By FroggyAds.
Ad targeting is the set of rules that makes an ad opportunity eligible, keeps unsuitable delivery out, and gives you evidence about which conditions helped reach a qualified buyer. With FroggyAds, you can work with documented controls such as geography, device, operating system, browser, carrier, category and source. Start with offer eligibility, then test optional filters against accepted outcomes.
Quick answer: Begin with the buyer or context your offer can legitimately serve. Turn non-negotiable requirements into eligibility rules, such as supported country or device; treat every other targeting idea as a testable hypothesis. In FroggyAds, choose from the controls documented for your campaign and format, record exclusions, verify tracking, and compare source-level delivery with the business event you accept. More filters do not automatically mean better targeting.
| Decision stage | What the buyer must establish |
|---|---|
| Offer eligibility | Which markets, devices and terms the offer can genuinely serve. |
| Test hypothesis | One supported difference that could change the next buying action. |
| Evidence and guardrail | Accepted outcome, review window, spend cap and rollback condition. |
For Google Ads-specific audience terminology, consult Google Ads: About audience segments. Its segment types, exclusions and campaign logic belong to Google Ads. Confirm the available controls and combination behavior inside your FroggyAds campaign before making them part of a test.
Ad targeting does not identify a perfect person. It defines eligible delivery using signals that can describe a market, device, browsing context, category, source or audience relationship, depending on the platform. A signal is a proxy with a limit. Treat it as a rule you can explain and audit, not proof that an individual wants your product.
Translate the offer into campaign eligibility, scoped exclusions and a comparison that can change your next buying action. Audience planning identifies a reachable group; the targeting plan also needs to explain the delivery rules, source evidence and customer outcome used to assess that group. Keep those decisions connected to the same actual offer.
Before you open a dashboard, write one sentence describing the customer problem and the event that makes the campaign commercially useful. Add the markets you can serve, the product or landing-page requirements, the prohibited or unsuitable traffic, the available measurement, and the person who can approve a change. This brief separates a hard eligibility rule from an unproven performance idea.
A hard rule excludes an ineligible opportunity: the offer cannot be fulfilled in that country, the product requires a supported operating system or the destination fails on that device. An optional hypothesis predicts a difference between otherwise eligible conditions, such as source groups or browsers. Enforce the requirement and test the hypothesis. Ineligibility is distinct from invalid or fraudulent traffic.
Our documented controls include country and city, device, operating system, browser, carrier, category and source selection. Use the available options to match your offer or investigate a delivery pattern. Those attributes do not establish a person's occupation, purchase intent or CRM relationship. Confirm any additional audience-data capability separately before relying on it in the campaign brief.
For each candidate rule, ask three questions: Does the offer require it? Could its effect be measured separately? Would you change a bid, creative, destination, exclusion or budget if the result moved? If the answer to the last two is no, the signal probably does not deserve its own campaign cell. Keep the setup small enough to explain and large enough to learn.
Targeting controls interact. A country, device and source rule may each be reasonable alone but leave too few eligible opportunities together. Platforms also differ in whether multiple choices within one category broaden eligibility or whether categories intersect. Read the current setting help, verify the resulting estimate or delivery, and document the exact logic instead of assuming that settings behave identically across DSPs.
Give every test a stable name that records the campaign objective and the changed dimension, for example `trial-country-control` and `trial-country-city-test`. Keep a separate identifier for creative and landing-page versions. Consistent labels make your tracker, analytics and source report joinable, and they prevent a copied campaign from silently blending two different targeting rules into one total.
Exploration and scale answer different questions. Exploration asks whether an eligible segment or source merits further study; it uses a capped allocation and a clear learning question. Scale asks whether additional spend can maintain the required quality and economics. Do not treat delivery volume, a low click cost or a short positive run as proof that a broader audience will perform the same way.
Build one test cell around one buying question.
Choose the smallest set of rules that satisfies the offer, then decide which single optional dimension would change your next action. For example, keep the country and creative fixed while comparing supported mobile operating systems only if the product experience actually differs. Keep source reporting and the accepted conversion definition consistent so a result can be attributed to the setting you meant to test.
