I worked with a garden-art company whose Facebook ads attracted plenty of clicks. The company turned vintage silverware into handcrafted pieces for outdoor gardens. The products sold on Etsy, the photography was good, and the ads brought people to the website. We tried several campaign approaches. We never established a profitable, repeatable way to sell those products through paid traffic.
I still don't know exactly why. The website was weak. The audience may have enjoyed looking at the pieces more than buying them. Those are possible explanations, and the clicks don't settle either one. My previous successes selling other kinds of metal art didn't settle the question either.
That distinction has to survive every campaign report. A click tells you someone responded to an ad. A purchase tells you someone bought. A profitable acquisition depends on what it cost to get that purchase and what the business keeps after delivering it. Each is a different result, and a campaign can improve the first while failing to produce the last.
Paid acquisition becomes easier to judge when a campaign starts with a claim the business can check. Name the buyer, the offer and the result you're paying to produce. Decide what that result can cost, what evidence would count against the idea, and when spending stops. Then keep the platform's report connected to what happens after the click.
My paid-traffic work starts by understanding who's buying and why. I align the offer with that intent, then structure the campaign and its tests. Google, Meta and LinkedIn give that work different settings and reporting tools. The business still has to decide what success means.
Diagnose the buyer before changing the campaign
A home-decor client was copying a competitor whose founder appeared in the ads. The client sold personalized metal wall art. Copying the appearance left a question unanswered: who was actually buying from this business?
Compare the people who actually bought
I exported the Facebook data to CSV and worked through it in Google Sheets. I grouped the ads by creative, copy and headline, then looked at cohorts such as age, device and placement. Women aged 45 and older stood out in that account. Family and home mattered in the messaging, and ordinary people showing the product in their own surroundings gave us material to work with. One effective photograph showed a piece of metal art on a barn.
Gifts were part of the business, but plenty of people bought the pieces for their own homes. Reducing the audience to grandmothers buying presents would lose what the work taught us. The useful connection was between the product and something the buyer cared about in their home or family.
The device findings also belong to their period. I used iPhone cohorts in that work and later described why changes to the available data made that tactic harder to transfer. An old targeting setting needs fresh evidence before it becomes advice for a current account. The reusable part is how we investigated the buyer.
Follow the campaign through the sale
An account audit extends that investigation through the sale. My forensic audit process starts with exports, separates performance by campaign objective and funnel stage, and reconciles the results with revenue. Audience overlap, old campaigns and unclear reporting are things to investigate. Their presence alone doesn't tell you how much money to cut.
A lead-generation campaign needs a trace through qualification and sales. An ecommerce campaign needs a trace through purchase and the costs attached to fulfilling it. A missing connection can make a cheap lead look valuable when the sales record says otherwise. It can also make a useful campaign look weak while its customers are still moving through the buying process.
Trace that connection before changing the creative. Establish which buyer the evidence describes, whether the offer fits their intent, and which downstream result the account can observe. Those findings give the next campaign something specific to test.
Set the business threshold before the campaign spends
A test needs to connect an audience with an offer and an expected business result. Someone comparing possible solutions may need more explanation before making a commitment. Someone ready to buy needs an offer they can act on. Sending both to the same page can leave the campaign testing several unresolved assumptions at once.
The hypothesis should identify which assumption you're changing. It might concern the buyer, the promise in the ad, the offer or a step between the click and purchase. Keep a control where a comparison is possible. Changing the audience, creative and landing page together may produce a useful commercial result, but it makes the reason for that result harder to isolate.
Set an acquisition cost the business can afford
A platform's cost-per-acquisition target needs a business definition behind it. Specify the acquisition first. A form submission, a qualified opportunity and a paying customer have different values. A lead-cost target also depends on what share of those leads becomes customers and what it costs to serve them.
Lifetime value can inform the limit, but projected revenue won't pay an invoice before the customer pays you. Account for contribution after delivery costs and the time it takes to collect it. A business may have attractive lifetime economics and still be unable to finance a long payback period. Use the evidence available for that business, with uncertainty visible, rather than impose one revenue-to-acquisition ratio on every account.
Operational capacity belongs in the same decision. The solar work involved geography, qualification and installer capacity as well as advertising. A region that generates leads needs a sales and fulfillment operation able to handle them. More demand can create a different problem when that capacity is already committed.
