Daily Call-Log Review Routine
What this is: the specification a scheduled cloud agent follows each weekday morning to review the previous day's calling against canon and report the exceptions.
This document is the routine's instructions. The scheduled agent is a thin pointer at this file, which means the review changes by editing canon rather than by editing a job nobody can see. If you want the routine to check something else, add it here.
Why it exists. The readiness plan §7 requires recorded calls reviewed by the Implementation Lead and delivered at the top of the next block. Before this routine that happened once, by hand. A trial criterion that depends on someone remembering is not a criterion.
It reads logs. It cannot listen. Transcription runs locally and the cloud agent has no access to it, so the listen-back half of §7 stays manual. The routine's job is to tell you which calls are worth listening to.
1. Scope
Sub-account vOyZYmChRuyjFM0yUtba. Yesterday's calls only. Ignore calls to the John Doe test contact.
Who the review is for
Resolve the agent from the Contractor Registry, the Vame entry's Placed VA row: their profile link gives the slug (contractors/vame/<agent-slug>-profile.md). The seat changes hands, so read it every run.
- Never infer the agent from the phone line's label, a folder that happens to have a
reportssubfolder, or the GHL user on the calls. All three have lagged a change of placement. - If their
reports\folder does not exist, create it insideTeam Members\<agent-slug>\, on the internal half, never inside the shared subfolder. - If the registry names no placed agent, or the slug has no
Team Membersfolder, write the report toTeam Members\_unassigned\reports\and say so on its first line. Do not pick a folder by guesswork.
Pull with the GHL connector: export-messages-by-location, channel: Call, a date window covering yesterday.
⚠️ The export caps near 200 messages per window and its cursor paging is unreliable. If the reported total exceeds what came back, narrow the window and stitch — and if the counts still do not reconcile, say so in the report. A partial count presented as complete is worse than no count. The same caution applies to get-campaign-stats, which disagreed with message history on 2026-10-29 and was wrong.
The pull, and the four traps in it
Verified against the live sub-account 2026-11-02. Each of these cost a wrong turn the first time.
| Operation | Use it for | The trap |
|---|---|---|
export-messages-by-location | The call log. channel: Call, a startDate/endDate window, limit 100 | sortBy accepts only createdAt or updatedAt. Anything else returns 422. A wide window returns a total with a 2-minute cursor and pages at 100; a window covering only the calling era returns whole with nextCursor: null. Report which you got |
search-contacts-advanced | Contacts by tag or assignee | Filter field names are camelCase; snake_case is rejected. pageLimit caps at 100 and the response carries no usable cursor — to cover up to 200, run the same filter asc and desc and take the union of the ids |
search-opportunity | Pipeline state | It embeds the contact (companyName, phone, tags), so joining a call to its opportunity needs no second read. contactId is the join key on both sides |
update-contact | Correcting a field | Send customFields only. Including tags in the body overwrites every tag on the record. Canary one contact and read it back before any bulk write |
What a call record carries: contactId, direction (inbound/outbound), status
(completed/no-answer/busy/voicemail), meta.call.duration, userId (which caller),
from/to (which line), a recording URL, and messageType — TYPE_CALL is a human dial,
TYPE_IVR_CALL is the voice agent. Counting the second as calling overstates the day.
2. What to check
Read the rules fresh each run rather than trusting a copy — they change.
