Commentary
AI Won't Be Biggest Thing Since Sliced Bread -- But WILL Change TV Advertising
- by Dave Morgan , Featured Contributor, January 18, 2018
However, to believe that we will all wake up in three years and find that our lives as advertising and media professionals will have been transformed by AI is ludicrous. The impacts of self-learning computers will take many years to really manifest themselves. Besides, it’s hard to imagine that an industry like advertising, embracing basic empiricism and data-based decision-making, is going to be changed very much or very quickly just because computers will be able to give people better numbers sooner.
That said, however, I do believe that the TV ad industry, where I spend my time these days, will actually see some not-insignificant effects from AI in the near term. Here are two examples that jump to mind:
advertisement
advertisement
Digital-like attribution. Today, most TV ad attribution to sales is based on mix models that rely heavily on correlation and regression-based models. They look to the past, not the future, cover long time frames, and are primarily useful only for strategic media and marketing allocation decisions.
With AI’s capacity to run increasingly sophisticated predictive models against massive disparate data sets of media exposures and consumer actions and purchasing, linked at the impression and household level, we are already starting to see media attribution at the campaign level with causation provable at levels more akin to how the pharmaceutical industry works. Thanks to AI, we can now build and run powerful synthetic tests and controls on campaigns. Finally, TV is now able to be compared to digital when it comes to ad attribution.
TV plans and buys that actually match up. It’s a running joke in TV advertising that what’s planned is never bought. How can it? TV plans are created months and quarters before campaigns are executed. At that point, no one even knows what the shows will be. Instead, the industry falls back on defining everything with really simple — and extraordinarily blunt — audience metrics like demographics, gross rating points, networks, days, dayparts and average rating points.
That approach was fine for the TV world of the 1980s dominated by three broadcast networks. It’s worthless today.
Buying demo-based GRPs in today’s fragmented TV world, where 75+% of national viewing occurs on episodes with a rating below 0.5, is a recipe for buying massive amounts of wasted frequency. Campaign plan optimizers powered by artificial intelligence today can actually evaluate and optimize against potential audience duplication at the household level for every potential combination of spots on all networks across all dayparts.
To put this into perspective: There are 250,000 different ad spots shown on all Nielsen-rated national networks every day. The number of potential different audience reach and frequency combinations that could be generated by those spots over a three-week campaign is N to the power of 34.
How big is that? That number is greater than the number of atoms in the known universe. However, an AI-powered plan generator — with the right data available — can spin those calculations in less than 10 minutes. Most importantly, the AI-generated plan is likely to deliver 25% more reach on average than legacy plans, and probably twice the ROI.
Not ready to talk AI yet? Don’t like acronyms invading your business? No worries. Most of this work will go on beneath the surface of the systems that you use. All you need to worry about is demanding the double ROI that AI might bring you.
What do you think?
Dave, I agree with you about the hyping of AI. Hoewver the fact that the vast majority of TV show airings are now low rated---due to the huge number of competing channels----does not necessarily corelate with excessive frequency. The basic reason why most TV shows now earn very low per-telecast ratings is because those who sample them watch them less often---not more often.
That said, we should, perhaps, redefine what we mean by "waste" when thinking about audience tonnage ---or "impressions". The plain fact is that 20% of all adults consume about half of all TV content, while, at the opposite end of the TV spectrum, another "quintile" accounts for barely 3-4%. The former are mostly older consumers and are heavily oriented to low brow groupings by education. income, etc. This makes them "less desirable", although not worthless targets for many advertisers. But the key point is the fact that in most cases all of the competing brands in a given category use the same media---with TV as their mainstay. Consequently, most brands' TV schedules contain a lot of "wasted" impressions. Which usually means that what really counts, aside from the product positioning and creative execution of the ads, is share of voice. If your brand has an overall share of TV ad impressions for its category of 10%, chances are that it will be in this approximate position an all of the TV quinties---heaviest viewing to lightest. And, in most cases its share of sales will not vary that much either. It is possible to verify whether this is or is not the case via Simmons or MRI. In order to reconfigure TV GRP weight so the lighter---younger/upscale----consumer gets more impressions, you must accept that you will sacrifice impressions---and share of voice---among the heaviest viewing group. Whether this is a good or bad idea, varies from advertiser to advertiser but there is no solid evidence that the trade-off---losing three to five heavy viewer impressions for every one gained among light viewers--- always pays out in ROI.
