Data interpretation mistakes that sabotage digital launches

This is the most classic mistake and also the most dangerous, because it often confirms what you already wanted to believe. When two events happen together, the human brain automatically assumes one caused the other. In the context of launches, this shows up in subtle and convincing ways.

Picture this scenario: you ran a Stories poll asking "What is your biggest struggle with paid traffic?" the day before the cart opened. The launch sold 40% more than the previous one. Immediate conclusion: the poll warmed up the audience and drove the extra sales.

But was it really the poll? Or was it the fact that, in this launch, you had 35% more leads in the base? Or that the last CPL had a much higher retention rate? Or that a strong competitor did not launch the same week? The poll and the sales happened together: that is correlation. To establish causation, you would need to isolate the variable, that is, compare a group that saw the poll with a group that did not, keeping everything else the same.

How to avoid it: before attributing a result to a specific action, ask yourself: "What other variables changed at the same time?" If you cannot isolate the variable, record the hypothesis, but do not treat it as a confirmed fact. In future launches, test the same action in isolation to validate it.

Survivorship bias happens when you analyze only the success stories and ignore the failures. In the digital market, this bias is epidemic and fed by social media.

Hypothetical scenario: 5 digital creators share that they ran a 5-day challenge before the launch and had incredible results. You conclude that 5-day challenges are the ideal strategy. What you do not see are the 30 creators who ran the same 5-day challenge and got mediocre or negative results. They did not post about it. They were not invited to podcasts. They did not write threads on Twitter.

This bias also shows up when analyzing your own data. If you only look at the profile of those who bought to understand "what works", you are making the same mistake. As important as knowing what buyers have in common is understanding why non-buyers did not buy. Leads who watched every CPL, visited the sales page three times and still did not buy carry valuable information about unresolved objections, price problems or communication failures.

How to avoid it: whenever you analyze results, deliberately include the negative cases. Compare buyers with non-buyers. Analyze who abandoned the cart. Study who stopped opening your emails. The data of the "survivors" tells only half the story.

This mistake costs real money, and a lot of it. Ad platforms (Meta, Google, TikTok) have structural incentives to report optimistic results. It is not necessarily dishonesty; it is how attribution models work.

Hypothetical example: Meta Ads reports a ROAS of 8.5x for your launch campaign. You celebrate, raise the spend and plan the next launch around that number. But when you cross it with the real data (sales confirmed at checkout, minus refunds and chargebacks), the real ROAS is 4.2x. A 141% gap.

Where does that difference come from? Three main factors inflate the reported numbers:

• View-through conversions: the platform attributes the sale to someone who only saw the ad, without clicking. A person who saw your ad in the feed, ignored it, and later bought on a friend's recommendation shows up as a campaign "conversion".
• Extended attribution window: with 28-day windows, the platform links sales that may have been influenced by dozens of other factors between the click and the purchase.
• Duplicate attribution: if the lead clicked ads on Meta and Google before buying, both platforms claim the conversion. Adding up the numbers of both, you count the same sale twice.

How to avoid it: always cross the platform's ROAS with real revenue data. Use consistent UTMs and a centralized attribution system. Treat the platform ROAS as a directional indicator, not as absolute truth. And when making investment decisions, use the real number, even if it is less exciting.

The attribution window is the period the platform uses to link a click (or view) to a conversion. Changing that window changes the numbers dramatically, and many launchers do not even know which window they are using.

The same data set can tell completely different stories depending on the chosen window. A campaign that generated 45 conversions in the 28-day window may show only 12 conversions in the 7-day window and a mere 3 conversions in the 1-day window. None of the numbers is "wrong": they answer different questions.

The problem appears when you compare campaigns or launches using different windows. If you used a 28-day window in the previous launch and a 7-day window in this one, the numbers cannot be compared. An apparent drop in performance may simply be a change of window.

The same goes for comparisons between platforms: Meta uses a 7-day click window by default, Google may use 30 days, TikTok uses 28 days. Comparing ROAS across platforms without normalizing the window is comparing kilometers with miles.

How to avoid it: set a standard attribution window for all your analyses and keep it consistent across launches and platforms. If you need to change it, recalculate the previous data with the new window before comparing. Always document which window each report uses.

In the heat of a launch, there is enormous pressure to decide fast. The problem is that early data with small samples is statistically unreliable, and it can lead to disastrous decisions.

Scenario: you are testing two creatives. After 4 hours, Creative A has 150 impressions with a 3.2% CTR and Creative B has 150 impressions with a 1.8% CTR. The temptation to pause B and scale A is huge. But with 150 impressions, the statistical margin of error is so large that both creatives may actually perform the same. What you are seeing may be just random noise.