| # | Check | Rule lives in |
|---|---|---|
| 1 | Every dial carries a disposition. Count the blanks | Execution SOP Step 6 |
| 2 | Each disposition's opportunity moved as the matrix says, or correctly did not | Execution SOP Step 6 |
| 3 | Call Attempt Count moved on every contact dialled. It is set by the C-021.Prospecting - increment call count workflow, not by the caller — so a contact dialled whose count did not move means the workflow has stopped firing, which is an infrastructure exception, not a caller one. Report it as such | Execution SOP Step 3 |
| 4 | No prospect over 5 attempts lifetime, over 2 voicemails lifetime, or dialled twice in a day on the same line. Several numbers for one business, dialled in one run, is one attempt — do not report it as an exception. Compute attempts from the call log as distinct days on which the contact received at least one outgoing dial. Do not read the cap off Call Attempt Count — that field counts dials, including each number in a multi-number run and each busy or declined line, so it over-reads against this cap by design | Execution SOP Step 7 |
| 5 | No dial to an excluded state | Compliance standard |
| 6 | No dial outside 08:30–20:00 recipient local, weekdays only | Compliance standard |
| 7 | Every dial from the caller's own assigned number | Phone Number Register |
| 8 | Inbound calls: answered, and by whom. An inbound during a calling block that the voice agent took is an exception | Execution SOP Step 6, rule 3 |
| 9 | A CALL note exists per dial in the documented format | Execution SOP Step 6 |
| 10 | No opportunity has a closed status in an open stage, or an open status in Closed Lost. Sweep the whole Sales Pipeline, not just yesterday's calls | Execution SOP, stage and status |
| 11 | Audit Finding Used is set on every contact dialled, as the rating and review count the gate was passed on. Report the ratio, not just exceptions | Cold Call Scripts §4 |
Check 11 is the evidence that the pre-dial gate ran. Every call now uses the same opener, so the field no longer records what the caller led with; it records the rating and review count the prospect was checked against before dialling. A dial with the field empty is a dial that may have skipped the gate. On 2026-10-29 it was empty on all 18 contacts dialled, so report it as a ratio every day rather than only when it is zero.
Check 11 used to cover Call Attempt Count too. It no longer does, because that field is now written by a workflow rather than by the caller. Asking a person to fill a field a machine fills produces noise, not compliance. Check 3 watches the machine instead.
Check 10 is a whole-pipeline sweep and it is cheap. search-opportunity on pipeline 9YS2qLIOFzJHr6Ouz47p returns status and stage together. It is here because a one-off backfill did not hold: two mismatches were corrected on 2026-10-07 and both directions had reappeared by 2026-10-30. Report the opportunity name and which field is wrong; do not guess which one is right — only the call history settles that.
Two judgement checks, not mechanical:
- Short connects. Any answered call under 20 seconds, listed with the contact, so a person can decide whether to listen. Day 1 produced several and one of them was the prospect saying "not really" to a banned opener.
- Dispositions that look wrong against duration. A
voicemail-leftlasting two minutes, or ano-answerwith a duration, is probably mis-logged. On 2026-10-28 a live conversation was loggedVoicemail. Flag the pattern; do not assert what happened — only the recording settles it.
3. The report
Write to the internal half of the agent's team-member folder (the agent resolved in §1), named
YYYY-MM-DD-call-log-review.md and dated for the day reviewed:
G:\Shared drives\sc-docs-restricted\10-people-operations\Team Members\<agent-slug>\reports\
Internal, never the shared half. This is an assessment. What reaches the agent is the coaching points drawn from it, spoken in the session — not this file. See the Team Member Shared Folder SOP §2 for the split and the Inside Sales Coaching Loop SOP §5 for which artifact goes where.
Exceptions first. If there are none, say so in one line and stop — a report that pads a clean day teaches people to skim it.
# Call-Log Review — <date>
**<n> dials, <n> connects, <n> exceptions.**
## Exceptions
<one line each: what, which contact, which rule>
## Worth listening to
<short connects and suspect dispositions, with a reason>
## Counts
<dials, connects, connect rate, dispositions used, opportunities moved>
Three rules for the writing:
- Do not speculate about why. Report what the log shows. "Logged
voicemail-left, duration 2:14" is useful; "she mis-logged this" is not, because the log cannot tell you that. - Say what you could not check. An unretrievable page, a field the connector would not return, a count that does not reconcile — name it. Silence reads as a pass.
- No praise padding. If the day was clean, one line. The genuine positives belong in a coaching conversation, not a compliance sweep.
4. Schedule and failure modes
Weekday mornings before the caller's block opens, so the review is ready at the top of it.
Routine identity
This is an account-level routine, not a local job. It follows the Claude account, not the machine — a new device inherits it already running, and there is nothing to reinstall. Recorded here because a silent routine and a routine that does not exist look identical from a fresh clone.
| Name | Inside-sales daily call-log review |
| Trigger id | trig_01SRjgRWVL38eYUcqssKqN89 |
| Cron | 37 12 * * 2-6 (UTC) |
| Environment | env_011Urvqqk36QhuNCBNHTMiTJ |
| Source | This repo, cloned fresh each run |
| Connectors | GHL MCP Server · Google Drive |
| Where to check it | https://claude.ai/code/routines |
To check whether it ran, list its recent runs rather than looking for the report — a fire refused before a session exists (paused routine, a cap, a failed pre-creation check) leaves no run record at all, so an empty list does not prove it never fired. Read the routine's own state as well.
| Runs | Tuesday to Saturday, reviewing Monday to Friday |
| Clock | Fixed UTC. It drifts by an hour at each daylight-saving change — the slot is early enough to absorb that, but check it after a change |
| Model | Set on the routine, not here |
| Repo | This repo is attached, so canon is read from the clone |
If the routine goes quiet, it has not necessarily failed. A fire refused before a session exists leaves no run record. Check the routine itself before concluding it never ran.