Create My Free Account
A useful targeting scorecard gives every decision a control, a reason and evidence. A country rule may enforce service eligibility; device may address product compatibility; category may test the environment in which an offer appears; a source exclusion may follow a mature quality problem. Record both the positive inclusion and its negative boundary so future reviewers can see what was allowed and why.
| Decision layer | Operating question | Evidence to retain |
|---|---|---|
| Offer eligibility | What must be true for this ad opportunity to be valid? | Approved terms, service area and compatibility check. |
| Optional targeting signal | Which single difference could improve a decision? | Written hypothesis, control cell and test variable. |
| Source treatment | When is source-level inclusion or exclusion justified? | Source report, conversion maturity and scope of evidence. |
| Message and destination | Does the visitor see the offer promised for this cell? | Creative version, landing page and verified terms. |
| Measurement contract | Which business event counts as accepted? | System of record, attribution window and deduplication rule. |
| Scale or rollback | What marginal result authorizes another change? | Cost and quality boundary, capacity check and saved baseline. |
Signals differ in certainty. A country or device value describes a delivery attribute; a category describes an inventory classification; a source ID identifies a supply path; a remarketing segment, where a platform offers one, represents a defined prior interaction. None alone verifies identity, need, intent or conversion quality. Describe exactly what the signal means and avoid more personal interpretation than the evidence supports.
Pair inclusion with exclusion. If a market is outside the offer's fulfillment area, exclude it as a hard rule. If a source appears weak, first check tracking, destination, conversion maturity and source mix; then apply an exclusion only when the evidence supports the scope. Overbroad exclusions can remove legitimate buyers, while missing exclusions can send spend toward customers you cannot serve.
A broader control or a holdout can provide context for a narrow segment, but only when budget, creative, destination, date range and conversion rules are comparable. When you cannot randomize, label the comparison observational. A difference between two targeting cells may reflect supply, time, device mix or user selection rather than the filter alone. Preserve those limits in your decision note.
Define the commercial event and its acceptance rule. State whether the team values a completed order, approved lead, paid subscription, install or another verifiable outcome; identify its source system, owner, attribution window and rejection reasons. Clicks and sessions can diagnose the path, but only the agreed business event should authorize a lasting acquisition-cost decision.
Translate the offer into eligibility. List required country or city, supported device and operating system, relevant carrier or category constraints, and any excluded source or environment. Separate rules that block an ineligible transaction from optional dimensions you want to compare. Check that the selected combination exists in the campaign interface and record its exact scope.
Choose one primary targeting hypothesis and prepare the experience it requires. Explain the mechanism rather than naming a segment: a device split may reveal whether onboarding completion differs because the app supports one operating system better. Keep the offer constant, adapt only necessary instructions, and use the same accepted event to judge both cells.
Test the whole journey before paid delivery. From each intended device and market, open the destination, complete a safe test conversion, and confirm campaign and source identifiers persist. Verify consent behavior, duplicate-event handling, currency and downstream acceptance. If the target device cannot complete the path, repair compatibility before interpreting low conversion as a targeting failure.
Set a learning-loss limit, a delivery review point and an outcome-maturity date before launch. Allow for the event volume your own decision requires, but do not invent a universal sample minimum. Pause immediately for a broken destination, policy issue or tracking defect; evaluate ordinary performance only after the agreed conversion window has elapsed.
Compare cohorts only after they reach the same maturity. Read eligible delivery, sessions, qualified actions and accepted outcomes by dimensions you can actually export. If a country looks weak on one browser or source, investigate that interaction before excluding the entire country. Mark low-volume results uncertain instead of promoting the most favorable cell.
Scale or roll back using a written rule. Increase one optional dimension at a time and review the incremental traffic separately. If cost, validity, refunds, support load or delivery quality crosses a guardrail, restore the last stable settings and investigate the earliest break in the path. Keep the campaign objective and conversion definition fixed while you diagnose.
Make creative specific to the eligible situation. If geography is a fulfillment boundary, name the supported service area; if a device split exists because of compatibility, describe the supported experience honestly. The ad should qualify a suitable visitor without exposing sensitive inferences or making the audience feel watched.
Vary the message only when the audience difference justifies it. For a country test, localize language, currency and terms you can verify. For an operating-system comparison, keep the offer constant and adapt only instructions that differ. Assign a distinct creative identifier to each change so reports can separate message effects from targeting effects.
The landing page completes the targeting promise. A visitor from an eligible market should see the correct shipping, service or legal terms; a mobile visitor should get a usable page; a lead should understand qualification and follow-up. If the ad says one thing and the destination reveals another, the resulting bounce or rejection does not prove the audience was poor.