Decide what happens at the limit
A usable campaign brief leaves these decisions written down before launch:
- Which audience and offer are being tested, and what changes against the control.
- Which business outcome counts, and where its record lives.
- What that outcome can cost within a payback period the business can finance.
- How much spending is authorized and when the outcomes will be assessed.
- What result pauses spending, what supports continuation, and who records the decision.
A stop rule protects the testing budget even when the evidence is incomplete. Reaching a spending cap may mean pausing while existing leads mature. It doesn't necessarily prove that the audience won't buy. A broken conversion feed calls for a measurement repair. An offer that fails after an adequate observation window calls for a different commercial decision.
Keep those outcomes distinct. Record whether the idea failed, the test couldn't answer it, or spending stopped because the business had reached its risk limit. The next campaign needs to know which of those happened.
Separate credited sales from sales the campaign added
Connecting an ad to a sale answers an attribution question. It gives you a record of the campaign's involvement in a purchase. Return on ad spend, or ROAS, compares attributed revenue with advertising spend. It still leaves open whether the sale would have happened without the ad, and revenue alone doesn't tell you what remained after costs.
That limit changes how to read a successful report. A campaign reaching people already close to purchase may receive credit for sales it did little to change. A campaign introducing an unfamiliar offer may produce customers whose later purchases appear elsewhere. Neither possibility tells you the size of the effect. You need evidence suited to the question.
Use a holdout to investigate added sales
An incrementality test compares outcomes for a group eligible to receive the campaign with a comparable group held out from it. The difference estimates what the advertising added. Group size and pre-existing differences matter, so comparing raw purchase counts from unequal groups won't answer the question.
Plan how the groups are assigned, what outcome counts and how long purchases need to mature. Decide what size of effect would change the budget decision and whether the available audience can support a useful test. A result too uncertain to distinguish a meaningful effect from variation is inconclusive. It doesn't establish that the campaign had no effect.
Incremental cost per acquisition divides the relevant test spend by estimated additional acquisitions. That denominator differs from the conversions a platform credited to itself. A small or uncertain estimate of additional customers makes a precise-looking cost especially hard to use. Keep that uncertainty attached when deciding whether to expand.
Check what the next customers cost
Average customer acquisition cost describes the customers already acquired across the spending you included. Marginal acquisition cost asks what additional customers cost as spending increases. An affordable account average can coexist with an expansion that costs too much.
Compare the added spending with the added customers over a suitable observation window. Check what else changed, including season, offer, pricing and sales capacity. A before-and-after increase in customers doesn't isolate the advertising's contribution on its own. Use the comparison to investigate expansion, and keep a causal claim within what the test can support.
Marketing mix modeling addresses a broader allocation question. It uses aggregate observations of spending and outcomes, alongside other relevant factors, to estimate channel contributions. The model needs enough variation and a defensible account of influences outside advertising. Channels that change together can be difficult to separate.
A model can help examine how a budget might be distributed across Google, Meta and LinkedIn. Its assumptions still need scrutiny. Holdouts, expansion tests and aggregate models answer different questions at different scales. Choose the method that can inform the decision in front of the business, and retain the limits with the result.
Keep the context that makes a test result reusable
A decision rule helps one campaign. A result record lets another operator understand why the decision was made. My programmatic split-testing work used structured naming, automated campaign creation and templated experiment setups, with results collected in a shared log of wins, losses and context. The machinery reduced the repeated work of setting up and retiring tests.
Setup cost affects which ideas a team can afford to investigate. A test that requires extensive manual rebuilding competes with account maintenance and delivery. Reusing a sound setup gives the team more room to test a different offer or audience. The setup still needs to preserve the comparison. Creating more campaigns while losing the reason each exists produces more records to untangle.
Record what changed and what remained unknown
The log needs enough detail to reconstruct the decision. Record the hypothesis and control, what changed, the spending authorized, the outcome definition and the assessment window. Include the result and the next action. A measurement failure or a test stopped at its loss cap needs a different entry from a failed commercial hypothesis.
Context includes the market and the operation serving it. The solar portfolio work made that visible across accounts. I used Supermetrics to collect data, BigQuery to store it and Google Data Studio to present it. Comparing accounts gave us a way to investigate patterns without treating each client's results as unrelated.
