Radio reaches almost everyone and leaves almost no trace. Measure it with your sales, not the spike

Radio is measured worst where it works best: in the car, hours before anyone can act. Judge it on sales, not a web-traffic spike. Give each station a number read only on air and match its calls to closed sales for a floor, ask new customers how they heard, and switch the schedule on and off to estimate the rest.

Radio has the strangest reputation in advertising. It reaches more American adults than any screen, it is the medium most local business owners grew up hearing their competitors on, and it is routinely ranked by marketers as the least effective thing they could buy. The gap is not about whether radio works. It is about whether anybody can show that it did.

That is a measurement problem, and it has a particular shape. A billboard has an address. A podcast has a download log. A radio spot has neither: it goes out across a whole metro to people who are mostly driving, and whatever they do about it happens later, somewhere else, through a door that does not say "radio" on it. This post covers what the usual radio measurements actually count, why the most popular one misses most of the effect, and how a business with its own sales records can measure the revenue instead.

Why radio is hard to measure: the listener is driving

Radio is hard to measure because most of its audience hears the ad in the car, where nobody can click, write down an address or act on the spot. The response arrives hours later through branded search, the main phone line or a walk-in, credited to whichever route the customer happened to use.

Read those two figures together and the measurement problem explains itself. The audience is enormous, and a large share of it is listening at exactly the moment it is least able to respond. A driver who hears a roofer's ad on the way to work does not pull over to call. She remembers the name, or half of it, and acts at lunch or that evening, by searching for it, by calling a number she already has, or by mentioning it at home to the person who handles the house.

Each of those routes is recorded, and none of them is recorded as radio. The search collects the credit we warned about in organic search, where branded queries are rewarded for demand created somewhere else. The call to the main line lands in Direct / Unknown. The conversation at home becomes, a week later, something that looks exactly like word of mouth. Radio did the work; three other lines in the report take the credit, and one of them is "nothing".

Radio's problem is not reach and not, usually, effectiveness. It is that the response is delayed and arrives through other doors, so any method that looks only at the radio door will undercount it.

What spot-response attribution measures, and what it misses

Spot-response attribution matches the time each ad aired against web traffic in the minutes or hours afterwards, and credits radio with the rise above a baseline. It measures website sessions, not sales, and its answer depends almost entirely on how long a window it allows.

This is the method most radio attribution services sell, and it was borrowed from television, where a spot interrupts somebody sitting in front of a screen with a phone in their hand. The radio industry has published a good deal of it, and the published work is useful mainly for what it reveals about the method's limits.

The radio industry's own trade body in the UK went further, and in doing so made the strongest argument against the very spike charts its American counterparts publish. Its research asked how long a radio spot's effect on web traffic takes to arrive, and the answer was: most of a day.

Two cautions about both studies, which apply to almost everything published about radio's effect. They were commissioned by organizations that sell radio, so the campaigns studied are the ones whose owners agreed to share data, and that is not a random sample. And both measure website sessions. For an online retailer, a session is close to the thing that matters. For a plumber, a law firm or a car dealer, a session is a step on the way to a phone call, which is a step on the way to a sale, and each step loses people.

The 21-hour finding is still the most useful fact in this post for anyone buying radio, because it does not depend on who paid for it. If the effect takes most of a day to arrive, a 20-minute window throws away nearly all of it, and a window long enough to catch it is long enough to catch everything else that happened that day too. The spike chart cannot be fixed by widening it. It has to be replaced by something that looks at sales over weeks.

A spike chart measures the listeners who could act immediately, which on radio is the minority. Judge radio on a window of weeks and on closed sales, or you will cut the medium for doing exactly what it was bought to do.

A number read on air is the cheapest exact measurement in radio

A phone number used only in one station's ads turns the calls that station produces into a lead source. Matched to closed sales by the caller's phone number, it gives an exact floor for that station's revenue: sales that can be traced to it one by one.

This is the same instrument we recommended for a billboard and a podcast, and on radio it is both more valuable and harder to use. More valuable, because the people who call a number they heard on the radio are, overwhelmingly, people who would not otherwise have been recorded at all. Harder, because a number heard once at 60 miles an hour is a number most people will not remember.