I happen to believe in better targeting methods---when they are people, not household based---- and would like to see a system where advertisers allocate part of their TV spending to massive, broadly targeted corporate upfront CPM-driven buys as an efficiency hedge, but also place some of their money purchases in a brand-specific upfront as well as scatter deals. Here we would see to what extent real improvements could be made, though even the most refined targeting techniques will, no doubt garner more GRPs against those ever present heavy viewers.
Ed, great points. I totally agree about the share of voice argurment. However, what I should have been clearer about is that it is not fragmentation that creates heavy frequency buys, it is buying on large average rating points that creates frequency heavy campaigns in a fragemtned world. Owning share of voice is great, but it's not so great if you only own it with one quintil of TV viewershp. That is what is happening today.
"Instead, the industry falls back on defining everything with really simple — and extraordinarily blunt — audience metrics like demographics, gross rating points, networks, days, dayparts and average rating points.”
Forget sliced bread, think bagels with a schmear - buy the content not the audience. What if you could buy programs with high correlation to current customers and lookalikes. Or programs with a high affinity (sentiment, emotion, tone) to the brand position; e.g., your customers love a particular story arc and emotional narrative. (https://goo.gl/WtCoiA)
For example, GoPro cameras and/or Red Bull appeal to risk takers, dreamers, action-oriented, fun loving and creative people. There are lots of tools to understand brand persona and appeal.
Using a variety of AI tools (IBM Watson, Identv, Graymatics) every video could be analyzed and clustered according to attributes like sentiment, tone, concepts, entities, keywords, and more. AI would find patterns in the types of programming that are consistent with the brand position and over time determine which program attributes are key to engagement.
Great points Larry. Yes. There is no question that AI tools will help marketer buy TV content according to indexes of target audiences who are likely to watch that content. And, more importantly, it will insure that the campaigns can be predictively tuned to optimal reach or frequency, since one problem with buying according to audience indexing is that you tend to end up with high frequency campaigns - continueing to hit the same target members. This is where AI's ability to sort through mathamatically massive amounts of combinations makes a big difference.
Dave, when an advertiser buys TV audience tonnage on a GRP basis---as this is the only way the sellers will guarantee GRP delivery-----this does not necessarily mean that the resulting ad schedule is much more heavily weighted towards frequency. Actually, if the buy follows a sensible plan that takes into account a mix of network types and dayparts, you can come pretty close to the reach and frequency goals you have defined. Also, when I speak of share of voice, I refer to all quintiles, not just the heaviest one. Exceptions aside, if you look at the typical brand's share of TV GRPs in each of the quintiles, in cases where all of the brands use TV in the same manner, the most common outcome is relative parity. This means that most or all of the brands have the same share of GRPs---or approximately so---in each quintile.
Returning to the frequency issue, if one compares current TV ad schedules with what was true 30 years ago, the main difference is not so much a loss of monthly reach but rather, a more reduced frequency---GRPs. This is a function not only of rating attrition but to a far greater extent of greatly increased costs---or cost per rating point. The typical ad dollar now buys many fewer rating points than before---hence consumers are being "reached" less often by the tyoical TV ad campaign not more frequently..
I do think that we are on the same page regarding the desirability of improving upon the buying and selling process---if more of the buys are freed from the corporate CPM rules everything trap----but this is a battle that has yet to be won, I'm afraid.