To have reasonable statistical confidence in a CTR difference, you need at least 1,000 to 2,000 impressions per creative. For conversions (rarer events), the number needed is even higher. Deciding before that is essentially flipping a coin and believing you are doing science.

This mistake is particularly dangerous at the start of launch traffic campaigns, when every hour feels crucial and anxiety pushes for immediate action.

How to avoid it: before starting the test, define the minimum sample size you will wait for before deciding. For CTR, wait for at least 1,000 impressions per variation. For conversion rate, wait for at least 100 conversions per variation. If the numbers look very good or very bad on a small sample, record the observation, but do not act until you reach the minimum volume.

"This launch made R$ 320,000 against R$ 280,000 for the previous one. We grew 14%." On its own, that sentence sounds like great news. But the context can reveal a completely different story.

If the previous launch spent R$ 40,000 on traffic and this one spent R$ 80,000, revenue went up 14% while spend doubled. ROI was cut in half. If the lead base doubled in size between one launch and the next, the conversion rate actually dropped significantly. If the previous launch was in January (a weak month for sales in the digital market) and this one was in March (a strong month), part of the growth is simply seasonality.

Raw revenue comparisons between launches are almost always misleading because launches never happen under identical conditions. The market changes, the base changes, the spend changes, the offer changes, the competition changes. Attributing the difference in results to a single variable, "it was because we changed the CPL format", ignores all the other variables that changed too.

How to avoid it: when comparing launches, normalize the data. Compare ROI (not absolute revenue), conversion rate (not the absolute number of sales), cost per lead adjusted for quality. Document the conditions of each launch (base size, spend, seasonality, offer, competition) so future comparisons have context.

Vanity metrics are numbers that look impressive in the report but do not translate into financial results. They feed the ego and the success narrative, but obscure the reality of the launch's performance.

Classic scenario: the post-launch report highlights 2.3 million ad impressions, 45,000 likes on the content and 12,000 comments on the lives. The presentation looks beautiful. But revenue came in 30% below target.

Impressions do not pay the bills. Likes are not conversions. Comments are not sales. These numbers measure reach and surface engagement, which matter as intermediate indicators, but should never be confused with result indicators.

The real danger of vanity metrics is that they can mask serious problems. A launch with massive reach and low conversion has an offer, price or sales communication problem, but if the team is celebrating millions of impressions, that problem can go unnoticed until it is too late to fix.

How to avoid it: before the launch, define your result metrics (revenue, ROI, conversion rate, average order value, CAC) and your process metrics (impressions, CTR, engagement). Reports should always start with the result metrics. Process metrics are useful for diagnosis when results fall short, but should never be presented as a substitute for the result.

This is the quietest mistake of all because it distorts not only the analysis of one launch, but your whole long-term strategy.

A seed launch that made R$ 35,000 may look like a failure next to production and traffic costs. But if the 800 buyers from that seed launch form a warm base that, in the second launch, contributes to R$ 210,000 in revenue, the first launch was not a loss: it was an investment in customer acquisition.

The cumulative effect shows up in several dimensions that are often ignored when each launch is analyzed in isolation:

• A base of repeat buyers: each launch feeds the next with a share of previous buyers whose conversion rate is 5 to 10 times higher than cold leads.
• Accumulated social proof: testimonials, cases and results from previous students strengthen the communication of the following launches.
• Brand recognition: every launch, even the smaller ones, raises recognition and lowers the trust barrier for future launches.
• Data learning: insights about what works and what does not in communication, offer and traffic accumulate and make each following launch more efficient.

When you analyze each launch in isolation, without considering how it feeds the next ones, you tend to underestimate early launches and overestimate the performance of mature launches (which benefit from all the previous work).

How to avoid it: keep a cohort view of your launches. Track the LTV (Lifetime Value) of leads and buyers across multiple launches, not just the revenue of each individual launch. Treat seed and pilot launches as acquisition investments, with the return expected over time. Document which assets each launch produced for the next ones: base, testimonials, learnings, waitlist.

All these mistakes have something in common: they do not happen for lack of data, but for lack of method in interpreting it. The numbers are there. The problem is what we do with them. Confirmation bias, the rush for results, the pressure for success stories: all of it conspires to make us read data conveniently instead of accurately.

The good news is that, once you know these mistakes, they become much easier to spot day to day. Question automatic correlations. Include the failures in the analysis. Cross platform data with real data. Wait for meaningful samples. Normalize comparisons. Prioritize result metrics. Think long term.

With Kiron, you bring traffic, capture, engagement and sales data into a single panel, eliminating the fragmentation that feeds many of these mistakes. By automatically crossing data from different sources and showing real funnel metrics, Kiron Funnels helps you read your data clearly and make decisions based on evidence, not illusion.