0. Summary
- What you doA professional-services firm serving local clients.
- TeamAbout 30 people. The owner is still the bottleneck for email, follow-up, reports, and decisions.
- Time sinkDecisions waiting on the owner (hours per week not given).
- Switch on firstA weekly approval batch.
- Retainer fitGrow — see §7. Illustration only. This is not a paying client.
1. Current tools
| Tool | Used for | Connect | Note |
|---|---|---|---|
| Email (shared inbox and owner inbox) | Client mail, referrals, vendor requests | Yes | Agents may label and draft. Sending needs a human yes. |
| Calendar | Appointments and internal meetings | Yes | Booking or canceling stays human. |
| Practice system / CRM | Client records, follow-ups, status | Partial | Read and draft updates. Writes that change a client record need a yes. |
| Accounting | Invoices and collections follow-up | Partial | Draft reminders only. No spend or refunds without a yes. |
| Phone / SMS | Client contact | Partial | Scripts and drafts only. Live send needs a yes. |
| Shared files | Templates, reports, policies | Yes | Agents may read and draft. Deleting or publishing stays human. |
ToolEmail (shared inbox and owner inbox)
Used forClient mail, referrals, vendor requests
ConnectYes
NoteAgents may label and draft. Sending needs a human yes.
ToolCalendar
Used forAppointments and internal meetings
ConnectYes
NoteBooking or canceling stays human.
ToolPractice system / CRM
Used forClient records, follow-ups, status
ConnectPartial
NoteRead and draft updates. Writes that change a client record need a yes.
ToolAccounting
Used forInvoices and collections follow-up
ConnectPartial
NoteDraft reminders only. No spend or refunds without a yes.
ToolPhone / SMS
Used forClient contact
ConnectPartial
NoteScripts and drafts only. Live send needs a yes.
ToolShared files
Used forTemplates, reports, policies
ConnectYes
NoteAgents may read and draft. Deleting or publishing stays human.
AI today: what worked: A chatbot for short internal drafts. What didn't: Anything that had to reach a client without review. What we will not repeat: Unsupervised send, publish, or spend.
2. Time sinks
| # | Recurring task | Frequency | Hours/wk | Rating |
|---|---|---|---|---|
| 1 | Clearing parked approvals | Ongoing | not given | Agent drafts + human yes |
| 2 | Reading a morning brief and making the day’s decisions | Weekdays | not given | Agent drafts + human yes |
| 3 | Inbox triage | Daily | not given | Agent drafts + human yes |
| 4 | Signing into tools and repairing broken connections | As needed | not given | Keep human |
| 5 | Nudging status across the team | Weekdays | not given | Agent alone |
| 6 | Client-facing follow-up and content | Weekly | not given | Agent drafts + human yes |
#1
Recurring taskClearing parked approvals
FrequencyOngoing
Hours/wknot given
RatingAgent drafts + human yes
#2
Recurring taskReading a morning brief and making the day’s decisions
FrequencyWeekdays
Hours/wknot given
RatingAgent drafts + human yes
#3
Recurring taskInbox triage
FrequencyDaily
Hours/wknot given
RatingAgent drafts + human yes
#4
Recurring taskSigning into tools and repairing broken connections
FrequencyAs needed
Hours/wknot given
RatingKeep human
#5
Recurring taskNudging status across the team
FrequencyWeekdays
Hours/wknot given
RatingAgent alone
#6
Recurring taskClient-facing follow-up and content
FrequencyWeekly
Hours/wknot given
RatingAgent drafts + human yes
3. Top automations
Open each item. Effort marks are inferences, not measurements.
01 Weekly approval batch+–
- Replaces
- Task #1
- How it works
- A routing seat collects parked items into one ranked list. The owner clears the list in a single sitting. Nothing in the batch executes without an exact yes.
- Tools
- Email, practice system / CRM, shared files
- Effort
- S (inference: our estimate, not measured)
- Running cost
- None: uses existing tools
- Payback
- Payback cannot be computed from intake; hours were not given. Track hours on this task in week 1.
- Guardrail
- Nothing executes without an exact yes.
- Risk
- If the batch is skipped, items age. One reminder, then stop.
02 Inbox triage in shadow+–
- Replaces
- Task #3
- How it works
- An agent labels each message as needs-owner / maybe / noise, in shadow, while the live inbox stays unchanged. After a proving window, only labels go live. Replies stay draft-only.
- Tools
- Effort
- S (inference)
- Running cost
- None: uses existing tools
- Payback
- Payback cannot be computed from intake; hours were not given.