1. One number per station. Not one for all radio: a shared number tells you radio produced calls but not which of your stations earned its rate. Split by daypart only if a station's volume is large enough that each number will still ring often.
2. Never print it anywhere else. The moment the station's number is on your website, your trucks or your Business Profile, its calls stop meaning anything about the station.
3. Say it at least twice, and make it easy. A vanity number or a local number with a memorable pattern does more on radio than on any other medium, because the listener has nothing to look at.
4. Record the caller's number. Export the call log with the dialed number, the caller's number and the date. That export is what gets matched to your closed sales.
5. Report the result as a floor. It is the revenue the station can prove, not the revenue it produced.

What the ad asks people to do matters as much as which number it uses. The radio industry's own research on search is blunt about it.

That study is old and measured search rather than sales, but its practical lesson matches what local advertisers have found for decades: tell people where you are, what it costs and what number to call, and leave the web address for media people can look at.

Ask every new customer, because radio is the channel people remember

A required "How did you hear about us?" question at first contact, with "Radio" as an option, captures listeners who never dialed the station's number. Recorded as the lead's source, it turns a memory into a row that can be matched to a closed sale like any other.

Radio is unusual among offline media in that customers name it. People rarely say they saw you on a bus shelter; they often say they heard you on the radio, sometimes with the host's name attached. That makes the intake question more productive for radio than for almost any other channel, and we set out how to design it, with no default answer and the people separated from the routes, in the post on referral revenue.

• Ask at first contact, not at the sale. A receptionist's first question on a call, a required field on the web form. By the time of the sale, memory has blurred.
• Offer "Radio", not a list of stations. Most listeners cannot tell you the station, and a list invites guesses. The station split comes from the numbers.
• Keep the answer in the source column. A lead whose source is "Radio" can be matched and counted. One whose radio mention lives in a notes field cannot.
• Keep it separate from the station numbers. Calls to a station's number and leads who answered "Radio" are different evidence. Label them differently so a report can show both.

Self-report has known faults. People blend media together, credit the last ad they remember, and say "radio" for a TV spot with a jingle. Those errors run in both directions and do not cancel neatly, which is why the answer is a source to record rather than a verdict to trust. Its value is that it catches the listener who called the main line, which the station's number never will.

Switch the schedule on and off to measure the rest

A flighted comparison alternates weeks with radio and weeks without, then compares closed sales between them. Because a metro station covers the whole market, there is usually no unexposed area to compare against, so time has to be the control, and the schedule has to be designed for it before it airs.

For a billboard, geography does the work: the board is visible in some ZIP codes and not others. Radio does not give a local business that option. A station's signal covers the metro, and a business that serves one metro has no untouched region to compare with. What it can control is when the ads run. Radio is bought by the week, and many local advertisers already run in flights. The only change is to arrange them so the off weeks are a measurement rather than a gap.

Nielsen gives radio advertisers using marketing mix models three pieces of advice in the same Audio Today report, and all three apply at the scale of a single business: use as-run data rather than planned delivery, look at delivery market by market, and analyze by week. The as-run log, sometimes called an affidavit, is the station's record of when each spot actually aired. Ask for it every month. A schedule that was planned for twelve spots a week and ran nine is a different test.

Here is what the whole method looks like for an HVAC company running radio on two stations in an eight-week flighted schedule, two weeks on and two weeks off, at $3,000 per on week: $1,800 on a news-talk station and $1,200 on a country station. Its average closed job is $4,800.

The arithmetic: four on weeks at $1,800 is $7,200 for the news-talk station and four at $1,200 is $4,800 for the country station. Six matched jobs at $4,800 is $28,800, or $28,800 ÷ $7,200 = $4.00 per dollar; three is $14,400, or $14,400 ÷ $4,800 = $3.00. The intake answers add four jobs, $19,200, that cannot be split between stations. In total, $62,400 ÷ $12,000 = $5.20 of matched revenue per dollar. That is the floor, and every line of it is a named customer.