Ed, 10 years ago you would have been correct to assume that basic planning techniques could keep GRP-based TV buys from deliering too much of their weight in frequency. Today, however, it is absolutely the case unless the buys are heavied up against mid and long tail cable. 75% of national TV viewing volume now occurs on episodes with a rating under 0.5. If a buy is GRP focused and aggregates spots with high average rating points (a best practice approach for decades) the campaign will dramatically over deliver frequency against heavy TV viewers. Erwin Ephron warned us about this decades ago. Today, his warnings are reality.
Dave, I'm afraid that I must disagree with you. The way that most media planners handle the reach vs frequency issue is by specififying a certain mix of network types and dayparts, in the process garnering the approximate reach and frequency levels that their budgets allow for. Of course, if all that a buyer does is buy GRPs, without regard to program/channel spread or dayparts, a CPM-driven purchase may well concentrate the brand on a few channels and a few programs, which could compromise the balance between reach and frequency. I believe that when corporate buys are allocated to the brands, that most advertiser media mavens try to take into account reach/frequency considerations as well as demos and other factors. If not, then those brands may have a problem.
As regards Erwin, my friend who I brought into the media business and partnered with for a long spell, I don't think that his notion that all you need is a one frequency per week ---"the "recency theory"----has anything to do with what we are discussing. You seem to be saying that today's TV advertiser is wasting too many of his/her "impressions due to excessive GRPs at the expense of reach. I don't think so. If any waste is involved it's caused by failure to more effectively target consumers, not only in terms of defining them as possible product buyers but far more important, by failing to consider their minsets and how these relate to what the brands are actually saying about themselves.
Ed, we don't agree as much as it seems. There is nothing wrong with the planning. However, it is no longer true that what is planned is bought. The majority of most national TV campaigns GRP's today are delivered through "packagin," not the planned-out spots that were hoped for those many months before.
On the Erwin issue, he is the article I am referring to. It is not about recency, but about maximizing reach in an increasingly fragmented world: https://www.warc.com/content/article/esomar/learning_to_live_in_lilliput,_the_media_land_where_small_is_beautiful_optimizing_reach_with_low_ratings_and_other_thoughts_on_tv_fragmentation/9286
Thanks so much for the great back-and-forth!
Dave, just to be clear, is it your position that an upfront corporate TV buy, once it is allocated to a given brand is so heavily weighted towards getting "impressions" at a low CPM that a brand, within the corporate umbrella, can't really come close to its reach/frequency goals because so many of the corporation's GRPs are redundant. And this is because so many placements run in shows with low ratings? For example, if a brand specifies that it needs 300 GRPs in a month, this will probably yield it about a 65% monthly reach with an average frequency of 4.6. Do you believe that what is really happening under the corporate buying umbrella is that the brand that wants a 65% reach for its 300 GRPs actually is getting a reach of only 40-45% with an average frequency of about 7?If that's is actually the case and this is happening on a widespread basis, I'd have to agree with you....but I have not seen such extreme imbalances between planned and delivered schedules in my work with clients. In any event, despite this disagreement, I still feel that we are in agreement in princilpe, but have very different ways of getting our message across.
Ed, ... oops ... just realized that my last comment should have started ... "we don't [dis]agree as much as it seems" ... sorry about the typo!
Ed, Yes. that is my point. When the big coporate deals, which are heavy in cheap GRP's finally delivery, they have much higher frequency than was planned, with a lot of that dispartiy between the heavy and light TV viewers. And, it's getting worse. This is why no one "posts" on de-averaged person level reach. The results are too ugly. Everyone now just looks at the averages. For most TV campaigns, the dispartiy between all quintiles of TV viewing and exposure to TV ads at the campaign level is massively weighted to heavy TV viewers, with the lightest TV viewers getting very few top brand oriented TV ads. Instead, they tend to get heavy direct resposne schedules. More intelligent computer planning can help a lot. Actually making sure that the buys deliver on the promises of the planning would help incredibly.
Thanks for that clarification, Dave. I'll have to take a closer look at what is going on in this regard, currently;. I have noted a tendency in the big upfront deals to limit the number of sellers that are given big schedules---as a way to earn slightly lower CPMs----so it will be interesting to ask some of my contacts to compare the schedules their brands are getting vs. what was planned---or hoped for. More on this later.