- Guardrail
- No send without a yes. Shadow-only until the owner agrees the labels are trustworthy.
- Risk
- A missed important message. That is why it stays shadow-first.
03 Start with a small seat set+–
- Replaces
- Overhead of too many overlapping jobs
- How it works
- Stand up the four seats in §4. Do not add more until the first seats run cleanly for a week. Each new seat, merge, or retirement needs an exact yes.
- Tools
- Shared files, practice system / CRM
- Effort
- S (inference)
- Running cost
- None: uses existing tools
- Payback
- Seat cost is not itemized in this sample, so no dollar figure.
- Guardrail
- Each change to the seat set needs an exact yes.
- Risk
- Adding seats before the first ones are clean recreates the bottleneck.
04 Request-to-fix loop+–
- Replaces
- Ad-hoc repair requests that stall on the owner
- How it works
- A written request becomes a tracked card. The owner approves the card. One agent drafts the change. The owner merges or rejects. No silent retries.
- Tools
- Shared files, practice system / CRM
- Effort
- M (inference)
- Running cost
- None: uses existing tools
- Payback
- Payback cannot be computed from intake; hours were not given.
- Guardrail
- No send, publish, or live change without a yes.
- Risk
- Blocked if sign-ins in task #4 are still broken.
05 Written intake and follow-up pipeline+–
- Replaces
- Owner-written replies to every new inquiry
- How it works
- A completed form becomes a draft reply and a checklist. The owner approves the reply. Follow-ups stay in draft until a yes.
- Tools
- Email, practice system / CRM
- Effort
- M (inference)
- Running cost
- None: uses existing tools
- Payback
- Payback cannot be computed from intake; hours were not given.
- Guardrail
- No client email without a yes.
- Risk
- A draft that sounds finished can still be wrong. The owner reads it.
4. Seat map
Default: start with these seats. Add more only after the first week runs cleanly.
| Seat | Job | Reports to | Approver |
|---|---|---|---|
| Routing | Collects requests, parks blockers, prepares the weekly approval list and the morning brief. | Owner | Owner |
| Inbox | Labels mail and drafts replies. Does not send. | Routing | Owner |
| Follow-up | Drafts client follow-ups and reminder sequences from the CRM. | Routing | Owner |
| Reporting | Drafts the weekly written status. Does not publish. | Routing | Owner |
SeatRouting
JobCollects requests, parks blockers, prepares the weekly approval list and the morning brief.
Reports toOwner
ApproverOwner
SeatInbox
JobLabels mail and drafts replies. Does not send.
Reports toRouting
ApproverOwner
SeatFollow-up
JobDrafts client follow-ups and reminder sequences from the CRM.
Reports toRouting
ApproverOwner
SeatReporting
JobDrafts the weekly written status. Does not publish.
Reports toRouting
ApproverOwner
5. 30-day plan
| Week | Switch on | Check |
|---|---|---|
| 1 | Automation #1 in draft-only mode. Log hours spent on parked approvals. | Is the approval list usable without major edits? |
| 2 | #1 live with an approval. #2 in shadow / draft mode. | Zero sends without approval. Owner responds in one batched session. |
| 3 | #3 held to the four-seat set. #5 in draft mode. Review guardrails. | Any near-miss → tighten the rule. Do not add seats yet. |
| 4 | Keep or stop each automation. Compare week-4 hours to the week-1 log. | Hours saved are measured, not estimated. |
Week1
Switch onAutomation #1 in draft-only mode. Log hours spent on parked approvals.
CheckIs the approval list usable without major edits?
Week2
Switch on#1 live with an approval. #2 in shadow / draft mode.
CheckZero sends without approval. Owner responds in one batched session.
Week3
Switch on#3 held to the four-seat set. #5 in draft mode. Review guardrails.
CheckAny near-miss → tighten the rule. Do not add seats yet.
Week4
Switch onKeep or stop each automation. Compare week-4 hours to the week-1 log.
CheckHours saved are measured, not estimated.
Success is measured from the week-1 hour log, not from a promise made in this sample.
6. Guardrail rules
- Agents may do alone: Research, drafts, labels, and files the owner already allows.
- Needs your yes: Send, publish, spend, delete, sign, book, cancel, or change a client record.
- Escalation: Blocked items go on one list. The owner clears that list in one batched session.
- Never: Paste secrets. Retry a declined send another way. Invent hours or results.
7. Retainer fit
Grow ($249/mo) matches the pattern: several automations, seats still being shaped. Illustration only. This sample is not a paying client. The report is yours either way.
Get the audit
Your report will follow this structure, filled from your intake — not from this sample.