How do big companies assess whether their TV ads were useful or not?
In: 5390
So people have given you some halfway decent answers. I will try and be a little mote comprehensive:
Depending on the advert goal, it is measured in a variety of ways, now perception based ads I will come to later, these arent meant to make you buy anything, but simply change your view of the company.
The rest basically comes down to a metric called ROI (Return on investment)
Now, each platform will have its own way of measuring things, but the most basic overview is this:
For each £/$ i spend on platform X, i recieve Y £/$ back.
Usuqlly this is meaured in decimal integers, 1.7 or 12.5 etc. So a TV advert with 1.7 ROI gives you 0.7 £/$ back for each 1 £$ spent.
Now thats the most basic way. The next thing will be a “diminish returns curve”, so its not actually true that it gives back 1.7x, what it does it give back 1.7x at that amount of spend, in reality it will be a much higher number for the first $ spent, and at some point, the return starts to drop off, giving less and less per dollar spent, most brands tend to hover much lower tgan a total investment.
So lets say TV gives you 10 back for the first dollar, then 9 for the next, then 8 etc.. and radio gives you 6 back for the first dollar, when you go to put that dollar into tv that will give you 5 back, you should actually put it in Radio. And then do this optimisation across all channels.making sure each dollar gives the most return.
Now, you do this across all platforms, TV, Radio, Social etc, and you then do a weighted calculation called your Media Mix, and it all gets condensed down to a single value, ROI (Return on investment).
Now after youve been doing this for a long time, you get a baseline, seasonally adjusted, so you know that most of your adverts give you back lets say 3x. Now if you sont change your media mix and have the same number/amound of advert coverage, but you change your advert, new colours, message, branding etc, you can then use thr baseline to see whether the visual changes you made moved the needle and improved or reduced your ROI.
So thats the top level for adverts meant to “drive some behaviour” whether is be making a purchase or clicking a link.
The perception based adverts are slightly different, you take a survey of peolle (a few thousand), at some point in time, then you take part of that group and make sure they dont see your adverts, and the rest, you expose to different groups of adverts (TV, or TV+Radio rtc) and then, you over a period of time, resurvey them in chunks, first batch immediately after (focus groups) and then the next group a day after, then a werk, then a month etc, and you see if the perception of the brand has changed relative to the group exposed to none of your advertising.
On a more targetted level, there are thousands of measures, , click rate, bouncerate, dwell time on page, skip time, time spent watching video, frequency of exposure, how many people it reached etc. Tbese are all what you would call standard metrics. These are all either attribution based (person saw advert and did a thing, or cohort based, these group/demographic did x or y less than this other demographic)
Then there are A/B tests, where you change one small thing and see how it compares to an unedited version, there is also Eyetracking, brain scans, and all these other methids that are a bit more hands on and invasive. But they are more about colour psychology and attention.
Thats the overall view of it.
The most important thing to understand really though is the marketing funnel, with awareness of the brand at the top and consideration/purchase at the bottom, thats how most of it really works and how marketers tend to think.
Nielsen ratings + sub second telemetry on subscription streaming data.
They still exist after having adapted to digital streaming and subscription platforms. Now they have second grain details of what is being watched which they sell to digital platforms as a data service.
The subscription platforms have even finer grain details and can use the Nielsen data to correlate switching streams to the point in time of ads. So they can model whether people watch the ads.
Source: Telemetry Data Engineer.
I am a data scientist and work in a market research company, and we exactly deal with this shit. We collect data ( approximation) on how many people in the locality did watch the ad when it was aired and how many people ended up purchasing it. It’s usually geographical computation. You take a region and you calculate since, calculating over ,let’s say a locality would have high chances of it being null.
Correlate GRPs or essentially reported viewership of ads with sales uplifted above a normalized baseline (ie. What would sales be if all else were equal, accounting for changing weather or price) the more accurately you can link a sale to someone seeing (or not seeing an ad) the better.
Edit: But as someone else mentioned, most companies seem to think that the artistic perception or resonance of their ads is enough. It isn’t.
It’s something which can be proven to have an effect on the small scale, and on the larger scale it becomes much harder to prove but can be assumed to be having SOME effect.
A big example was an isolated experiment by the tobacco companies in the US. They picked a few towns and ran a series of experiments. In a normal town that they all spend a fair amount of money on advertising in, they all had roughly equal shares of the market. Each company was assigned one town that ONLY they would get to advertise in, the others would stay out. And what they saw was that after a few months about 80-90% of the market was JUST buying the advertised brand even though the other brands were available, just not advertised. Then they doubled up with each experimental town shared by two advertisers. After a few months, they both had an almost perfectly even split of the same 80-90% of the market.
