- Oct 26, 2026
- 4 min read
The Sunday Outbound Review: One Hour, One Decision
The outbound pipeline from the earlier series turns map listings into a short list of companies, each with one measured problem on its own website. Building it took a few weekends. Running it is a weekly habit: one hour every Sunday, ending in one decision.
The loop
flowchart LR
X[Export from the<br/>sending tool] --> L[Log replies<br/>by hand]
L --> F[Rebuild the funnel]
F --> R[Read 3 numbers]
R --> D[Change one thing<br/>+ write it down]
D --> B[Build Monday's batch]
B --> X
1. Export (5 minutes)
The sending tool’s plan I’m on has no API, so the data comes out as files: one lead export per campaign (status, sequence step, bounced, replied) and the send logs (how many went out each day, per inbox). They go into a dated folder as downloaded, never edited. That folder is the bronze layer of a small local DuckDB file.
2. Log what the export can’t see (10 minutes)
I read every reply myself and log it with one command:
make event [email protected] E=optout # or bounce | autoreply | positive | call
An opt-out also goes on the sending tool’s global blocklist, and on a local suppression list that the next batch is checked against. Reply text is never loaded into the database; only the label.
The autoreply label exists because of week one: an office
auto-acknowledgement counted as a reply, and the tool stopped that lead’s
follow-up. Now auto-replies are logged as what they are and don’t count.
3. Rebuild the funnel (1 minute)
make funnel
Silver tables are rebuilt from scratch every time (one row per business, per send, per outcome). Gold is counts only: per batch, per finding type, per inbox, per day. Gold is the only part committed to git, so the history of the funnel survives the 90-day deletion of the personal data under it.
A detail that matters: a lead in my batch file but missing from the tool’s export is marked not sent. Imports quietly drop a few rows (duplicates, addresses the tool rejects), and rates should be over what actually went out.
4. Read three numbers (10 minutes)
| Number | Threshold | Action |
|---|---|---|
| Bounces + opt-outs | above 3% | Stop and check the list |
| Human replies | under 1% after ~100 sends | Change one thing |
| Positive replies | any | Answer the same day; that’s the whole point |
5. Change one thing, and write it down (15 minutes)
When replies are flat, change one variable, in order of cost: the subject line first, then the type of finding, then the trade, then the offer. Change two at once and you’ll never know which one worked. Each change gets a short decision record: what changes, what stays, and how it will be judged, before the results exist.
6. Build Monday’s batch (20 minutes, plus the scrape running)
make runs TRADES="…" AREAS="…" MAX=55 # scrape → enrich → audit
make send RUNS=… BATCH=b5 # batch rules, writes bN.draft.csv
make check BATCH=b5 # no repeats, nothing suppressed, every field filled
make mark-sent BATCH=b5 CAMPAIGN="…" # after the import and launch
Between send and check I read every row by eye. The rules catch the
legal and technical problems; only a person catches “that’s a supplier,
not the trade I searched for”.
What two weeks returned
| Companies contacted | 117 |
| Emails sent (with follow-ups) | ~190 |
| Bounces | 2 (1.7%) |
| Human replies | 1 (an opt-out) |
| Auto-replies | 1 |
| Positive replies | 0 |
Zero. Not a verdict at this volume (cold email reply rates of 1–2% are normal, so ~100 sends can easily return nothing), but enough to trigger the rule: change one thing. The next post in this series is about the limit I hit first, and it wasn’t leads.
What the review caught besides the numbers
- The campaign timezone was wrong. Emails went out between 8 and 11 in the morning instead of across the working day.
- Open tracking was back on in campaigns made by duplicating an old one.
- The daily cap was full Monday to Wednesday and half-empty on Thursday and Friday.
None of these shows up in a reply rate. All of them showed up in the send logs, which is why I keep exporting them.
Takeaways
- One hour, one decision, every week. The decision is the output.
- Log replies by hand, with labels; never store the text.
- Rates over what was really sent, not what you uploaded.
- Write down how a change will be judged before you make it.