Choose the execution format from the visitor's next step.
Push, native, display, pop, interstitial and video present different attention patterns; format and targeting are separate decisions. Select a format that fits the creative, destination and offer, then check asset and approval requirements in the live campaign workflow. Do not assume a targeting dimension or inventory volume is identical across every format.
Create My Free Account
Measure the path from eligible delivery to accepted value: opportunity, impression or click, landing session, qualified action, accepted conversion and downstream value. Each stage answers a different question. Clicks without sessions may point to redirects or page performance; valid visits without accepted outcomes may indicate offer fit or qualification. Diagnose the stage before changing targeting.
Use a short report that names the primary business event and a few diagnostic measures. Break down only by supported dimensions that could change your next action, such as country, device, source or creative. Excessive splits create tiny cohorts and unstable rates. Show counts beside rates and mark cells that lack enough mature observations for a dependable comparison.
Reconcile the media dashboard with web analytics, an affiliate tracker or your CRM. Clicks, sessions and accepted leads can differ because systems record different stages and use different attribution rules. Check redirects, consent, deduplication, time zones and conversion lag. Note known variance instead of forcing totals to match or crediting targeting with a tracking artifact.
For every metric, define its numerator, denominator, time window, source, currency, attribution scope and maturity rule. Accepted CPA is media spend divided by accepted conversions in the chosen period; it is not cost per click or cost per raw form submission. Use one definition across test cells before comparing their economics.
| Measure | Definition to specify | Diagnostic interpretation |
|---|---|---|
| Eligible delivery | Available opportunities under the selected rules, as reported by the platform. | Low volume may reflect eligibility, bid competition, approval or supply. |
| Qualified landing sessions | Sessions that meet your stated validity and qualification checks. | Compare with clicks to find redirects, consent or page-path gaps. |
| Accepted conversions | Business outcomes that pass the agreed approval rule. | Compare cohorts at the same maturity and attribution window. |
| Accepted CPA | Media spend divided by accepted conversions for the same scope. | Do not substitute raw leads, clicks or unapproved affiliate events. |
| Source quality | Source-level evidence joined to downstream acceptance where possible. | A campaign average may conceal meaningful source differences. |
| Accepted value in the added cohort | Accepted value from spend added after the baseline period. | Keep added spend separate so the original cohort cannot hide dilution. |
Set your ceiling from contribution economics rather than an industry benchmark. Estimate the contribution from one accepted outcome after variable costs, refunds, sales or service expense, and affiliate approval adjustments. The maximum sustainable acquisition cost cannot exceed that contribution and the share of value your business is willing to reinvest. Record uncertainty instead of guessing a conversion rate.
A useful break-even check is: maximum acceptable cost per accepted outcome must stay at or below expected contribution from that outcome. For a subscription, use retained value only when your records support the period and costs. For an affiliate offer, use approved commission after reversals. The illustrative calculation later is arithmetic practice, not a FroggyAds price or performance forecast.
Average performance can hide a weaker marginal audience. When budget grows, review the added spend and its accepted outcomes separately from the earlier base. A country that looked efficient at low volume may have a different source mix after expansion. Keep an unchanged reference where practical and preserve a rollback point before widening delivery.
Quality checks begin with the destination and supply evidence. Confirm campaign approval, creative, tracking and landing pages match the offer. Review source-level reports, invalid-traffic signals and unusual click-to-session gaps. FroggyAds provides Adscore-supported and internal traffic-quality controls; you should still validate downstream sessions and accepted outcomes against your own records.
Use data only when you have the right to collect and activate it, and follow applicable law, platform policy and consent requirements. Avoid sensitive-attribute inference, discriminatory exclusions and claims that a device, location or category reveals identity. Keep only the identifiers needed for the test, document retention and access, and have a qualified reviewer assess regulated offers.
Accessibility and compatibility are targeting considerations too. Check keyboard access, readable copy, contrast, motion, controls and forms on the devices your campaign includes. If a landing page fails on one browser, a temporary exclusion may protect the experience, but repairing the page is the durable answer when that audience matters to your business.
Common targeting failures begin when an offer boundary, a platform setting and downstream evidence do not agree. Separate policy or compatibility faults that need immediate correction from performance hypotheses that need a capped comparison. Then find the earliest point where observed traffic stopped matching the campaign brief.