Now the comparison. In the four on weeks the company closed 29, 31, 30 and 32 jobs: 122 in total, 30.5 a week. In the four off weeks it closed 26, 27, 25 and 26: 104, or 26 a week. The four weeks before the campaign averaged 25.5. The gap between on and off weeks is 122 − 104 = 18 jobs, and 18 × $4,800 = $86,400, or $86,400 ÷ $12,000 = $7.20 per dollar.

Three things keep that figure honest. The 13 matched jobs sit inside the 18, so the two numbers overlap and must never be added. The off weeks are not clean: they averaged 26 against 25.5 before the campaign, consistent with radio's delayed effect spilling into them, which makes the comparison conservative rather than generous. And HVAC demand follows the weather, so a heat wave in one on week would ruin the test; short alternating flights inside one season are the defense, and a test that coincides with a weather event should be rerun, not reported. Our post on incrementality tests covers how much volume a comparison like this needs before its answer is more than noise.

Report $62,400 as attributed and $86,400 as the estimated effect, side by side and never summed. The first is a list of customers; the second is an argument about eight weeks. Both are worth having, and they are not the same kind of fact.

What radio genuinely cannot be measured for

Some of radio's effect cannot be traced to a person or a station by any method a local business can run: listeners who searched your name or called your main line without mentioning radio, the split between stations sharing a schedule, and brand effects that build over months.

It is better to say this plainly than to let a vendor fill the silence with a model.

• The individual listener who took another door. A customer who heard the ad, searched your name and clicked your search ad is, in your records, a search customer. Nothing short of tracking that person's movements can say otherwise, and you should not want to.
• The station behind an unprompted response. The on-off comparison measures radio as a whole. Two stations running in the same weeks cannot be separated by it; stagger them if the split matters enough to be worth the delay.
• The long tail. Some of what radio does is make your name familiar enough that a customer picks you in eight months, when the furnace fails. No two-week comparison sees that, and a mix model needs years of data and a much larger budget to estimate it. Our comparison of attribution and mix modeling covers when that becomes worth doing.
• A schedule too small to register. If the station's number barely rings and the on weeks look like the off weeks, the test has not proved radio fails. It has proved the schedule was too light to measure, which is often the same as too light to work.

That last point is the arguable one, and we will argue it. Many local businesses buy radio the way they buy a lottery ticket: a few spots a week on several stations, renewed because canceling feels risky. A schedule that cannot be seen in matched calls or in a flighted comparison is a schedule nobody can defend. Buy fewer stations with more weight, measure them properly for one season, and let the result decide the next one. Radio that works will show it; radio that does not is cheaper to find out about than to keep.

CloseRev matches people, not broadcasts. If your lead file has a row for each radio-driven lead, such as a call to a number used only on air or a "How did you hear about us?" answer of "Radio", with its phone or email and "Radio" (or the station's name) in the source column, it matches those leads to closed sales like any other source and reports them under exactly that label. A listener who never became a lead stays in Direct / Unknown; it does not estimate reach, lift or what the spot did elsewhere.

What the radio line in a report should look like

A defensible radio report has three parts: matched revenue per station number and per intake answer, stated as a floor; the on-off comparison beside it, with its weeks, as-run logs and caveats; and ratings or reach figures only as the basis for choosing stations, never as evidence of revenue.

The mistake to avoid is the one radio invites most: collapsing evidence of different kinds into a single return. A station rep's reach figure, an attribution vendor's web-session lift and your own matched sales are three answers to three different questions. Added together, they produce a number that nobody can check and that will be believed exactly until somebody asks where it came from.

Kept apart, they make a strong case. The matched sales show that specific customers came through the station's number and through the intake question. The comparison shows the size of everything the number missed. The rest of what radio does, like the rest of what a direct mail drop or a billboard does, stays honestly unattributed, and the size of the unattributed bucket is itself evidence: watch whether it grows in the weeks the schedule runs.

Radio is not unmeasurable. It is badly measured by methods built for screens. Measure it the way the listener actually responds, later and through other doors, with a number, a question and a schedule that switches off, and the argument about radio becomes an argument about evidence.