Kira is an AI teammate who runs your Amazon ad account. She is software. Busy hires her out by the month, the way you would take on an agency.
Your ad account doesn’t have a budget problem.
It has a waste problem, and no record of what changed.
At 7:05 every morning she tells you where the month stands against the plan. In between she works the account every day, inside guardrails you set once and she cannot change.
What lands at 7:05 tomorrow
This month we’ve driven $312.4K (vs $340.0K plan) in sales on $33.8K ad spend (vs $37.0K plan) for a TACOS of 10.8% (vs 10.9% plan).
- Hydrating Serum$141.0K / 3,120 units
- Repair Balm$96.4K / 2,410 units
- Travel Set$21.8K / 610 units
MTD sales are up 6% YoY.
Nothing needs a decision today.
She works for whoever owns the account. If that is you rather than your founder, nothing changes hands: the plan, the targets and the guardrails stay yours, she cannot change any of them, and you pick who the morning note goes to.
The rules in these playbooks came off accounts an operator actually ran, which is where the awkward ones come from: record the bids before pausing a stocked-out product, keep branded terms in their own bucket, never open a campaign without its keywords in the same run.
The week
What she actually does,
Monday to Friday.
Ten playbooks, each with its own trigger and its own cadence, all of them installed and running on a live ad account. Nothing below is planned, coming soon, or on a roadmap.
- Every morningBudget pacing
Yesterday's spend sorted into its buckets and measured against the month's plan. Drift small enough to correct, she corrects; drift big enough to be a decision, she flags once with a proposed action.
- Every morning, 7:05The daily note
Sales, spend and TACOS against plan, month to date. A few lines, then it stops.
- The day it happensOut-of-stock pause
A product hits zero and its ads stop that day. The bids it was running at are recorded first, so when stock lands the ads come back at exactly those bids.
- Every search-term pullBranded-term hygiene
Searches for your own brand name negated out of category campaigns, so branded spend stops being reported as growth.
- On your scheduleBids to target
Bids recomputed from what each target actually returned. No single bid moves further than the step you set, so nothing swings the account in one run.
- WeeklyHarvest and negate
Converting terms graduate to exact match and get negated at the source, so the spend moves rather than doubling. Terms that only spend become negatives.
- Weekly sweepCut the losers
Campaigns past your loser threshold get bid down or paused. The old auto campaign that quietly became the account's biggest spender is the one this catches.
- WeeklyThe report
Decisions first, numbered with reply codes; then spend against plan by bucket; then one line per playbook that acted.
- Before the month startsThe monthly plan
Last year's month, a growth multiple and your TACOS target give the budget envelope. She drafts it, you approve or edit it, and those numbers stand for the month.
- Underneath all of itHouse rules
Plain English, no jargon. A problem is flagged once with a proposed action, never re-raised the next morning.
Daily — spend against the plan
Weekly — the search-term sweep
Harvested to exact, negated at the source
- hydrating serum for combination skin11 orders · 19%
- niacinamide serum travel size7 orders · 22%
- fragrance free face serum6 orders · 16%
Negated — spent past threshold, no orders
- serum recipe homemade$68 · 0 orders
- face serum for dogs$41 · 0 orders
- cheap serum wholesale lot$37 · 0 orders
A single search term quietly spending $60 a week with no orders costs $3,120 over a year. Most accounts carry several at once, and they stay invisible until someone pulls the report, which is why that sweep runs on a schedule rather than when someone has time.
The harness
Nobody cares about your margin the way you do.
So she works inside a harness you set.
In the first hour, every number that could hurt you is set with you and written into the sheet: the target ACOS, the bid floor and ceiling, the most any single bid may move in one run, the clicks a target needs before it is judged at all.
She works inside those numbers. Outside them she asks, and waits.