After writing the report, the run sends a push notification with the one-line summary and the file link, so the Implementation Lead sees it without opening Drive.
If the GHL connector is unreachable, the run should say so and stop rather than report a clean day. A silent pass on an unreachable source is the failure this routine exists to prevent.
Related Documents
- Inside Sales Coaching Loop SOP — what happens to this report once it lands
- Prospect Phone and Call Execution SOP — the dispositions, the matrix and the stop rules
- SC Outbound Calling Compliance Standard — windows and excluded states
- SC Phone Number Register — which number belongs to whom
- Sales VA Trial Readiness Plan — §7, the obligation this discharges
Document History
| Version | Date | Changes |
|---|---|---|
| 1.6 | 2026-11-12 | Check 6's window is 08:30 to 20:00, following the compliance standard. The Implementation Lead set the first dial at 8:30 am on 2026-11-12 |
| 1.5 | 2026-11-12 | The review resolves its agent from the Contractor Registry. On 2026-11-11 the run had no rule for <agent-slug>, guessed from the phone line's stale label and the only folder with a reports subfolder, and filed the current caller's review in a departed caller's folder. §1 gains Who the review is for: read the registry's Placed VA row every run, create the reports folder if missing, and write to _unassigned with a first-line warning rather than guess. Check 11 is reworded for the one-opener scripts (call scripts v1.20): Audit Finding Used now evidences the pre-dial gate, not the opener. §4 records the push notification the run already sends. last_updated had not moved with 1.4 |
| 1.4 | 2026-11-02 | Made portable. §4 gains the routine's identity — name, trigger id, cron, environment, connectors and where to check it — and states that it is account-level, so it follows the account rather than the machine. Recorded because a fresh clone on another device had no way to know the routine existed, and a silent routine looks identical to no routine. §1 gains the pull recipe and its four traps, all of which had to be rediscovered: sortBy accepts only createdAt/updatedAt, contact-filter field names are camelCase with no usable cursor, the opportunity read embeds its contact, and update-contact overwrites every tag if tags is sent. §3 says which half of the team-member folder the report goes in — internal, never shared — which it had left unstated |
| 1.3 | 2026-11-02 | Call Attempt Count moved from caller discipline to infrastructure. A workflow now increments it on every outgoing dial, so check 3 becomes a test of the automation and the Call Attempt Count half of check 11 is dropped. Check 4 gains the arithmetic it was missing: the cap is counted in attempts, defined as distinct days with at least one outgoing dial, computed from the call log — explicitly not read off the field, which counts dials and so over-reads on the multi-number runs that 1.2 established are legitimate |
| 1.2 | 2026-10-30 | Check 4 stops reporting number-tier working as a stop-rule breach. Its first run flagged a caller who dialled three published numbers for one contractor in 33 seconds and reached a decision maker for 2 minutes 33 seconds — the longest call of the day. The rule was wrong, not the caller; both this check and the SOP's stop rule are corrected. Adds check 11, Audit Finding Used and Call Attempt Count as a daily ratio, because the first field is the only record of what the caller led with and it was empty on all 18 contacts dialled |
| 1.1 | 2026-10-30 | Adds check 10, a whole-pipeline sweep for opportunities whose status and stage disagree in either direction. It is not a call-log check and it is here anyway, because the mismatch recurs: two were corrected by backfill on 2026-10-07 and both directions had reappeared within three weeks. Instructs the routine to name the mismatch and not guess which field is right |
| 1.0 | 2026-10-30 | Created so the readiness plan's daily call review stops depending on someone remembering. Written as the routine's own instructions rather than inlined into a scheduled job, so the review changes by editing canon. Carries the two data traps found the week it was written: the message export truncates near 200 and its paging is unreliable, and the campaign-stats endpoint disagreed with message history and was wrong |