$1,500 in media, one offer, one Google Search campaign, 28 days: that is the constraint, and I am holding it for the full four weeks. A management fee that covers Google, Bing and ChatGPT does not turn that money into three usable test budgets.
The temptation is understandable. OpenAI reported that WeightWatchers’ attributed CPA on ChatGPT Ads came in 15.3% below its blended paid-search benchmark, measured by Rockerbox. That is an encouraging result from an early partner test. OpenAI also says its incrementality work with Haus, Measured and WorkMagic is still in the early stages.
The distinction matters: attributed CPA assigns credit to an ad interaction. It does not, by itself, establish what the advertising added that the business would not have won anyway.
For a business with identifiable, high-intent Google demand, I want one usable acquisition baseline before paying for another channel’s learning period. Coverage is not a budget. Split $1,500 three ways and you buy three small samples instead of one useful test.
0. Before day one: ring-fence the media and check the fit
Cost against the constraint: $0 in media before launch. Management stays on a separate line.
What it buys: a test whose costs—and limits—you understand before the first click.
The $1,500 goes to Google. It does not include management. I keep those numbers on separate lines in the same spreadsheet so nobody confuses buying clicks with paying for work.
Freelance management typically runs $500 to $2,500 a month, or 15 to 20% of spend under $10k with a $500 to $1,500 minimum. A $500 fee on $500 in spend is 100% overhead before a click is bought. A $1,000 fee on $1,500 in spend substantially raises what the test has to earn back.
For these 28 days, I favor self-management unless the owner will not actually do the checks. Allow about three hours in week one and one hour a week after that. A tightly scoped freelancer quote for setup plus two check-ins can also work, provided the scope and price are capped.
At $1,499 a month, groas buys continuous execution plus a human strategist. That becomes the right purchase when the waste it addresses justifies the fee. It is not the right purchase when the fee effectively doubles this test’s cost.
Before committing even the media budget, I check three things:
- Identifiable demand: people already search for the commercial version of the problem, with a wallet in hand.
- Affordable clicks: the core terms leave enough room to buy useful evidence. At $18 a click, this budget can get thin fast.
- A proven offer: someone has paid full price, and the landing page has converted cold traffic.
If those conditions are missing, I skip this plan. It buys evidence. It does not manufacture demand, fix pricing, or rescue an offer that has never sold.
The allocation is straightforward: about $350 in each of the first three weeks, then the remaining $450 in the final week. Those are spending targets, not promises of exact daily delivery. I track actual spend against the $1,500 ceiling throughout; the daily budget setting is not my test ledger.
What gets cut: an ongoing retainer that consumes the test, and any launch that cannot meet the basic demand, affordability, and offer checks.
Keep management separate. If it doubles the test’s cost, run the test yourself and hire once there is a baseline worth protecting.
1. Days 1–7: buy one slice of intent and test the plumbing
Cost against the constraint: about $350 of the $1,500.
What it buys: an initial read on whether the closest buyers click—and whether their actions get recorded correctly.
I pick one commercial intent and one landing page that matches it. Emergency plumber in Austin goes to an emergency page with a phone number above the fold, not a homepage advertising six services. SaaS time tracking for agencies goes to a trial page for agencies, not a generic features page.
One campaign keeps the budget and conversion evidence together. The arithmetic gets ugly when a small account fragments: at $60 per enquiry, $1,500 buys about 25 enquiries in one campaign versus about five each if split five ways. Structure cannot rescue a lack of evidence. It can spread thin data thinner.
The lean budget playbook for bootstrapped companies follows the same one-campaign logic.
Before spending, I test the conversion paths that apply:
- Submit the form.
- Call the tracking number.
- Buy the product with a test code.
- Check that each intended conversion records once, not twice.
I have watched Smart Bidding chase a thank-you page that fired twice and a call asset that never rang through. The account looked brilliant while the business starved. Tracking is not an administrative task to finish later. It is what tells the system which outcomes to pursue.
I start with a $50 average daily budget, keep the location tight to where the business can actually fulfill, and check search terms every 48 hours. I block clearly irrelevant intent: jobs, free, DIY, salary, where those searches do not fit the offer.
Clicks can start within hours of approval, but a usable signal takes about two weeks, with stable judgment closer to 90 days. This is a 28-day baseline test, not a promise of mature performance.
What gets cut: Performance Max, Display, another Search campaign, Bing, and a ChatGPT test. Splitting the first week’s $350 into three roughly $115 samples defeats the exercise.
If tracking lies in week one, every bid change in week three optimizes the lie.
2. Days 8–14: diagnose waste before touching the bids
Cost against the constraint: about $700 spent cumulatively by day 14.
What it buys: a way to separate three failures that get lumped together as “the ads aren’t working.”
I diagnose them in this order:
- Tracking failure: the lead happened, but Google did not record it—or recorded it incorrectly.
- Poor query fit: Google recorded a click from someone who was never a likely buyer.
- Conversion failure: the right person arrived and did not act.
The order matters. There is no point rewriting a page because a working form stopped reporting submissions. There is equally little point adjusting bids to attract more of the wrong people.
If the search terms show job seekers, I add negatives and tighten matching. If clicks match the intended demand and forms sit silent, I investigate the page or offer before blaming the bid.