And the biggest experiment of all was a set of towns where all the companies entirely removed all advertising. Their products were still on the shelves, but no billboards, tv/radio ads, no posters, etc. And the result? The same as when they were all spending loads of money on advertising, a near even split of the town’s market for tobacco products.
So the result in these small scale tests was that the tobacco companies realized they had a bit of a game of “Prisoner’s Dilemma” on their hands. If everybody advertised (and spent money) then everyone got an equal share. If NOBODY advertised (and saved money) then everyone got an equal share. But if ONE company advertised when nobody else did, that company won the lions share of the market.
The result?
The tobacco companies immediately stopped lobbying against laws meant to restrict tobacco advertisements. If it was legally enforced that NOBODY could advertise, they didn’t have to worry about anyone breaking ranks for individual profit.
One thing many of these answers is missing: data science. Companies will build media mix models which essentially take in media spend as one set of many sets of variables that are used to predict sales quite accurately. Based on the coefficient for TV advertising spend in that model, you can establish a return on advertising spend.
Or better yet, run a test with a holdout region.
As someone who works in sales/marketing, I can tell you that most marketing campaigns (where we display the client’s content across webpages) involves “placing a pixel” on landing sites like their homepage, a product page, etc. We’ll then track how many targets were reached with their ads, how many clicked on the ad (impressions) and how many went to 3 or more pages on the client’s site (engagements).
I work in the industry and can tell you it depends on how advanced the marketing team is. The less sophisticated way is running “tests” where you place ads in one market and not another and see whether it made a difference in sales. The problem with that is there are so many other things that could impact sales other than ads – maybe your competitors ran a sale in one market and not another. Maybe you also had digital ads running. Maybe your product is weather dependent and was nicer in one market and not the other. Maybe your sales people are better at their jobs in one market.
That’s where more advanced techniques like multi-touch attribution or mixed media modeling comes in – basically using statistics to estimate how much each possible explanation (ads, competitors, weather, sales staff, inflation, etc…) likely impacted your sales.
(Edited to include an attempt at an actual ELI5 answer since people seem genuinely interested in the topic:)
Let’s say you have a lemonade stand that you run in your front yard every Saturday in the Summer. Everyday you write in your diary about your day with all the details about the lemonade stand including how many customers you had, what the weather was like, if there was a BBQ or block party on your street and whether that annoying girl Suzy a few blocks down was also running her lemonade stand (she uses powdered lemonade mix not fresh lemons btw). After one summer I looked back at all my notes and learned that on a typical summer afternoon I can expect 20 customers. When it rains I only get around 5 customers. When Suzy is also running her stand I get around 15 customers. When there’s a BBQ I get around 30 customers! Knowing this information I could start to estimate how many customers I can expect on any given day. Now I had the idea to start putting out posters at the front of my neighborhood that say I have FRESH squeezed lemonade. Ever since then I average about 30 customers on a typical summer afternoon, that’s 10 more than before! When it rains I get about 10 customers now and when Suzy is out I get about 26 customers. So I can estimate the posters (advertising) is helping make up for the competition and the rain by about 5-6 customers and gives me around 10 more customers on most days.
This is what the Nielsen rating (and other similar types of data gathering) is for. It’s a survey of ‘typical’ US households, that determines what people are watching at any given time.
TV stations then use that data (not just how many are watching, but also their demographics) to sell ad spots. So ad spots can be priced based on the amount of viewership, and can also be used to target specific kinds of people.
So part of the answer to the question is that companies can actually choose how many eyeballs are exposed to their ad, based on the prediction for that show and time slot. And they get the ‘real’ results later.
And the second half is that they compare it to their sales/revenue figures. If they ran an ad at a specific date, and it corresponds to a big spike in their sales (accounting for delay), they’ll know that the ad was effective.
Sometimes the business will have a customer survey that asks how the customer became aware of the product, in order to try to determine which marketing campaigns are working.
Some ads will have specific promo codes offering a discount. The real purpose of these promo codes is to pinpoint which channel and show the customer was watching or listening to at the time.
Other than that, they will just look at sales data. The data doesn’t specify which marketing campaign worked, but they can try to line up the graphs with the schedule of ad slots. For example, they can look at the February sales graph and see if there’s a noticeable increase in sales in the week after their ad played during the Super Bowl.
Of course, TV ads rarely work, because the vast majority of people treat them as bathroom and snack breaks.
Latest Answers