The first failure is stacking filters because each sounds relevant. The combination may shrink eligible delivery, restrict source choice and leave too few observations to answer your question. Start with offer requirements; add one optional constraint only when there is a business reason and measurement plan. Check delivery before deciding that a market has no demand.
The opposite failure is broad expansion without source visibility or audience-overlap review. A campaign average can hide weak inventory and duplicate exposure across cells. Keep campaign and source naming stable, inspect the breakdowns actually available, and set a spending cap. If your platform cannot report a dimension, acknowledge the limit instead of claiming that the average proves segment quality.
Turn the eligibility rules into a reversible comparison.
Launch one defined test, not a matrix of every country, browser, device, category and source. Hold the offer, destination, conversion event and review window constant. Compare one optional targeting dimension, or a few cells whose differences you can explain. Stop for a technical or policy fault; otherwise wait for the prewritten outcome-maturity point before judging results.
Create My Free Account
Check the combined eligibility rules before concluding that the market has no demand. Remove an optional restriction only when the offer permits it, then test whether delivery changes while the creative, destination and accepted-event rule remain stable.
Compare identified sources with accepted customer outcomes before widening delivery. A useful campaign average can conceal a source whose own result deserves investigation.
Record where the targeting cells overlap and which identifiers your reports can deduplicate. Keep unknown overlap explicit instead of adding cell totals as unique demand.
Confirm service areas, supported devices and offer restrictions before launch. Exclude the ineligible condition and preserve the requirement that explains the boundary.
Use permitted delivery attributes for a clear offer requirement. Do not treat a location, device or category as proof of a sensitive personal trait.
Verify the visitor's actual progression and business acceptance. A segment name describes the buying rule; it cannot establish that every recipient wants the offer.
Days 1–4: confirm the offer's real eligibility, accepted event and exclusions; verify the destination and choose a baseline. Create a campaign record with source, creative and targeting identifiers. Ask the product, sales or affiliate owner to confirm terms and disqualifiers before launch. Do not invent requirements from early click data.
Days 5–10: launch the baseline with the fewest necessary controls. Monitor pacing, destination health, identifier persistence and campaign status. Fix mechanical faults promptly and annotate each change. Avoid daily audience edits because a dashboard moved; keep a stable observation period so ordinary variation is not mistaken for a repeatable mismatch.
Days 11–20: when tracking and delivery are sound, test one optional dimension that could change a decision. Compare supported mobile operating systems only if the offer or journey differs. Keep the control cell, budget logic and maturity window visible. If evidence is too sparse, record that limit and gather more only within the approved loss cap.
Days 21–30: review mature accepted outcomes, downstream value, quality signals and operating capacity. Keep, broaden, exclude a defined source or pause. If results are mixed, preserve the last stable version and write the next discriminating test. Thirty days is a planning sequence, not a universal statistical minimum or performance guarantee.
When you expand, change one dimension at a time: budget, geography, device, category, source or creative. Review the added spend as a separate marginal cohort so the original base cannot mask weaker new traffic. Keep the accepted-event rule fixed. A setting deserves expansion only if the additional audience still meets the same cost and quality boundary.
Confirm that the business can serve the additional customers before increasing delivery. Check sales follow-up, inventory, support capacity, payment handling, fulfillment and any regulated-offer review. A lower media cost does not help if the extra leads cannot be contacted or the product is unavailable. Treat operational capacity as a real targeting boundary, not an afterthought.
Store the last stable campaign configuration, creative, exclusions and measurement rule. If marginal cost, lead acceptance, refund rate, tracking variance or source quality crosses your guardrail, return to that version and investigate the earliest failed stage. Record who can authorize a restart. A reversible campaign gives you a useful learning path without locking the business into a poor segment.
FroggyAds supports a buyer-led workflow for country and city, device, operating system, browser, carrier, category and source controls. You can manage campaign budgets and bids and inspect source-level reporting, then use source allowlist or blocklist actions where supported. These controls help you act on actual paid-delivery evidence without implying that a location or device proves personal intent.
We recommend FroggyAds to affiliate marketers and media buyers who want to buy traffic directly, test the eligible market or device, and review results by source. You bring the lawful offer, creative, destination and accepted conversion definition; our self-serve workflow helps you control the delivery decisions. That is a practical fit when you value hands-on testing and source-level follow-up.