A wonderfully insightful exchange between two of my favorite analysts, Dave and Ed. What you both seem to be saying is that it is important to know, understand, and optimize against "cost per reach point" as opposed to "cost per point" as the latter focuses on cheap(er) impressions as opposed to cheap(er) reach. Focusing on cheaper impression generates high frequency -- and Erwin called frequency the "crabrass" of media. This is also what Turner and Barry Fischer were saying a decade ago with "Media at the Millennium." You can optimize against a demo target as we have conventionally, or optimize against category buyers (if you have good enough data to do so), but optimizing against reach (cost per reach point) rather than impressions (cost per point) is the winning strategy. Using broad dayparts and/or cross platform exposure is only part of the solution as the "law of double jeopardy" teaches us that heavy users/viewers are the ones who watch lower rated shows. Like small brands, low rated shows suffer from two problems, few viewers and those that watch them watch lots of TV. A show with a high rating attracts viewers that low rated shows don't -- light viewers. Heavy viewers watch lots of shows -- high and low rated.
I'm with Jack - loving it!
One of my regrets is that I never got to NY to have that lunch with Erwin. I admit I proselitised much of what he wrote within the agency staff and client base I worked with. OK, I might have even bored them stiff.
Jack we used to call it CPIRP - Cost Per Incremental Reach Point. One client who didn't really "get it" wanted to optimise the buy so that as soon as the incremental reach cost was increasing then stop the buy. I politely explained (well, for me) that would mean a single spot campaign as it was inevitable that the second spot had audience duplication with the first and that the cost per reah point for that spot would be higher.
Dave & Ed, could the answer lie in that what Dave is seeing is the impact of the increase in programmatic buying? It's focus is on lowest campaign cost by targeting low cost inventory. Sure there can be parameters (such as capping share by channel) but what are your thoughts on whetheer we are now relying too much on algorithms and not enough on knowledge, experience and intuition.
John, "programmatic" time buying, where all or most of the available GRPs are fed into an automated buying/selling system simply does not exist in "linear TV". If it did, you and Dave would be absolutely right. Having no way to address issues like reach, program quality or merchandisablilty, the computers would ---if allowed to---buy the lowest CPM shows and/or channels and shun primetime, news and sports. The result, if such a nightmare scenario took place would be a brand that once ran five or six 30s in broadcast network primetime per month along with late night and basic cable prime would suddenly find itself appearing daily on the likes of the "Jerry Springer Show" and lots of daytime/weekend cable, but little else. But this isn't going to happen----fortunately.
The bigger issue which we have identified in this dialog, is whether CPM-driven corporate buying is overly concentrated on too few channels and especially on shows which low brow couch potatoes favor to a greater than typical degree. As I mentioned, I suspect that some of the major buys are so CPM focused---due to client bean counter insistence----that they are not providing enough reach generating variability so when the spoils are divvied up and allocated to the brands, they are not getting what their plans called for. I will try to investigate this, however, as the truth---whatever it is---may be deemed harmful to the agencies, I may have a tough time getting straight answers. Of course, any client who cares enough to check could make a specific determination for its brands.
Very true Ed. I did take a more Aussie perspective.
We have progranmmatic in our Foxtel Subscription TV service (around a 20% share of viewing), and it is being taken up in part with the FTAs. Basically, we negotiate the core programmes (target audience, content quality and fit, sponsorship and cross-marketing opportunies - oh, and cost of course) and then 'backfill' on a target-audience CPM basis. When the daily ratings come out we can then monitor campaign performance and reach and adjust if need be.
Jack, you are right on target (as I would expect) and perfectly capture the issue that Ed and I are advocating from different sides of the coin. If the industry would focus away from bulk GRP CPM to cost-per-reach point against tighter TRP's the TV ad industry could not only better compete with digital, it could also recapture some below the line money.