# AI SDR for lean teams

> A play for one-to-three person GTM teams that want outbound running without an SDR hire — Jason AI in Approval mode, one Offer, one Playbook object, a knowledge base with Re-engagement follow-ups, and Autopilot held to its defaults.

**Start Jason AI in Approval mode, where every generated message waits on the Pending approvals page until you approve, edit, or regenerate it. One Offer, one Playbook object, and a knowledge base ground what it writes. Keep Autopilot sourcing at its default ten contacts a day, and switch to Automatic mode only after approvals stop needing edits.**

_Status: Reviewed — human-edited, facts not yet confirmed against the product._

## Situation

Your GTM team is one to three people wearing every hat. Outbound matters, but nobody has
the hours to research, write, send, and follow up — and hiring an SDR is premature. You
want the outbound motion running this quarter without adding headcount, and without
letting an unsupervised agent speak for the company.

## Who this is for

| **Role** | Founder, first marketer, or a one-person sales team |
| **Company size** | Startup with a working product and at least a few reference customers |
| **Motion** | Wants a steady outbound motion without an SDR hire |
| **Prerequisite** | A sellable offer you can describe honestly, and one hour a week for review |

## The play

### 1. Targeting

Define one ICP and hold it. Jason AI can suggest the ICP from your website plus a few
inputs like pain points and value, then source prospects against it in real time from
LinkedIn data using filters such as location, industry, and job title — or you bring your
own list; [find prospects from an ICP](/workflows/find-prospects-from-an-icp) covers both
routes. With Autopilot on, new contacts are added daily; the Sales Navigator Autopilot
source defaults to 10 contacts a day and 3 contacts per company (500 a day is the
maximum). Keep the defaults while trust builds; a second ICP comes only after the first
produces meetings.

### 2. Sequence structure

Jason AI generates the sequence from your inputs — emails and LinkedIn steps are added
automatically, and you can add calls, WhatsApp, SMS, or manual tasks. The table below is
an illustrative generated shape, not steps you author by hand.

| Step | Channel | Day | Purpose |
| --- | --- | --- | --- |
| 1 | Email | 1 | Opening message generated from the Offer and knowledge base |
| 2 | LinkedIn | 4 | Profile touch or connection request |
| 3 | Email | 7 | Follow-up angle: proof point or objection preempt |
| 4 | Email | 14 | Closing note with a clear ask |

### 3. Messaging angles

- The Offer carries who you sell to and what you offer them — audience, pain points, priorities; the knowledge base carries the product facts the AI must not invent
- When you create a sequence, Jason generates five Offers from your website and business inputs — one pre-selected, four more in the Offers section — each a different angle on the same product, differing by ICP and pain points
- Vague inputs produce vague messages — "we help companies grow" in the Offer guarantees generic output
- Fix problems upstream: editing an Offer with **Update original** improves every sequence that uses it; **Keep separate** forks a variant for one sequence so you can test without side effects
- The Playbook object sets the strategy layer — channels, structure, pacing — so generation stays consistent

### 4. Timing and cadence

The sending mode is the supervision dial, chosen per sequence. Start in **Approval
mode**: every generated message lands on the Pending approvals page (AI SDR → Pending
approvals) and nothing sends until you approve it. In the queue you can edit a message
directly, click **Regenerate** with a short prompt to steer a new version, and approve
one by one or in bulk; approved messages still respect the sequence's schedule and
limits. Deleting a queued message removes that prospect from the sequence entirely — it
is a rejection, not a skip. Your likes, dislikes, and edits feed Jason's AI Learnings —
inferred writing-style rules you can view, edit, or disable from the AI SDR tab — so
early reviewing compounds. Switch a sequence to **Automatic mode**, where Jason generates
and sends without waiting, only after the queue has stopped needing edits. Hold a fixed
weekly review either way; the model is described in
[the AI SDR model](/learn/ai-sdr-model).

### 5. Handling replies

Jason AI handles incoming replies your way: it either saves drafts for your review or
sends responses automatically — keep drafts longest, because this is where trust is won
or lost. It can also cover common replies like meeting requests and objections. For
threads that go quiet, configure Re-engagement in the knowledge base: each follow-up
fires a set number of days after the last reply (for example one after 2 days, another
after 5), in Auto or Draft mode, referencing the previous conversation on email or
LinkedIn. See [reply classification](/learn/reply-classification) for how the sorting
works.

## Templates

You do not write the messages — Jason AI does. What you control is the style: the
knowledge base and Re-engagement settings take instructions, a sample answer written in
your voice, and a tone of voice picked from confident, persuasive, witty,
straightforward, or empathetic. The example below is illustrative — a style reference you
might store, not a sourced product template.

```
Subject: {{PainPoint}} at {{Company}}

Hi {{FirstName}},

Noticed {{Trigger}} — in our experience {{PainPoint}} is usually close
behind.

{{MyCompany}} does {{OneLineWhatItDoes}}. {{CustomerExample}} used it to
{{ConcreteOutcome}}.

Open to 20 minutes next week?

{{SenderName}}
```

## Expected results

| Metric | What to expect | Conditions |
| --- | --- | --- |
| Email accuracy | Jason checks each email address before sending and aims for up to 98% accuracy | Product behavior — a deliverability guardrail, not a reply-rate promise |
| Approval workload | Front-loaded, then drops — AI Learnings turns your likes, dislikes, and edits into standing writing-style rules | Requires actually working the queue, not rubber-stamping it |
| Reply quality and meetings | Results depend on list quality and offer — Reply does not publish benchmark rates | The Offer and knowledge base set the ceiling on output quality |

## Common mistakes

- Starting in Automatic mode before Approval mode has proven the output
- Writing a vague Offer and expecting specific messages
- Correcting individual drafts forever instead of fixing the Offer and knowledge base
- Deleting queued messages casually — deletion removes the prospect from the sequence, it is not a skip
- Skipping the weekly review once things "seem fine"
- Running three ICPs at once before one of them works

## Build this in Reply

1. Follow [build an AI SDR](/workflows/build-an-ai-sdr) end to end — it is this play as a step-by-step workflow.
2. Create one [Offer](/specifications/offers): paste your website and let Jason generate the angles, then edit the audience, pain points, and proof points honestly.
3. Create one [Playbook object](/specifications/playbooks) for strategy and one [knowledge base](/specifications/knowledge-bases) with product facts; add Re-engagement follow-ups there.
4. [Connect a mailbox](/how-to/connect-a-mailbox), keep [sending limits](/how-to/configure-sending-limits) conservative, and leave Autopilot at its default 10 contacts a day and 3 per company while trust builds.
5. Launch with Approval mode selected as the sending mode — the [AI SDR capability](/capabilities/ai-sdr) page covers the modes — and hold the weekly review before considering Automatic.

## Related

- [Build an AI SDR](/workflows/build-an-ai-sdr)
- [The AI SDR model](/learn/ai-sdr-model)
- [AI SDR capability](/capabilities/ai-sdr)
- [Playbook object](/specifications/playbooks)
- [Outbound for seed-stage founders](/playbooks/outbound-for-seed-stage-founders) — the manual version of this motion

## Build with Reply

- REST API: [docs.reply.io](https://docs.reply.io/api-reference/introduction) — offers, playbooks, and knowledge-base endpoints
- MCP: [agents.reply.io/mcp](https://agents.reply.io/mcp) — agents create Offers and Playbooks and work the approval queue
- CLI: [agents.reply.io/cli](https://agents.reply.io/cli) — scripted setup and review pulls
- Agent skills: [agents.reply.io/skills](https://agents.reply.io/skills) — the build-an-AI-SDR workflow as an installable skill