What "Monday review" used to mean
Every Monday at 8am, the on-call strategist opened a portfolio spreadsheet, then opened each client's Google Ads account in a separate tab, then opened the call-tracking dashboard in a third tab, then a notes doc in a fourth, and started the round.
For each account: pull Saturday and Sunday spend, compare to the daily plan, flag anything pacing over 110% or under 60%, eyeball search terms for new junk, check whether any campaign tripped into Learning over the weekend, glance at call duration, write three bullets in the notes doc.
Across 39 accounts, that ritual averaged 2.5 hours. Mondays were over by 11am. We knew the portfolio was healthy, but we had spent half the morning learning that. Anything important that was not in the standard pull was hidden in plain sight.
The spreadsheet bonfire
The first thing we did was delete the spreadsheet. Not move it, not refactor it: delete. This sounds dramatic and it was supposed to be. The spreadsheet was the symptom, not the disease. It pulled data on a schedule, it had 14 tabs, and nobody on the team trusted any single number on it — every cell came with a story about whether the formula had drifted.
We replaced it with a single output we now call the Daily Briefcase: one page, per agency, generated overnight, that surfaces the four things a strategist actually needs to act on Monday morning.
- Top movers — accounts where spend, CPA, or conversions changed by more than a threshold since the previous business day
- Hidden risks — accounts that are not at the top of the movers list but where a leading indicator is degrading (low impression share spike, search-term drift, learning re-entered)
- Opportunities — accounts where one specific lever (budget, tCPA, a single keyword) would unlock more conversions at the current efficiency
- Standing actions — the 4 to 6 things that actually matter this week, by account name, that the strategist needs to ship before Friday
That is it. Four sections. One page. We can read the whole thing standing up with a coffee.
What we automated
The briefcase is generated by a small overnight pipeline. Apps Script pulls the data from each MCC at 4am local. A Python script normalizes the columns. Claude writes the prose in the four sections from the normalized data, using a frozen prompt and a deterministic source-validation step so it cannot invent numbers. By 7am the file is ready in our shared workspace.
| Step | Who does it | Time |
|---|---|---|
| Pull data from MCCs | Apps Script (automated) | ~ 14 min |
| Normalize and reconcile | Python (automated) | ~ 2 min |
| Write the four sections | Claude with frozen prompt | ~ 3 min |
| Validate every number against source | Validator script (automated) | ~ 1 min |
| Strategist reads, sanity-checks, sets priority | Human | ~ 12 min |
The strategist still does the last step. We tried automating that too. We stopped because it crossed a line we did not want to cross.
What we kept manual on purpose
This is the part most "AI for agencies" pitches skip. There are three things we will never automate in this workflow, even though we technically could.
1. The final call on priority
The briefcase tells the strategist what changed and what is at risk. It does not tell the strategist what to do first. That decision has context the data does not have: which client is in a quarterly review next week, who is mid-renewal, who is currently asking for more touch points. Putting an LLM in that seat would optimize for the wrong thing.
2. The first client call of the week
Mondays still include a 15-minute call with the two largest accounts. The briefcase is the input; the conversation is the output. Removing the call would save another 20 minutes a week and cost us a lot more than that in trust.
3. Anything that touches live bidding
The briefcase recommends levers. It does not pull them. Every change to budget or tCPA still gets a human eyes-on review, with a 14-day cool-off enforced after a Smart Bidding change. The cost of one bad automated nudge across 39 accounts is higher than the cost of 20 minutes of strategist time.
The rule we ended up with
Automate the look, never the verdict. The pipeline tells the strategist what to look at; the strategist decides what to do. That split is what made the new Monday actually faster, not just shorter on paper.
The numbers, twelve months in
We kept timestamps for a year before and a year after we shipped the briefcase. The averages are not cherry-picked from a good week.
| Metric | Before | After |
|---|---|---|
| Monday review (avg, per strategist) | 2h 30m | 12 min |
| "Hidden risks" caught before client noticed | ~ 1 per month | ~ 4 per month |
| Accounts per strategist (sustainable) | 10 to 12 | 18 to 22 |
| Strategist Monday end time | ~ 11am | ~ 9am |
| Spreadsheets in active use | 14 | 0 |
The Monday savings are the headline, but the second-order effect matters more: a strategist with 22 accounts has more brain left for any one client than a strategist with 12 accounts and a Monday hangover. We expected to free up time. We did not expect the quality of the work to go up at the same time.
What it cost to build
Six weekends. Roughly $20/month in API spend. No engineering hires. The hardest part was not the code — it was deciding what the briefcase should and should not say. That part took longer than the automation itself.
The framework is the same one we now ship inside actcenter. You can read the public version of the build guide in our portfolio dashboard post, or skip the build and let us generate the briefcase for you every morning.