Build an automated sales process in seven steps: write it by hand, automate research first, ramp volume, set stop rules and loosen approvals in stages.
An automated sales process is a written sequence of selling steps that software runs for you, with people approving the decisions. Build it by doing the process by hand first, automating research and sorting before writing and sending, ramping volume slowly, and loosening approvals only after a stage hits its pass mark. Write the stop rules before anything sends.
Most guides to an automated sales process tell you which tools to connect. That's the easy half. The hard half is deciding how much freedom each step gets, and when. This is the build order we'd follow, with the approval points marked. If you want the mechanism first, read how B2B sales automation works, then come back here to build one.
An automated sales process is a documented sequence of selling steps, from choosing accounts to booking a meeting, where software performs the repeatable steps and a person approves the judgement calls. It isn't a tool you switch on. It's a process you've already written down, with some steps handed to a machine one at a time.
G2 puts the starting point plainly: map every step a lead takes from first touch to closed-won, with each step's owner, tool and outcome captured before any automation is configured (G2). We agree, and we'd add one rule. Every step gets an autonomy level, and it starts low.
You need four things: a named buyer and a one-line reason they'd answer, one place that records accounts and stages, a sending setup with warmed inboxes and authenticated domains, and one person who owns approvals. Without those, you'd be automating guesses. Check your domains with the email domain checker before step 4.
Build it in seven steps, in this order. Each step has a pass mark, and you don't move on until you've met it. The order matters more than the tools: reading and sorting come before writing and sending, because mistakes in the first two are cheap and mistakes in the last two reach real inboxes.
Write each stage on one page: who it's for, what triggers the next step, who owns it, and what "done" looks like. Then run it manually on about 50 accounts. Sending 50 emails by hand feels slow, and that's the point. You'll find the segment that's too broad, the claim you can't back up and the handoff nobody owns, before software multiplies any of them.
Pass mark: you can explain in two sentences who you target and why they'd reply, and you've read every reply by hand.
Remove anything that doesn't move a deal forward. Three follow-ups that say the same thing become one better one. Artisan's guide makes the same point from the tooling side: start with the small set of repetitive tasks that eat most of your time, not everything at once (Artisan). Automating a bloated process just gives you a faster bloated process.
Automate account research, contact verification and the sorting of replies before anything that writes or sends. Mistakes here stay inside your system. A bad research note gets caught in review. A bad email gets sent. Salesforge lists unverified lead lists among the common pitfalls of outbound automation (Salesforge), which is why verification belongs first.
Pass mark: spot-check 25 records. If you find more than a couple of wrong contacts or stale facts, fix the data source before continuing.
Let the system draft emails from the research, but approve each one yourself in this stage. You're testing whether the drafts say true things in your voice. Our post on how AI SDRs personalise emails covers what a good draft is built from. Keep a log of why you reject drafts, because the reasons become your written claim rules.
Pass mark: you approve at least 9 in 10 drafts without edits, over a run of 50. That threshold is our rule of thumb, not an industry figure. Set your own, but set one.
Start small and raise volume only while the numbers stay clean. Google's guidance says to increase volume gradually, and warns that suddenly doubling what you previously sent can bring rate limiting or reputation drops. It also says to reduce volume if messages start bouncing or being deferred (Gmail sender guidelines).
Here's a worked example using our published limits of up to 40 emails per inbox per day. With 5 warmed inboxes the ceiling is 200 a day, about 4,000 over 20 sending days. You don't start there. A plan we'd use: 50 emails in week one, then 100 a day, then the ceiling, only if bounces and complaints stay under your limits. The numbers are ours, so adjust them. The cold email capacity calculator shows how many inboxes a target volume needs.
Decide in advance what pauses the system, who gets told and who restarts it. A rule nobody owns is a rule nobody follows. Put the thresholds in a table you can read at 7am on a bad day.
| Signal | Threshold | Action | Source of the number |
|---|---|---|---|
| Spam rate in Postmaster Tools | Keep below 0.3% | Pause the inbox, review the copy and list | Google's sender guidelines |
| Bounce rate | Above 2% | Pause the campaign, re-verify the list | Outbound Labs auto-pause setting |
| Opt-out request | Honour within 10 business days at most | Suppress the contact across every campaign | FTC CAN-SPAM guide |
| Any interested, pricing, legal or security reply | Every one | Stop the sequence, hand to a person | Our rule |
| Draft rejection rate | Your own pass mark, from step 4 | Return to approve-every-message | Our rule |
The FTC states that CAN-SPAM makes no exception for business-to-business email, so the opt-out rule applies to you (FTC). We'd set a stricter rule than 10 days for your own process: suppress the contact the moment they ask. G2 recommends the same on the sequence side, stopping a sequence as soon as a prospect replies, books or unsubscribes.
