, long conversations or curious browsers. Only the verified actions your agents complete count toward your plan.
- Generated with the agent. When you describe an agent in plain English, Webless proposes its outcomes alongside its instructions and skills, based on what the agent is for, the content it knows and the tools it can use.
- Reviewed and edited by you. Each outcome has a name and a clear rule for when it counts. You can add, edit, reorder or discard outcomes, and restore any you discarded.
- Checked for gaps. Before an agent goes live, Webless checks that it has at least one measurable outcome, and that everything the agent can do on its own is covered by an outcome. If you remove “Lead submitted” but the agent still offers standalone sales follow-ups, Webless flags the mismatch for you to fix rather than rewriting your setup.
- Confirmed before going live. You see every active outcome, how each one is counted, and what changed since your last approval. You confirm explicitly, and that confirmation is stored against that exact version of the agent.
- Verified in every conversation. System outcomes, like a booking or submission, are verified against the system that completed them. Conversational outcomes, like a delivered quote, are checked against the conversation itself. The agent can suggest that it completed something, but it can’t authorize a count.
- Visible in your dashboard. Each conversation shows its verified actions, with names, times and evidence. Your plan usage shows completed actions for the period, kept separate from actions still being verified and from preview testing, which never counts.
Editing an agent later doesn’t change how past conversations were counted. Each conversation stays tied to the version of the outcomes that was live when it happened, and meaningful changes need a fresh confirmation before they take effect.
Removing an outcome is a real change to what the agent does, not just a way to lower a count. If an agent keeps doing something you discarded, Webless surfaces it for review instead of charging for it, so you can decide whether to restore the outcome or remove the capability.
How to define good agent actions
Whether you use Webless or another platform, the same rules make agent actions useful:
- Name the result, not the activity. “Meeting booked” is an outcome; “asked qualifying questions” is a step.
- Tie each outcome to evidence. If you can’t point to a record that proves it happened, such as a booking ID, a CRM entry or a delivered document, it isn’t measurable yet.
- Fold prerequisites into the outcome they serve. Collecting an email to book a meeting is part of the booking.
- Keep standalone outcomes separate. If the agent can submit a lead on its own, without booking, that is its own outcome.
- Make “counts when” specific. A rule like “a confirmed booking exists” automatically rules out test conversations, unused links and duplicates.
- Count each outcome once. A second quote for a different product may count; re-sending the same quote shouldn’t.
- Start with two or three. Most agents need only a few outcomes. Add more once you see how visitors actually use the agent.
For more on the vocabulary, see our Agentic Glossary for Marketers. For examples of outcomes by team, read website agent outcomes by team.
Measure agents by what they get done
Conversations show that people are talking to your agent. Agent actions show what it actually got done for them and for you. With Webless, conversations are unlimited on every plan, and you measure agents by the actions they complete. Defining outcomes up front gives the agent a clear job, gives your team numbers it can stand behind, and keeps your spend in line with the value you get.
New to website agents? Start with What Are Website Agents?
Ready to launch agents measured by outcomes? Get started with Webless Website Agents.
