Write the reason to stop before the ad test starts
A broken phone route, a full schedule and weak results call for different decisions. Give the owner and agency a written rule for each before spending.
TL;DR
- Define the full pilot cost, the work you want, what the shop can handle and the evidence you will review.
- Separate protective limits from performance conclusions. Reaching a cash limit can justify pausing without proving that the market will not buy.
- Decide who can pause spend, repair a problem or approve an extension. An average daily budget is not the written all-in spending agreement.
- Use a stated observation period for job outcomes. Booked, accepted, completed and paid work should not be combined to make a test look conclusive.
A pilot reaches its spending limit while several estimates remain undecided. The agency wants more time; the owner wants to know whether to stop. If neither party agreed what evidence would settle the question, the next conversation starts with competing interpretations.
Write the decision rules before buying the traffic. They cannot remove uncertainty, but they can identify which uncertainty needs another observation and which boundary the business will not cross.
Define the purchase you are testing
Start with a specific service, customer and area. An emergency repair campaign and a replacement-estimate campaign can produce different job timelines. A mixed total can hide that difference.
Then list what the pilot actually buys: media, management, setup and any other applicable costs. State the period, currency, payment dates and whether a cost continues after the pilot. Keep a website installment or ongoing subscription visible rather than treating its monthly line as the whole commitment.
The agency-quote guide helps resolve scope before this agreement. The billing reconciliation helps compare it with what is later charged.
Calculate what the specified cost would require
Consider an invented pilot with $1,200 media, $600 management and $300 setup. Its specified cost is $2,100. These are teaching inputs, not Lonta prices or a recommended budget.
Suppose a comparable completed and paid job brings in $800 and has $500 of job-dependent cost. It leaves $300 before marketing, fixed overhead and taxes.
Covering the specified $2,100 pilot cost would require seven additional comparable jobs at $300 each. Dividing only the $1,200 media by $300 produces four jobs, but leaves out $900 of the cost you agreed to pay.
| Invented checkpoint | Job contribution | Less $2,100 pilot cost | What remains unresolved |
|---|---|---|---|
| Six completed and paid jobs | $1,800 | −$300 | Whether observed jobs were additional |
| Six completed and paid; two accepted but pending | $1,800 from the completed group | −$300 so far | Completion, collection and costs of the pending work |
| All eight later completed and paid at the same cost | $2,400 | +$300 | Whether the pilot caused additional work |
Do not assign the pending jobs their full contribution at the checkpoint. The last row is conditional: it applies only if both jobs later complete and pay at the assumed cost.
Even the positive row is not proof of incremental advertising return. Some work might have happened without the campaign. Keep the break-even requirement, observed outcomes and causal question separate. The control-group guide addresses that last question and its practical limits.
Agree on the evidence before choosing the verdict
Google Ads conversions reflect configured conversion actions; the primary column can include modeled results. They are not automatically a count of distinct customers or completed jobs. Write down the actions being counted and how they connect to local inquiry and job references.
For this pilot, a useful review might inspect service fit, repeat contacts, recorded opportunity stages and cost evidence. Keep unknown outcomes visible. A missing CRM record should trigger a check rather than become an assumed failure or a claimed sale.
Use the first-20-inquiries review for an initial record check. Twenty is a manageable review size in that tool, not a statistically justified threshold that makes a pilot conclusive.
Use different rules for different problems
| Observed condition | Decision to agree before launch | What the condition does not establish |
|---|---|---|
| Agreed spending or funding boundary reached | Authorized person pauses or prevents further commitment | That the market has no demand |
| Phone, form or routing cannot be verified | Repair and verify the customer path; decide whether to pause affected spend | That reported inquiries are necessarily fake |
| Work falls outside the agreed service or area | Inspect targeting and qualification evidence before continuing unchanged | The precise cause of every unsuitable contact |
| Shop cannot serve additional work | Confirm availability and delivery constraints before buying more | That the advertising has become more efficient |
| Outcomes still immature or unknown | Set the next evidence check within the agreed financial boundary | Permission to extend spend automatically |
| Cost and mature outcomes miss the stated requirement | Review scope, baseline assumptions and alternatives; record the stop or change decision | A causal estimate from a simple before-and-after count |
A pause can protect the shop while a technical problem is repaired. Stopping the purchase can also be sensible when evidence remains inconclusive but the owner does not want to fund another observation. Neither decision requires pretending uncertainty has disappeared.
Put one review agreement in writing
Before launch, the owner and provider should be able to complete these fields together:
- Scope: service, customer type, area and exclusions.
- Cost boundary: included charges, dates, total commitment and who can approve a change.
- Operating boundary: available appointments or crew time and who updates availability.
- Evidence: platform definitions, local record references, stage rules and unresolved fields.
- Review: observation-through date, decision date and person responsible for each check.
- Actions: who can pause, what must be repaired and verified, and what requires a new approval.
Check how the actual account settings enforce the intended boundary. A platform's budget setting and a contract's total spending limit are different controls. Do not assume a general daily setting implements every fee and date in the agreement.
Allow time without writing an open-ended extension
Choose the observation window for the work being sold and state its end date. A group of leads received yesterday has had less time to progress than one received weeks earlier. Our conversion-delay example shows why unequal ages can distort the comparison.
Preserve the original checkpoint when reviewing again later. If a pending job becomes completed, record when it happened instead of rewriting the earlier snapshot. If the provider proposes an extension, ask what new evidence it is expected to provide and what the additional commitment is. Record a new decision rather than treating silence as approval.
What this doesn't cover
These are proposed operating rules and invented arithmetic, not an experiment establishing advertising effectiveness. No universal test duration, sample size, acceptable acquisition cost or platform learning deadline is recommended. The example excludes fixed overhead, taxes and collection losses and assumes comparable job costs. It does not replace your contract, cash plan or account-specific controls.
Questions owners ask
Is reaching the budget enough reason to stop?
It is enough to respect the spending boundary. Whether the evidence supports a performance conclusion is a separate question. The owner can decline more spending while leaving the outcome inconclusive.
Should a tracking fault count as a failed campaign?
Record the fault and its scope. Repairing measurement does not recover evidence that was never captured. Avoid converting an unknown outcome into either a successful or failed job.
Can we extend the test because two jobs are pending?
Review their actual state, the time needed to resolve them and the cost of waiting or continuing. Pending work is not automatic permission to buy more traffic. Keep the agreed approval requirement.
Sources and method
- Google Ads: Understand your conversion tracking data, read October 16, 2026, supports the distinction between configured conversion metrics and local job outcomes.
- The pilot amounts, job example, decision table and review agreement are original Lonta teaching material. No client observations or measured pilot results are used.
Lonta AI provides managed advertising. Discuss a pilot's scope and review rules before committing to it, or use these fields with your existing provider. The agency guides cover the related ownership, fee and reporting questions.