Move through four stages of autonomy and promote only when the pass mark is met. Stage one is everything by hand. Stage two is the machine drafts and you approve each message. Stage three is you approve the campaign (segment, claims, sample emails, limits) and the machine sends inside it. Stage four is the machine runs inside those limits while you watch the stop signals. Our guide to AI SDR human in the loop goes through what an approval card should show.
| Stage | Who sends | Who approves | Move up when |
|---|---|---|---|
| 1. By hand | You | You | You can state who you target and why they'd reply |
| 2. Draft only | You, after review | Every message | 9 in 10 drafts pass unedited over 50 |
| 3. Campaign approval | The system, within limits | Segment, claims, launch | Two weeks with bounces and complaints under the stop rules |
| 4. Monitored | The system | Exceptions only | Stays here. Drop back a stage if any stop rule trips |
Stage four isn't "set and forget". It's one person looking at the stop signals and the interested replies every day. Dropping back a stage after a bad week isn't a failure. It's the process doing its job.
Automate last anything that makes a commitment or touches a relationship: pricing, contracts, security and legal answers, renewals and strategic accounts. Never automate the live discovery call. One wrong answer on a security questionnaire can end a deal, and no hour you saved is worth that. We keep these with a person permanently, not just until the tool improves.
Check these before each promotion to a higher stage. If any line is a "no", stay where you are.
| Check | Yes / no |
|---|---|
| The process is written down with an owner for every step | ☐ |
| It was run by hand on about 50 accounts and every reply was read | ☐ |
| A 25-record spot check of the list found no stale contacts | ☐ |
| SPF, DKIM and DMARC pass and the inboxes are warmed | ☐ |
| Drafts pass your own approval threshold | ☐ |
| Volume ramp is written down, with the daily ceiling | ☐ |
| Stop rules have thresholds, an owner and a restart rule | ☐ |
| Opt-outs suppress the contact everywhere | ☐ |
| Interested, pricing, legal and security replies route to a person | ☐ |
The common ones are skipping the manual run, automating writing before sorting, jumping to full volume, and having no owner for stop rules. Each one lets a small error repeat at scale. A fifth is measuring only meetings, which lag by weeks, instead of the stage-level tests that tell you what broke.
Monday.com's seven-step flow ends with tracking performance metrics, and we'd go further and track them per stage (monday.com). A good meeting count on a bad bounce rate is a domain about to get burned. Our guide to deliverability guardrails for AI-sent email covers those checks in more depth.
Outbound Labs is built around the staging above. Six agents (Researcher, Strategist, Prospector, Writer, Sender and Closer) cover steps 3 to 5, you approve every campaign before anything sends, and sending auto-pauses above a 2% bounce rate. The AI SDR page shows the approval flow, and the trial includes 750 emails with no card, enough to run stage two properly. Start the free trial when your hand-run list is ready.
SDR agency vs in house SDR compared on speed to meetings, control, management load, contracts, meeting quality and attrition, with a switch-over checklist.
Eathan Brown · 2 Nov 2026 · 8 min readA Lindy AI alternative for every job: Zapier Agents, n8n, Gumloop, Relevance AI, Apollo, Salesforge, AiSDR and Outbound Labs, with a cost for 500 prospects.
Eathan Brown · 1 Nov 2026 · 9 min readAutonomous SDR defined: what the label means, the autonomy levels from drafts to autopilot, how it differs from an AI SDR or sequencer, and a 4-question test.
Eathan Brown · 31 Oct 2026 · 5 min readA go to market strategy for startups that starts with outbound: when it fits, how to pick the first channel, the capacity maths and a 30/60/90 plan with stop rules.
Eathan Brown · 31 Oct 2026 · 9 min read