A 60-second spot in the NBA Finals. A season-long sponsorship of the country’s biggest sports league. A celebrity partnership announced on every feed at once. A national TV push timed to your biggest launch of the year.

Everyone sees them. Nobody can say exactly what they returned.

The measurement gap

Most teams report these channels with reach, impressions and platform-attributed revenue. Those numbers are useful. They just don’t hold up when finance asks the real question: what did this channel actually add?

A geo test is the usual answer. But you can’t hold out a region from a channel that runs everywhere at once. And many brands, especially single-country ones, don’t have the 20 or more comparable regions a geo test needs.

So the channel keeps its budget on trust, or loses it on a hunch. Time Testing closes that gap.

What is a time test?

A time test measures the incremental impact of one meaningful change in spend, when you can’t build a test and control group by region. It works for national TV, radio, podcasts, influencer, sponsorships and organic social, and for brands with too few regions for a geo test.

Instead of comparing regions, it compares time. Before the change, Lifesight builds a forecast of what your KPI would do if nothing changed. You then pause, cut or raise spend on one channel for a set window. When the window ends, Lifesight compares what actually happened with the forecast. The gap is the channel’s incremental impact.

[Image: forecast vs. actual chart, with the incremental lift shaded during the test window]

There are two ways to run one:

  • Hold-out: pause or cut spend to measure what you would lose without the channel. A drop against the forecast means the channel was driving results.
  • Scale-up: raise spend to measure what more budget would return. A rise against the forecast shows the return on the extra spend.

No geo split and no user-level tracking. Just the KPI history you already have, for revenue, orders, conversions, sign-ups, leads or app installs.

Time Testing sits next to geo testing in Measurement > Experiments.

Know what your test can detect before you spend

The most expensive test is the one that comes back inconclusive. You pause a channel for six weeks, lose the revenue, and learn nothing. So Lifesight does the design work before any budget moves.

When you set up a time test, the design step shows:

  • The smallest lift the test can detect, at 95% confidence and 80% power. If you expect a 5% drop and the test can detect 2.3%, it’s well sized.
  • How long the test needs to run to find the lift you expect. Tests can run from 7 to 180 days.
  • Spend scenarios: the same check for spend changes 10%, 20%, 30% and 50% bigger or smaller than your plan, so you can find the change that gives a clear answer.

If the window is too short to read, Lifesight tells you. For example, a 30-day test on weekly data has only four data points, which is rarely enough. You can run it longer, use daily data, or make a bigger change, before you’ve spent anything.

The design step shows the smallest detectable lift, and how it is worked out.

A forecast built to hold up

Without a control group, a time test is only as good as its forecast. So Lifesight doesn’t rely on one model.

The forecast combines five models: Prophet, NeuralProphet, SARIMAX, CatBoost and a Bayesian model. Each one is scored out-of-sample, on recent weeks it did not train on. More accurate models get more weight. Models that miss by more than 40% are left out.

The forecast learns from your history. You need at least 16 weeks of weekly data or 56 days of daily data. Fifteen months is the default, so the forecast covers a full year of seasonality plus your recent trend. Up to 24 months is used.

Guardrails that protect the result

A time test counts anything else that changes in the window as lift. Lifesight is built to catch the common problems before they cost you a test:

  • It warns you that sales, stock-outs, price changes or news in the window will show up as lift.
  • It flags a test with too few data points, and a forecast built on old data.
  • It warns you when the planned spend doesn’t match the test type, such as higher spend on a hold-out.
  • It checks your history file and tells you which row to fix if a date is missing, the spacing is wrong, or a value isn’t a number.

How time tests are different

Time tests, geo tests and MMM each answer a different question. Here’s how they compare.

Time test Geo test MMM
What it answers What did one channel add during a set window? What did one channel add in test regions vs. control regions? How should budget split across every channel?
The control A forecast built from your own KPI history Matched regions with no change A model of all channels over time
Best for Channels that run everywhere at once, or brands with too few regions Channels you can split by region, with about 20 or more comparable regions Planning the full mix, sharpened with test results
Watch out for Anything else that changes in the window shows up as lift Spillover between regions and unreliable location targeting Needs test results to stay honest on hard-to-measure channels

 

And compared with a simple before-and-after look at your numbers, a time test is planned and sized in advance. Its forecast accounts for your trend and seasonality, so a slow month isn’t mistaken for a channel effect.

In Lifesight, all three methods sit in one platform. You pick the test that fits each channel, then enter the results into your MMM to check and calibrate it, and plan next quarter’s budget.

From lift to return on spend

When the test window ends, you upload your KPI for it, and Lifesight measures the result:

  • Lift: how far your KPI moved from the forecast during the test.
  • Incremental KPI: the extra or lost revenue, orders, sign-ups or installs.
  • iROAS and cost per incremental result: when you include spend in your history file.
  • A plain-language readout: what the result means for the channel.
  • Post-treatment effect: whether the impact continued after spend returned to normal. This matters for TV and podcasts, where the effect often outlasts the flight.

An inconclusive result doesn’t mean the channel has no effect. It means the effect was smaller than the test could detect. A longer test or a bigger spend change can find smaller effects.

Now you can show finance exactly what the channel returned, and decide whether to keep, cut or grow it.

When to run a time test, and when not to

Run a time test when:

  • The channel can’t be split by region.
  • You have too few regions for a geo test.
  • Regional targeting or spillover would make a geo test unreliable.

Choose a geo test, or move the time test, when:

  • The window overlaps a sale, a launch, a price change or peak season.
  • You can split the channel by region and have enough regions. A geo test will give the stronger read.

Measure your biggest channel

Time Testing is live in Lifesight today. Go to Measurement > Experiments > New Experiment > Time Testing. Set up, upload your history, size the test, and review. It takes four steps.

Know what your hardest-to-measure channel really drives. To see Time Testing on your own data, book a demo.