Never — not hers to change
- Change the target ACOSNever
- Change the monthly budget envelope or the bucket splitsNever
- Move held-back test budget into always-on spendNever
- Activate a monthly plan you have not approvedNever
Only with your yes, every time
- Create a new campaign or ad groupAsks
- Pause a branded campaign for high ACOS — branded economics are different and that call is yoursAsks
- Spend past the monthly envelope, or anything that approaches itAsks
- Resume ads at bids other than the ones recorded when they were pausedAsks
- Add a term to the branded listAsks
On her own, inside your numbers
- Move a bid inside your minimum and maximum, by no more than your step sizeAlways
- Pause the ads on a product that sold out today, and resume them at the recorded bids when stock landsAlways
- Negate a search term that has spent past your threshold with no ordersAlways
- Harvest a converting term into exact match, and negate it at the sourceAlways
- Nudge a daily budget to bring a bucket back toward its planned percentageAlways
What a bid run looks like inside the harness
| Target | Clicks | ACOS | Bid | New bid | Note |
|---|---|---|---|---|---|
| hydrating face serum | 412 | 41% | $1.42 | $1.22 | step limit |
| vitamin c serum dry skin | 168 | 18% | $0.95 | $1.04 | within step |
| B0C4X2 (product target) | 233 | 62% | $1.10 | $0.90 | step limit |
| serum for redness | 97 | — | $0.80 | $0.80 | below min clicks |
| korean face serum | 301 | 24% | $1.15 | $1.09 | within step |
Two of those moves hit the step limit and stopped there. A target badly out of line comes back over several runs, which is slower on purpose: one run cannot swing the account. Every row lands in the changelog the same day, with the numbers that justified it, so a wrong move is visible the next morning rather than in next month’s invoice.
A morning in the account
7:05 AM.
This is the whole report.
A short note that arrives wherever you already read things, written so it makes sense to someone who was not on any call. Most mornings it reads like the one at the top of this page and ends with nothing to answer. This is the other kind.
A morning with one decision
This month we’ve driven $188.2K (vs $200.0K plan) in sales on $24.6K ad spend (vs $22.0K plan) for a TACOS of 13.1% (vs 11.0% plan).
- Hydrating Serum$88.1K / 1,940 units
- Repair Balm$61.9K / 1,550 units
Repair Balm sold out yesterday — its ads are paused; restock lands ~Sept 12.
Spend is running ahead of plan while sales sit behind it. I suggest we hold budgets flat for the rest of the month rather than trimming mid-month. Reply Y to hold, N to trim.
The decision is the last line, never buried under the numbers, and there is only ever one. Reply Y and she holds; reply N and she trims. Either way it is settled before you have finished your coffee.
A problem is raised once, with a proposed action attached, and never raised again the next morning. A decision you have already made becomes planning input rather than a question she asks you twice.
What she plugs into
No new dashboard to check.
Four places, and one of them is a Google Sheet.
She works in the two places your ad account already lives, keeps one shared sheet current, and sends the reports to wherever you read things. Nothing to log into, nothing to remember to open on a Friday.
- Amazon Ads
Campaign and target performance, search terms, bids, budgets, negatives, pauses. This is where the work happens.
- Seller Central
Sales, inventory and Brand Analytics search query performance — so a stockout reaches the ads the same day and TACOS is measured against real sales.
- One Google Sheet
The plan, the pacing view and the changelog as three tabs in a sheet you already own. Want a chart from it? Any AI will draw one on demand.
- Slack or email
Where the daily note and the weekly report arrive. She reports into the place you already read things, at the time you pick.
Friday — the report and the call agenda
1 · Decisions
1 — Repair Balm restock slipped to Sept 12. Keep its ads paused (1A), or move its budget to the Travel Set until stock lands (1B)? My recommendation: 1A.
2 — The retargeting campaign has come back on this list three weeks running. Restructure it (2A) or kill it (2B)? My recommendation: 2B.
2 · Spend vs plan
| Bucket | Spent | Planned | ACOS |
|---|---|---|---|
| Branded | 14% | 15% | 7% |
| Category | 71% | 70% | 27% |
| Competitor | 15% | 15% | 44% |
3 · Playbooks ran
Harvested 6 terms to exact · 14 new negatives · 2 campaigns bid down · 1 paused (out of stock) — details in the changelog.
