Full sales outreach loop (research → draft → approve → send → log)

Full sales outreach loop covers five steps of cold outbound flow in a single orchestration: lead research, draft generation, manager approval, sending, and logging. The AI agent handles the routine; the human retains control over message quality and the decision whether to send or not.

What the automation does

1. Research. The agent collects lead data from the CRM and open sources: company website, contact role, recent events (releases, hiring, funding). Builds a brief profile of 4–6 points.
2. Draft. Based on the profile and tone template, generates a personalized first email: hook, value proposition, soft CTA. Does not use generic phrases like "hope you're doing well".
3. Approve. The draft goes to the manager in Slack, email, or the CRM interface. The manager edits, approves, or rejects with a single action. Without approval, the email is not sent.
4. Send. The approved email is sent via the connected email provider (the manager's mailbox or a dedicated outreach domain). Added to the follow-up sequence if configured.
5. Log. The agent logs the send, open, reply, and related activities to the lead's CRM card. Marks the funnel stage, creates a follow-up task if needed.

What the automation does NOT do

• Does not replace the manager at the approve step. The approval gate is built into the architecture — this is not a bug, it's a feature. The loop is designed for controlled scale, not for blast campaigns.
• Does not bypass spam filters. If the domain is not warmed up, deliverability will be low even with perfectly personalized content. Domain and mailbox warm-up is a separate preliminary task.
• Does not work with a dirty CRM. If the CRM has duplicates, outdated statuses, and missing required fields (role, industry, company size), the research step will return garbage and the draft will come out generic.

The result is a 10x increase in outbound contact volume (from 10–15 to 150–200 emails per day), open rate growth from 18% to 52%, and reply rate from 2.1% to 8.7%. The figures are achievable provided the domain is warmed up, the ICP is defined, and the CRM contains a minimum set of lead fields.

Under the hood — multi-agent orchestration on an agent-framework. Each step is a separate agent or node with clearly defined inputs, outputs, and responsibility boundaries. Grow2.ai assembles a loop from three classes of components: a state orchestrator, an LLM engine for generation, and integrations with CRM, email, and the approval channel.

Architecture diagram

1. Trigger. Entry into the loop — either on a schedule (daily batch of N leads) or on a CRM event (new lead in the «cold outbound» segment). Batch mode is used by default for controlled rate limiting.
2. Research agent. Retrieves lead_id from CRM, pulls company data (website, public profile, news), contact role and context. Uses external enrichment APIs and CRM fields. Output — a structured JSON profile.
3. Draft agent. Accepts profile + template guidelines (tone, CTA type, length) + examples of past successful emails. Generates a draft via LLM. Checks length, tone, and the presence of prohibited phrases through a rule-based filter.
4. Approval gateway. Draft + lead profile go to the manager. Format — a card in Slack, an email with inline buttons, or a task in CRM. The manager sees the context and one of three options: approve, edit-and-approve, reject.
5. Send agent. Triggered on approval. Connects to the email provider, sends the email on behalf of the manager (via OAuth or SMTP). Sets headers for open and click tracking.
6. Log agent. Writes to CRM: sent, opened, replied. Creates associated activities, updates the lead stage. On reply — moves the lead to the queue for manual follow-up.

System components

Implementation steps

1. Discovery (week 1–2). Audit of the current outbound process: where time is spent, which templates work, what information is collected manually. Defining ICP and segmentation criteria.
2. CRM hygiene (week 2–3). Deduplication, filling in required lead fields (role, industry, size), defining the status funnel for outbound.
3. Agent scaffolding (week 3–5). Building research/draft/send/log agents, connecting to CRM and the email provider. Manual testing on 20–50 leads.
4. Approval UX (week 5–6). Configuring the approval channel (Slack bot or CRM view), debugging latency between draft and approve. Goal — the manager spends no more than 30 seconds on approve.
5. Domain warmup (in parallel with weeks 1–6). If a dedicated outreach domain is used — warmup runs in parallel with the main track, gradually reaching the target volume.
6. Pilot + tuning (week 6–10). Launch on full batch of 100–200 leads per day, monitoring open/reply/unsubscribe, iterative revision of prompts and templates.

The architecture remains transparent: each step is logged, each agent decision is visible to the manager. The approval gate ensures that a bad draft never reaches the client.

To launch the full loop, you need lead data, access to working tools, and minimal team readiness.

Data and access

• CRM with filled lead fields: role, company, industry, size, funnel stage. Without this, the research step returns garbage.
• Email provider with OAuth access (Gmail Workspace, Microsoft 365) or SMTP. Preferably — a warmed domain or mailbox for outbound.
• Email templates and examples of successful cases. At least 5–10 outbound emails that worked, from which the agent can learn the style.
• A defined ICP (ideal customer profile): who we write to, what we look for, how we qualify.
• Approval channel: Slack workspace, corporate email, or CRM with inline-approve capability.

Team readiness

• One responsible manager (SDR or sales lead), who is ready to spend 30–60 minutes a day approving drafts during the first 2–3 weeks of the pilot. Later — 15–20 minutes.
• Technical role (in-house or from Grow2.ai): CRM integration, API access, agent-framework setup.
• Buffer for domain warmup — if the outreach domain is new, 3–4 weeks are needed in parallel with the main development.

Timeline

6–10 weeks from discovery to production launch for a team of 5–50 people. First 2 weeks — CRM audit and cleanup. Weeks 3–6 — agent build and testing. Weeks 6–10 — pilot with gradual scale and prompt tuning. Domain warmup runs in parallel, if applicable. After launch — 2–4 weeks of active reply rate optimization and personalization on real data.