Skip to content

How it works

A sales team that learns from every outcome

Wino AI’s agents write, send and read replies. Then they look at what happened and adjust. Here is the loop, step by step, in plain language.

The Wino AI learning loop1. Capture: Every agent run is logged with its context. 2. Match: Replies, approvals and edits are tied back to the run. 3. Learn: Wins, misses and your edits become learnings. 4. Apply: The most confident learnings shape the next draft. 5. Test: Improved prompt variants are measured on real outcomes. 6. Promote: A proven variant becomes the live version.Learnfrom every outcome01Capture02Match03Learn04Apply05Test06Promote
  1. 01

    Capture

    Every agent run is logged with its context

  2. 02

    Match

    Replies, approvals and edits are tied back to the run

  3. 03

    Learn

    Wins, misses and your edits become learnings

  4. 04

    Apply

    The most confident learnings shape the next draft

  5. 05

    Test

    Improved prompt variants are measured on real outcomes

  6. 06

    Promote

    A proven variant becomes the live version

At a glance

Six steps, running in the background

Agents do the work. The loop watches the results. You decide what changes for good.

  1. 01

    Capture

    Every agent run is logged with its context

  2. 02

    Match

    Replies, approvals and edits are tied back to the run

  3. 03

    Learn

    Wins, misses and your edits become learnings

  4. 04

    Apply

    The most confident learnings shape the next draft

  5. 05

    Test

    Improved prompt variants are measured on real outcomes

  6. 06

    Promote

    A proven variant becomes the live version

Step by step

The loop, explained

  1. Step 01

    Capture

    Every time an agent does something, Wino AI writes it down.

    Each run is logged with the campaign phase, the lead’s warmth and priority at that moment, and a short summary of what the agent did. That record is what later outcomes get attached to.

    Your part: Nothing to do. It happens on every run.

  2. Step 02

    Match to outcome

    When something happens, it gets traced back to the run that caused it.

    A prospect replies. You approve a draft. You change a line before sending. Each of those results is attached to the exact agent run that produced the message, so cause and effect stay connected.

    Your part: Approve, edit or dismiss drafts as you normally would. Every decision counts.

  3. Step 03

    Learn

    Clear patterns become lessons for your workspace.

    Clear wins, clear failures, your edits and phase-specific insights turn into learnings. Each one carries a confidence level, and they belong to your workspace, not anyone else’s.

    Your part: Your edits are one of the strongest signals. Rewrite a draft and future drafts drift toward your style.

  4. Step 04

    Apply

    Before the next draft, the agent reads what has worked.

    The most confident learnings for that agent and phase are added to its instructions, within a strict size budget so the signal stays sharp. The message writer also uses what the reply reader has learned.

    Your part: Open any draft in the approval queue to see which learnings fired, and how confident each one is.

  5. Step 05

    Experiment

    Wino AI tries better versions and keeps score.

    From those learnings it builds improved prompt variants, then tracks each variant’s reply rate and meeting rate against the baseline, with sample sizes, so a lucky week does not look like a trend.

    Your part: Watch variant results on the Agents page before anything changes for good.

  6. Step 06

    Promote

    A proven winner becomes the new normal.

    When a variant has clearly beaten the baseline, it can be promoted to the live version for that agent in your workspace.

    Your part: Promotion is a deliberate step, not something that happens behind your back.

Forget and fall back

Old lessons fade out

A tactic that worked last quarter should not run your outreach forever.

  • Confidence decays

    Learnings lose confidence week by week and are removed once they are no longer reliable.

  • Off switch, safe defaults

    Learning can be switched off at any time. Content is cleaned before it reaches a prompt, and agents fall back to their default rules if anything looks wrong.

Illustrative example of the “learnings that fired” note on a draft.

You stay in control

What you decide

The loop does the bookkeeping. The decisions that matter stay with you.

  • What sends on its own

    Approval rules by phase, first message in a phase, prospect title or warmth threshold decide what auto-sends and what waits for you.

  • What the agents learn

    Your approvals and edits are the training signal. Change a draft and you are teaching the agent, not just fixing one message.

  • What you can see

    Every draft shows the agent’s verdict, confidence, reasoning, the learnings it used and any red flags.

  • What goes live

    Variant results are visible with sample sizes. Promoting a winner is your call.

  • How hard it pushes

    Daily limits for connections, messages and profile visits, plus working hours and time zone.

  • Whether it learns at all

    Learning can be switched off at any time. Agents then run on their default rules.

FAQ

Questions, answered

Does Wino AI send messages without asking me?

Only if you tell it to. Out of the box, drafts wait in your approval queue. Approval rules let you decide what can auto-send, for example by phase, job title or warmth, and everything else waits for you.

What does the learning actually learn from?

Real outcomes in your workspace: replies and their sentiment, which drafts you approve, and the edits you make before sending. Those signals are matched back to the agent run that wrote each message.

How quickly does it improve?

It depends on volume. Learnings need real outcomes behind them, so a busy campaign gives the loop more to work with than a quiet one. Variant results show sample sizes so you can judge for yourself.

Can a bad lesson stick around?

Learnings lose confidence week by week and are removed once they are no longer reliable. Content from the learning system is cleaned before it is added to prompts, and agents fall back to their default rules if anything looks wrong.

Can I turn learning off?

Yes. Learning can be switched off at any time, and the agents keep working on their default rules.

Is my data used to train other workspaces?

Learnings are stored per workspace and applied to that workspace’s agents. Your edits shape your drafts.

What stops the agents from being pushy?

Campaigns move leads through phases, and meeting asks are held back until the right phase. A clear “not interested”, “wrong person” or out-of-office reply is caught by fixed rules that the AI cannot override.

Can I see how an agent will behave before going live?

Yes. Pick a test contact, move them to any phase and simulate a chat to see exactly how the agent would respond, including whether it would reply, wait, escalate to you or park the lead.

Still curious? Book a demo and we will walk you through it.

Further reading

New to AI SDRs?

All articles

Start the loop on your own pipeline

Connect LinkedIn and email, describe your ideal customer, and approve your first drafts. The learning starts with your first decision.