[cm2012]

With luxury goods, the brand is the product and the value is from the brand, so you can't easily iterate with one off sales.

If I'm selling a new type of toothbrush, or a software that helps teachers grade homework with AI, I can make some ads, see how close they get to the needed ROI and tweak it. I can test marketing different aspects or selling points of the product and getting results iteratively.

If I'm selling a new luxury handbag, I will never, ever sell it at a reasonable ROI from just a set of ads. You have to build the whole eco system at once (kind of a chicken and egg problem). People have to see influencers using it, celebrities having them, the right kind of feeling in the ads, the right news articles, being sold in the right stores, etc. My first marketing job was at a start-up denim brand trying to become big and I observed and researched a lot there about how it works.

[shostack]

As a marketer who came up on the performance side, I feel your pain.

What would you say the state of analytics is on that side of things? Is attribution sufficiently advanced to get some read on the impact of various influences, celeb and PR hits, product placement, etc?

Even as a senior marketer confident in my skills, so much of the luxury space seems very much like an exercise of needing to put all your eggs in the one basket of a big launch. None of the steady burn of some more performance-driven plays with the usual iteration on ads and funnel metrics and such.

[cosmie]

Attribution is as sufficiently advanced as needed to get approval to give BBDO another blank check for another hairbrained scheme that'll fund yet another award for BBDO. While concurrently being undermined and considered not reliable or advanced enough if the numbers it provides do not lead to another blank check for the creative agency.

I come from the performance side, but currently work with a bunch of CPG clients with products that run the gamut between commodity to luxury.

I specifically work on creating analytics and attribution frameworks, because these companies are fine with fuzzy hand-wavy "lift studies" for tv commercials and stupid in store display stunts. But they hold a double standard and anything digital has to be concretely measured to defend its budget.

It's actually pretty easy to create robust analytics and attribution in the space. But it's mainly a process thing, to be able to sprinkle around enough unique traits or identifiers along the way to measure at an aggregate level what the impact was. It tends to rarely be done though, due to a lack of that level of operational discipline for brand marketers and agencies, or due to the desire to deliberately sabatage the numbers because they don't paint a particularly flattering picture. So more often than not you end up with a botched execution on the small details that were required for proper attribution, then the resulting numbers being full of enough holes to spin the data however is convenient. Or someone slapping on some poorly integrated software that spits out a number that's taken as the holy grail, "cuz AI said so".

... which leads to a terrible cycle of distrust in analytics and attribution on the brand side, leading to fewer initiatives that prioritize it.

[shostack]

Interesting. We share a lot of frustrations and challenges.

What sort of spend levels and data volumes do you typically need to see for the lift studies you do with TV and CTV? Do you typically isolate to specific markets for that?

[cosmie]

It really just depends. In the case of one retail client, we have carte blanche access to all of their data, from marketing systems to POS data to app location data. It makes it incredibly easy to "lazy load" a lift study after the fact, by looking for anomalies in behavior that are correlated with the creative. Rather than a standard test and control, we can essentially tailor the model to a per-store or per-region level and rollup lift from there. It's less about the spend level and data volume, and more about the data completeness.

For CPG clients, it's more of a pain. Those usually involve really complex interagency relationships, with discontinuity in both processes and data access. And in a lot of cases, they may have access to a retail partners POS or loyalty data, but can't share it directly with us as a third party agency, and there's a game of telephone where we have to coach them on what to ask for and provide to us (in whatever form they're allowed), while being blind to the data and data/system structures. So a lot gets lost in translation, with the spend level having to be large enough to compensate for however dysfunction that process is for that program and client.

That said, I'm lucky enough to be a passive observe to that most of the time. Another manager under my boss is responsible for those more traditional lift studies. I have an unusual background in that I've done a lot of process development work, web analytics, and data engineering/management. So I'm only brought into those projects when we have more technically sophisticated needs.

[cm2012]

Since my first job, I've also tried to be in mostly performance roles, so I can't say for sure what the state of the art is in the brand side. I know there's lots of software trying to solve those issues.

[CogitoCogito]

This sounds similar to lamenting the difficulty of systems optimizations versus micro optimizations. Similar frustrations exist everywhere I guess. Interesting to see similar issues in the luxury/non-luxury goods space.

[alexpetralia]

Hey this was a really interesting post and I'm wondering if you'd have any ideas about a project I'm working on related to brand marketing. I couldn't find your email but I saw you live around NYC (as do I) - maybe we could connect? I'm at [e-mail].

[ahartmetz]

Grading homework with AI is an idiotic and destructive idea, just saying. "AI" may be able to see superficial similarities between the work to grade and a well-written text, but it won't (in its current state) be able to distinguish gibberish from a good argument. And what does it feel like to pour your heart into something, then have it graded by a machine that doesn't understand because your teacher can't be bothered to even give it a good read?

[humanrebar]

Most people don't distinguish between Artificial Intelligence and Augmented Intelligence.

Grading fill in the blank arithmetic exercises seems like a perfect piece of Augmented Intelligence. No gibberish problems until you get to "show your work", and even there, there are options if the business model is there. If the program is able to mark answers as "illegible", the grader can review before confirming. Make take-home grading something that can be done before the children go home for the day.

[ahartmetz]

But still, apart from really mechanical yes/no aspects like spelling, I wouldn't want to have anything I write graded by something with which I can't have a normal conversation (a human or strong AI). Would you?

[a_t48]

If you disagree with the grade, you would still have the option of taking it up with the teacher/TA running the grading program.

[justin--sane]

Really? That's what you get from this post?

[ahartmetz]

When somebody lists something awful as a positive example, it does disturb me and I don't want to let it stand. Is that hard do understand?

[barbecue_sauce]

The commenter did not list it as a positive example, just as an example.