Some creative approaches that worked in other markets didn't work everywhere. We tested locally recognizable imagery, including landmarks in thumbnails, and found useful results. There were also markets we never made profitable. Both findings belong in the record. Keeping only the successful creative would make the playbook look more transferable than the experience justified.
Test a finding again in the next account
Another account can reuse the question, the test setup and the record of where the idea worked. The new campaign still needs evidence that its own buyers respond. A local image may warrant a test in a new region. It doesn't warrant assuming that every regional campaign should use one.
Automate the repeated work when it consumes enough effort to justify maintaining the automation. Naming conventions and a shared log can be useful before a custom system is worthwhile. More testing is valuable when the business can interpret the results and act on them. A volume target alone says little about how much the team learned.
That record also gives automation a boundary. An operator can let software create or monitor a test while keeping the outcome definition, spending authority and stop decision visible to the people responsible for the account.
Check which outcome each platform receives
An automated bidding system works with the conversion events and values configured for it. The operator needs to know what those events mean in the business. Counting a submitted form leaves a gap when success depends on qualification, a completed sale or an installation. Improving the connection between those records makes the account's evidence more useful. It doesn't establish that the campaign will become profitable.
The same investigation applies across Google, Meta and LinkedIn. Start with the campaign's chosen outcome, follow a real customer record downstream, and identify what the platform receives back. Confirm which settings and data connections are available in the actual account before designing the implementation.
Trace the reported event to the business record
A Google lead-generation account needs a way to investigate the connection between an advertising interaction and the CRM's later result. Can the team trace the lead to an opportunity and a closed deal? Can the account receive the outcome needed for the bidding decision? A click identifier can help with that connection, but its presence alone doesn't prove the sales record arrived or matched correctly.
A Meta account needs the same clarity about an event's meaning. Does the reported conversion represent an inquiry, a completed application or a purchase? Does the business record agree? Check for missing or duplicate events before comparing campaign results. An audience suggested by the platform still needs to be judged against the people who actually buy.
A LinkedIn campaign for a business service needs to retain what happens after the initial response. A lead may qualify, be rejected or remain open while a buying decision develops. A report that stops at the inquiry can't answer what acquiring a customer ultimately cost.
Keep unmatched outcomes visible
Use the appropriate permissions and consent for the data involved. Record missing matches and delayed outcomes instead of treating them as zero sales. Check the connection with actual records, then assess whether the signal represents the outcome the campaign was meant to produce.
Automation can handle tactical choices within the account. People still decide which outcome deserves the budget, whether the evidence supports another test, and how much loss or uncertainty the business can carry.
Give one campaign a documented next action
Start with the campaign receiving the most spending. Write down who it targets, what offer it makes and which business outcome justifies continuing it. Use the actual account and sales records. A hypothesis that only exists in the operator's memory leaves everyone else trying to infer the plan from the settings.
Trace the evidence before setting the test
Work through the dependencies in order:
- Trace a reported result to the business record. Establish what was bought or how the lead progressed. Record a missing connection as a measurement problem that needs an owner.
- Set an affordable acquisition cost and a payback period using the business's costs, collection timing and capacity. Keep estimates visibly separate from observed results.
- Name the assumption the next test changes and the comparison that could help assess it. Set the authorized spending and assessment window before the test begins.
- Write the continuation and pause conditions. Include what happens if the test reaches its spending cap before sales mature or if tracking fails.
- Record the result and act on it. Continue within the agreed limits, pause for unresolved evidence, or retire the hypothesis when the result supports doing so. Preserve the reason for the next operator.
Give an extension a new limit
An extension changes the experiment's risk. Give it a reason and a new limit rather than let the campaign keep spending because a decision was postponed. Stopping also deserves a reason. A useful test can end with an unanswered question and a record of what would be needed to answer it.
Paid acquisition requires room for that possibility. A business unable to absorb the testing loss should resolve its offer, evidence or financing constraints before funding another attempt. Previous success in a niche doesn't remove the risk for a different product.
The campaign's job is to produce the business outcome named in its brief. Keep the clicks in the report. Base further spending on what customers bought, what acquiring them cost and what the available evidence can support.