Our published minimum deposit is $50. It is an account-funding minimum, not a minimum bid, campaign price or recommended test budget. A suitable test amount depends on the buying model, market, bid, available delivery, conversion frequency and approval lag. Set a loss limit your business can accept, and avoid dividing the budget into cells too small to interpret.
Ad targeting helps when a documented signal either blocks traffic your offer cannot serve or creates a comparison that can change your next buying decision. Use offer eligibility first, then test optional geography, device, category or source rules against accepted outcomes. A segment label or lower click cost alone does not establish that the traffic is commercially suitable.
Build the first plan from the actual offer: write the customer problem, hard eligibility, optional hypothesis, exclusions, matching creative and landing page, accepted event, spend cap and review date. Use only the controls available for your format. Keep the baseline and change one major dimension at a time so the result is explainable.
Detailed targeting can add work and reduce reach. Extra cells need separate setup, budget, creative, reporting and review; smaller eligible pools may deliver slowly. Audience data can also bring licensing, consent and integration obligations. Compare that cost with the decision value: if a split will not alter what you do next, use a simpler baseline and preserve the budget for a test that can.
Validate fit with business evidence independent of the targeting label: qualified sessions, valid lead fields, merchant approval, completed orders, refunds or retained value where your records support them. Compare sources and devices at the same conversion age. Click-through rate shows response to the message; call the traffic qualified only when downstream acceptance supports that conclusion.
Make the creative explain why the eligible visitor should act. Name a supported service area, explain device compatibility or clarify the relevant offer terms when those facts are verified. Keep the destination consistent. Do not suggest that you know an individual personally or expose a sensitive trait inferred from a location, device, category or prior interaction.
Before launch, check offer approval, market and format eligibility, creative requirements, exclusions, destination behavior, tracking identifiers and accepted-event definition. Test from the devices you plan to include. Verify current options in the live campaign interface rather than assuming that documentation or a past account view overrides changed settings.
Targeting quality requires a chain of evidence: eligible delivery, reachable impressions, landing sessions, qualified actions, accepted conversions, source quality and economic value. Choose the measures that match your objective and show counts beside rates. When sample volume is limited, label the result provisional and retain the original settings until evidence is sufficient for the decision you planned.
If a segment underperforms, confirm the signal definition and freshness, then check available reach, source mix, exclusions, creative, destination compatibility, tracking, conversion maturity and overlap. Test a competing explanation before removing an entire market or device. If the issue is isolated to one source or landing flow, use the narrowest supported action and preserve the rest of the campaign as a reference.
Use only permitted signals and documented platform settings. Avoid targeting that relies on sensitive personal attributes, unlawful profiling, deceptive personalization or a proxy chosen to evade policy. Keep exclusions tied to legitimate offer eligibility or a verified quality issue, not arbitrary personal traits. For regulated products or employment, housing, credit and similar offers, obtain the required specialist review first.
Increase spend only after the added traffic—not just the original base—continues to meet your written acceptance, cost, quality and service-capacity rules. Review mature cohorts, source composition, refunds or rejection, tracking variance and frequency where available. Expand one dimension at a time and keep a rollback configuration. If the evidence is incomplete, keep the test capped or pause rather than presenting uncertainty as a win.
The official sources already linked in this guide describe Google Ads audience segments, contextual and topic targeting, placement targeting, exclusions, and WCAG accessibility. They explain their own systems and standards; use them as reference material, not evidence that FroggyAds exposes identical options. The FroggyAds controls described here are grounded in its public platform information, and live campaign settings remain the final check.
Use this worksheet before building the campaign. It is designed to prevent the most common targeting mistake: turning a customer idea into several overlapping filters before confirming what the offer requires. Complete each line with the source of evidence, owner, exact setting and rollback decision. A blank field is a question to resolve, not permission to assume the answer.
Write the primary accepted event and exactly how your business counts it. Include the system of record, deduplication rule, conversion window, qualification or approval status, attribution scope, owner and next decision that the metric will control. Add a diagnostic event only when it helps identify where the journey breaks; label it separately from the commercial success event.
Keep the final brief reproducible. Save the approved targeting values, campaign ID, exclusions, format, creative version, destination, source identifiers, budget cap and date. State what your data can and cannot show. A second buyer should be able to rebuild the comparison from this record without needing to infer the original intent from a dashboard screenshot.