A single search term quietly spending $60 a week with no orders costs $3,120 over a year. Most accounts carry several at once, and they stay invisible until someone pulls the report, which is why that sweep runs on a schedule rather than when someone has time.
Everything she changes lands in the changelog tab the same day with the numbers that justified it. That tab is why a wrong move surfaces the next morning instead of at the end of the quarter.
Who she is not for
Three accounts
she would turn down.
Worth saying before you type an email, because two of these come up on the first call and it saves everyone the hour.
- Not a fitThe account is already winning
If ROAS is fine and you are growing faster than you can buy inventory, the ad account is not your constraint. Sellers in this position say so plainly, and they are right. Fix the inventory first.
- Not a fitYou want the bids moved but not the numbers set
The target ACOS, the bid floor and ceiling, the step size, the thresholds — those take an hour with you at the start. Without them she is another algorithm spending your money to its own taste, which is the thing you already decided against.
- Not a fitThere is not enough spend yet
Below a few thousand a month there are not enough clicks to judge a keyword on, and the guardrails would leave her mostly waiting. Come back when the account has signal in it.
Where this came from
The playbooks were written
by someone who has run the accounts.
The thresholds, the flag-once rule, the fixed shape of the daily note — none of it came out of Amazon’s help pages. You can see it in the specifics nobody invents: bids recorded at pause so they can be restored exactly, branded terms kept in their own bucket so nobody reports repeat customers as growth, a campaign never created without its keywords in the same run.
Busy was built inside a holdco that runs its own ecommerce brands, and those brands are the first account Kira works. When a line in the morning note reads wrong, it reads wrong to an owner whose own money is in the account.
This is provenance, not endorsement. No customer count, no case study and no result is claimed on this page, because Busy has none it can show you yet. What it has is the ten playbooks above, in full.
Month one, in order
- Guardrail numbers set with youHour one
- Branded term list confirmed, account cleaned upWeek 1
- Prior year backfilled into the pacing tabWeek 1
- First monthly plan drafted for approvalWeek 1
- Daily note starts arrivingWeek 1
- First harvest, negate and loser sweepWeek 2
- First weekly report and call agendaWeek 2
- Next month’s plan drafted from the dataWeek 4
Straight answers
The six questions
that come up every time.
How is this different from me just using Claude?
Claude answers when you open it. Kira runs whether or not you think of her: the pacing check before you are awake, the search-term sweep on its day, a stockout paused the afternoon it happens. She also has standing permission to change bids inside your guardrails, which a chat window does not.
AI bid managers burn money. Why is this one different?
Because the numbers that could burn money are not hers. You set the bid floor, the ceiling, the most any bid may move in one run, and the clicks a target needs before it is judged. A badly out-of-line target gets there over several runs, never one. And every change is in the changelog the same day, so a wrong move shows up the next morning rather than in next month's invoice.
Is this a replacement for my agency?
It answers the question a retainer usually cannot. Ask an agency what it did last Thursday and you get “we optimized bids”. The changelog answers with the keyword, the old bid, the new bid, and the return that justified the move.
I run the ads in-house. Does this replace me?
No, and it is not built to. It takes the work that never reaches the top of your list — the daily pacing check, the weekly search-term sweep, the report nobody has time to write — and runs it on a cadence. The plan, the targets and the guardrails stay yours, and she cannot change any of them.
Who owns the account and the data?
You do. She works inside your own Amazon account, on access you grant and can pull back at any time. Nothing is migrated anywhere. If you stop, the sheet stays with you: the plan, the pacing history and every change ever made.
What does it cost?
There is no price on this page yet, deliberately: accounts differ enough that a number here would be wrong for most people reading it. Most sellers pay an agency $2,000 to $5,000 a month. That is the anchor most owners bring, and it is a fair place to start.
Get a demo
Put Kira on the ad account.
The demo is her working a real account rather than a slide deck — yours, pointed at read-only, or one of ours if you would rather not hand anything over yet. Type an email and a person here replies, usually the same day, with two or three questions about the account and a time.