Chapter 01
How we learned the operation
Four weeks in the back office before anything was built. We interviewed seven people (two of them clients), read the systems the work runs through, and mapped each process twice: the documented version and the real one.
Sample deployment record. Tidewater Creative is a fictional client and every number here is illustrative.
Systems read and mapped
- QConnected
QuickBooks Online
Invoicing and books
API
- MConnected
Meta Ads
Spend and results
API
- GConnected
Google Ads
Spend and results
API
- GConnected
Google Sheets
Report template
API
- GConnected
Google Drive and Gmail
Filing, approvals, sending
API, shared mailbox
- NRetired
Notion
Onboarding checklist
Retired
7
People interviewed
6
Systems read
3
Processes mapped
11
Opportunities found
Where the time went
Pulling and re-keying numbers
34%
Chasing approvals and payments
22%
Reviewing results and deciding what to say
18%
Formatting and filing
14%
Talking to clients
12%
Agents and plain code Stays with people
70% of the time we mapped was pulling, chasing, formatting, and filing, which agents and plain code can do. The other 30% was judgment and client conversation, which stays with people.
How the four weeks ran
Week 1
Interviews and access
Owner, account lead, media buyer, designer, bookkeeper, and two clients. Read access to the ad platforms, QuickBooks, Drive, and the report workbook.
Week 2
Shadowing a month-end
We sat through one full reporting cycle. The documented process had four steps. The real one had eleven steps, five handoffs, and six and a half days of elapsed time for about two and a half hours of work.
Week 3
Exceptions and risks
Every exception from the last six months counted and sorted. Five of six types followed a repeatable pattern. Every invoice waited on the owner, and overages in three cycles were never billed.
Week 4
The sort and the plan
Every step labeled delete, plain code, agent, or human. Eleven opportunities ranked by value and complexity, grouped into three workflows in build order, and a quote for the whole map as one zone.
Opportunity map · 10 of 11 built · about 71 hours a month
| # | Opportunity | Workflow | Hours / month | Sorted as | Built |
|---|---|---|---|---|---|
| P1 | Platform export into the report template | Monthly client report | 14 | Plain code | Yes |
| P2 | Mismatch check before the report goes out | Monthly client report | 15 | Plain code | Yes |
| P3 | Change summary drafted from account history | Monthly client report | 9 | Agent | Yes |
| P4 | One-tap approval in email | Monthly client report | 6 | Agent | Yes |
| P5 | Send and file on approval | Monthly client report | 4 | Plain code | Yes |
| P6 | Invoice created the day the work ships | Invoice and collect | 5 | Plain code | Yes |
| P7 | Overage detection against the retainer tier | Invoice and collect | 3 | Plain code | Yes |
| P8 | Reminder cadence with a person at day 45 | Invoice and collect | 5 | Agent | Yes |
| P9 | Access requests and follow-up | New client onboarding | 6 | Agent | Yes |
| P10 | Asset collection and folder setup | New client onboarding | 4 | Plain code | Yes |
| P11 | First-month plan inputs assembled | New client onboarding | 3 | Agent | Not yet |
What we chose not to build
First-month plan inputs assembled
Two new clients in four months did not justify the build. The owner writes the plan from a short template instead. Revisit if onboarding volume doubles.
Automatic commentary without review
The commentary is how the account lead talks to each client, so a person keeps the final say. The agent drafts and the account lead edits and signs.