For every signal, record what it represents, why it is relevant, who approved it, where it came from, how current it is, and whether it is an inclusion or exclusion. Mark hard eligibility rules separately from performance hypotheses. If the source cannot be verified or its use is not authorized, remove it from the plan and use an approved alternative.
List the exact promise, evidence, creative variant and destination for each eligible cell. For a geography test, record localization and terms; for a device test, record compatibility and rendering; for source comparisons, keep the core offer consistent. Track each creative separately so a change in message is not mistaken for a change in audience.
Forecast conservatively using your own bid, delivery, conversion and acceptance history. Show the assumptions for impressions or clicks, qualified visits, accepted conversions, contribution per outcome and expected cost. Include a downside case such as delayed approvals or weaker source quality. Do not fill missing history with industry averages presented as facts; use a transparent hypothetical and cap the loss.
Preserve campaign, audience-rule version, source, country or city, device, operating system, browser, creative and time window wherever reporting makes those details available. Join them to analytics or business records with stable identifiers. If the interface omits a dimension, note that limit and avoid making segment-level claims your data cannot support.
Reconcile the platform's clicks and spend with landing sessions and the accepted business event. Record each system's count, date range, time zone, attribution rule, duplicate handling, known lag, variance and owner. Investigate material gaps before acting on source quality. Keep a difference log so a later report does not silently compare incompatible definitions.
For every material change, note the symptom, evidence, hypothesis, exact setting, expected effect, spend limit, launch time, review date and rollback trigger. Change one main targeting variable at a time when practical. If a policy or destination fault requires immediate action, fix it promptly and label the before/after periods so they are not mistaken for a controlled test.
Before expansion, confirm the marginal cohort meets the cost and acceptance boundary, source quality is stable, landing pages and fulfillment can handle demand, tracking is current, frequency or overlap is understood where reported, and the new setting can be reversed. Keep the last verified configuration and decide who can approve the next budget increase.
FroggyAds is a practical fit when you already have an approved offer and want direct control over paid traffic. Create your account, set up a focused campaign with the targeting, budget and source decisions available for your format, and review source-level evidence before expanding. You supply the lawful offer, creative, destination and accepted conversion definition; our self-serve workflow gives you hands-on controls for a testable delivery plan.
Create My Free AccountChoose the next action from the offer and observed delivery: enforce a genuine eligibility boundary, investigate a source concern or compare an optional rule. If you are operating a separate X campaign, use X Ads Targeting for that platform's controls. For your FroggyAds purchase, keep the supported settings, actual source evidence and accepted outcome together.
Use this broad framework alongside the specific FroggyAds controls in related resources. A market or device rule defines eligibility; audience planning asks which reachable group fits the offer; campaign optimization examines what to change after evidence arrives. Keep all three decisions connected through the same offer, accepted event and source identifiers. That continuity lets you move from a targeting idea to a campaign decision without treating every topic as a separate audience promise.
| Checkpoint | Targeting action | Evidence to retain |
|---|---|---|
| Eligibility mismatch | Exclude only the market or device that cannot receive the offer. | Verified offer terms and affected campaign scope. |
| Unproven performance concern | Run a capped comparison before changing a broad rule. | Control settings, sample maturity and accepted-event result. |
| Quality or tracking fault | Pause or roll back to the last stable configuration. | Source, destination, event logs, owner and restart condition. |
Hypothetical arithmetic: if a capped test spends USD 125 on media and produces five accepted conversions, media cost per accepted conversion is USD 125 divided by 5, or USD 25.00. Replace both inputs with your own campaign and acceptance records. Report other campaign costs separately when assessing complete acquisition economics. These illustrative figures are not a FroggyAds rate or forecast.
Create your FroggyAds account when you are ready to turn a documented targeting hypothesis into a paid campaign. Choose a supported format, set only the eligibility controls your offer needs, cap the initial exposure, verify the conversion path and decide in advance which mature evidence would justify keeping or changing a rule. Create your free FroggyAds account
Ad targeting is the disciplined control of which paid opportunities can receive an ad and how their results are evaluated. Define the offer's hard eligibility first, test optional signals one at a time, preserve source and conversion evidence, and increase spend only when accepted outcomes support the next decision.
