Run a channel test with a stop rule

A channel test is not 'run some ads and see.' That produces a platform dashboard and a story. A useful test buys evidence about a named audience, offer, and conversion path under a fixed cash and time limit.

The go-to-market chapter warns against channel diffusion and using paid acquisition before customer value is understood. Read the surrounding argument at go-to-market chapter.

Freeze one channel question

Write the question in one sentence: 'Can this search campaign produce qualified calls from US controller-led firms at a marginal cost we can recover within the observed contract contribution?' Keep audience, offer, landing path, and qualification rule stable enough to interpret. Creative iterations are allowed only if they answer the same question and are logged.

Google Ads documents experiments that split budget or traffic between a base and trial campaign over a specified period (Google Ads experiments). This is platform documentation and applies to eligible campaign configurations; it is not independent evidence that paid search suits your business.

Define the conversion below the form

A lead is not a customer. Use a sequence: visit, meaningful action, qualified conversation, accepted evaluation, and paid conversion. Set the primary decision event at the deepest stage likely to produce enough observations. If paid customers are too rare, use a qualified event but do not label its cost CAC.

Record cash spend, creative or agency cost, founder and sales time, number of qualified events, and eventual gross contribution. The existing economics research explains why different CAC denominators tell different stories at metrics and economics chapter.

Worked hypothetical budget

Hypothetical: A B2B founder tests one narrowly targeted search theme for 21 days.

• Media cap: $1,800
• Landing and creative cash cost: $300
• Founder setup and follow-up: 18 hours, tracked separately
• Primary event: a call with the target role, confirmed workflow pain, and decision timing inside 90 days
• Advance: at least six qualified calls, at least two accepted paid-pilot discussions, and no evidence that the query intent is mainly job seeking or education
• Stop early: spend reaches $900 with zero qualified calls, or 30 relevant clicks produce no meaningful action
• Do not scale: leads appear but none meet the buying trigger or budget path

If the test yields six qualified calls, the media cost per qualified call is $300. It is not customer acquisition cost, because no customer has been acquired. All figures are invented for the example, not channel benchmarks.

Separate attribution from incrementality

Google Analytics defines attribution as assigning credit across touchpoints using a rule or algorithm (Google Analytics attribution). Credit is not proof that the conversion would not have happened without the channel. Branded search, repeat visits, sales outreach, and word of mouth can all sit in one path.

Where volume allows, use a randomized or geographic holdout. Where it does not, compare against a stable baseline and make a weaker claim: 'the platform attributed these qualified events,' not 'the campaign caused them.' Analytics reporting can also involve sampling, modeled events, and differences between reporting surfaces (Google Analytics reporting limitations).

Channel decision card

• Audience, trigger, geography, and exclusions
• Offer and landing path
• Primary event and qualification rule
• Cash cap, labor log, start, and end
• Advance, stop, and inconclusive conditions
• Tracking test completed before spend
• Attribution window and model recorded
• Control or baseline, with known contamination
• Decision: stop, repeat with one change, or scale one increment

Limits

Small tests may be underpowered; Google itself notes that experiments can remain inconclusive when traffic, time, or split is insufficient. Auction conditions and creative fatigue change. A passed test at one spend level does not establish that ten times the budget will preserve cost or lead quality. Scale in increments and keep cohort quality attached to channel.