Methodology
How we count money.Conservatively, and out loud.
Every money figure on this site and in the product carries one of three labels. Here's what each means, and what it can't tell you.
Updated
Three numbers, never mixed
- Potential / estimated
- What a finding or a fix should be worth, calculated from your history before or after a change. Useful for prioritizing; not proof.
- Reduced spend
- Money you didn't spend. Only counts as an improvement if results held - spending less alone proves nothing.
- Measured
- Savings shown by a credible causal design (such as a holdout) with uncertainty stated and mature outcomes. The only number that may use the green color.
How potential waste is estimated in the free audit
Each of the 164 checks that finds a problem estimates the spend exposed to it in the period checked (usually the last 30 days): for example, spend on search terms with zero conversions, or spend in hours with no results.
Findings that overlap are de-duplicated so the same money is never counted twice. The total is labeled “estimated”, with its period, and is shown as a monthly figure.
Potential waste is not a promise of savings. Some of that spend may move elsewhere when it's blocked, and some findings need your judgment.
How estimated savings are calculated
These formulas are deliberately conservative. They are always labeled “estimated”.
| Type | Method | Formula |
|---|---|---|
| Waste stopped (negatives, pauses) | Counterfactual | Spend on the blocked items over the 14 days before ÷ 14 × days since × 0.7 - only if the campaign's conversions did not drop |
| Cost per result improved (budget, target, bid) | Before/after, seasonally adjusted | (CPA before − CPA after) × conversions after - only for kept changes with a significant difference (bootstrap, 90%) |
| Revenue gained (target ROAS, PMax) | Before/after | (ROAS after − ROAS before) × spend after × 0.5 |
| Spend protected (a critical alert that was fixed) | Counterfactual | Spend per hour at the time of failure × hours saved until end of day |
| Issue caught | Direct | Counted as an event, worth 0 in money |
What it takes to call savings “measured”
Measured savings require an identified experiment or a credible causal design: for example, an approved holdout with stable randomization, balance checks, a stated minimum detectable effect and enough time for outcomes to mature.
A confidence score or a discount factor never turns an estimate into a measurement. Without valid causal evidence, the measured total stays unavailable and estimates have their own total.
Negative outcomes are recorded as carefully as positive ones, and figures are retracted if the platform restates its data.
Your sales, not the platform's opinion
Where Taseli can see your real orders or leads (through UTM tags today; Shopify and GA4 connections later), results are reconciled against them, with the platform's own figure shown alongside.
Until a connection to orders exists, platform conversions are labeled “Not verified against orders”.
- Only purchases (or verified leads) count as sales. Add-to-cart and begin-checkout are shown as a funnel, never as revenue.
- Results that involve brand campaigns are also calculated without brand traffic. A brand growing on its own is not our saving.
- When you enter a profit margin, we show contribution (profit after media cost) and break-even ROAS, not ROAS alone.
What restoration can and can't reverse
Restoration puts supported settings back as they were. It does not recover money already spent during the monitoring window, and it doesn't necessarily reverse every campaign effect, such as a learning period.
Some changes have no exact restore: items that were removed come back with new IDs and no history, for example. Every proposal says whether its change type can be restored before you approve it, and every restore notice says what wasn't reversed.
Data freshness
Platforms report with delays, and conversions keep arriving for days after a click. Taseli shows the freshness of each data source next to the numbers, and judges changes on mature interaction-date cohorts rather than on the latest partial day.
Figures on this website
Product screenshots and example figures use demo accounts and are labeled “Example”. Aggregate figures (such as total potential waste found in free audits) appear only above a minimum sample of 100 audits, with their counting period, and are never animated upward.
An AI campaigner. You just approve.
Taseli does the work. You make the call.
- Free
- Nothing changes without your approval