I do not rebuild the account every morning. Thin conversion data already makes calibration difficult; conversion volume and learning time matter. Constant bid-strategy changes and budget swings add disruption when I need a cleaner comparison.
Nor do I declare learning complete because the calendar says day 14. Learning takes one to two conversion cycles, not a fixed number of days. For emergency service, a cycle may be a day. For a $20,000 remodel, it may be weeks.
This week, I judge inputs I can inspect:
- Are irrelevant searches becoming less common?
- Does each conversion path fire once?
- Is the cost per qualified start within a range the business can afford for another two weeks?
That last question keeps the test commercial. “Still learning” is not permission to ignore spending that the business cannot carry.
What gets cut: daily bid changes, budget swings, and new keywords added simply because the account feels quiet. I fix observed waste, not boredom.
Fix leaks in week two. Save the performance decision for week four.
3. Days 15–21: connect leads to money, not Google’s column
Cost against the constraint: about $1,050 spent cumulatively by day 21.
What it buys: the first honest read on business value.
I export the leads, not just the dashboard. Each lead gets a business status: qualified or junk, quoted or ignored, booked or dead. For ecommerce, I look at contribution after product and shipping costs rather than treating revenue as money kept.
Say $1,050 buys 18 form fills:
- Six are spam.
- Seven never pick up.
- Three receive quotes.
- Two book at $900 each.
The cost per booked job is $525. That is the acquisition number to compare with what a job can carry. The $900 booking value is not all available to pay for advertising.
Cost per click is plumbing. Cost per raw lead is a rumor. A form fill only becomes useful evidence when I know what happened next.
I also account for lag before calling the result final. Conversions can be credited back to the click date, which means an earlier week can look better as later conversions arrive. Looking only at the most recent seven days can make an account look worse precisely because those clicks have had less time to convert.
Instead, I review trailing 14- to 30-day windows matched to the sales cycle, using the history available within this test. If customers close in 48 hours, older clicks have had time to produce a more useful read. If they close in 30 days, much of the evidence is still forming. I am judging qualified starts, not pretending I already know final revenue.
That distinction determines what the next week can reasonably decide. A short sales cycle may support a booked-job baseline. A longer one may only support a decision to continue observing the same cohort.
What gets cut: decisions based on Google’s conversion column alone, and verdicts that treat recent, unfinished leads as failures.
Count qualified outcomes in the sheet, then give the conversion lag time to catch up.
4. Days 22–28: use the remaining $450 to make a defensible call
Cost against the constraint: the remaining $450, adjusted for actual spend so total media does not exceed $1,500.
What it buys: a decision about this campaign—not permission to open another one.
I pace the remaining budget against the ledger. The first three weeks target $1,050; the final week uses what remains. I do not assume that leaving the daily setting untouched will land on exactly $1,500.
By now, I have roughly three weeks of clicks, a sheet of qualified outcomes, and older click cohorts that have had more time to convert. If the last seven days look thin, I read them alongside the older data before reacting. A calendar deadline does not settle conversion lag, which is also why a 30-day guarantee should make you suspicious.
I make one of three calls:
- Continue the same test after day 28. Queries fit, tracking works, qualified-start costs remain defensible, and closes are still in progress. The evidence is incomplete, not necessarily bad.
- Stop and fix the offer or conversion path. The right searches arrive but do not convert, or qualified acquisition costs exceed what the job or order can carry. More traffic is not the repair.
- Keep the Google baseline and prepare the next budget decision. There is a usable cost per booked job or other qualified outcome. I document it before changing the test’s scope.
None of those calls sends money to Bing or ChatGPT during these 28 days. A second channel needs its own hypothesis, measurement, and budget the business can afford to lose. It should be judged on new customers or incremental bookings, not just platform-attributed CPA.
Early reports such as WorkMagic finding 67% of incremental purchases for Dose came from new customers illustrate the kind of outcome worth examining. They are not a reason to split this budget now.
What gets cut: a last-minute channel launch, a celebratory verdict based on raw leads, and extra spending just to make the test feel conclusive.
Day 28 buys a decision. It does not guarantee certainty.
5. When the constraint lifts: fund what worked before buying coverage
Cost against the constraint: $0 more within this test. Any further media or management needs a separate budget.
What it buys: a handoff based on evidence rather than optimism.
When more budget becomes available, my first move is not to add channels. It is to add weight to what worked. If the Google baseline earns that decision, I keep the campaign and raise its budget in steps fulfillment can absorb. If the evidence is still incomplete, I continue the same test rather than disguising uncertainty with expansion.
Before hiring ongoing help, I hand over a sheet, not a login and a hope:
- Search terms, with negatives marked.
- Conversion actions, with test dates.
- Cost per qualified lead and cost per booked job.
- The observed conversion-lag window.
Anyone who cannot read that sheet and explain what they will change in week one is selling coverage, not execution.
This is where groas earns the handoff for owners who do not want a second job. Self-management makes sense at $1,500 because the fee would consume the test. Once the baseline and budget justify ongoing management, daily query review, budget pacing, and bid calibration become work a monthly check-in cannot cover. groas runs that execution continuously, with a named strategist and logged actions.
Prove one channel can pay, then pay for someone to run it around the clock.

