This is from my corporate days
Polaroid has such a rich and interesting history. It started in the 30’s, becoming one of the biggest customer products giants, before going bankrupt throughout 2000’s. Fortunately some really cool people resurrected it, and eventually these same people hired me some years later!
When I joined, Polaroid was doing very well → sales were growing, new cameras just launched and it was a great time of growth. However, one thing was missing - concrete insights. While there were some loose competitive reports and plenty of financial figures, it missed one thing - a holistic definition and understanding of the audience.
That’s where I come in. This wasn’t my first project - it took nearly 2 years to get buy-in for a sizable investments into a foundational audience definition. But when I got a green light, it ended up being one of the largest and most fulfilling projects in my career! And it made me a little locally famous to - about once a quarter I run into someone who knows me because of this piece of work despite getting hired way after I left!
With a company restructuring to put product at the fore-front AND rethinking marketing strategy, a huge need emerged to fully understand both current and ideal audience. It’s really hard to make concrete decisions about the future without having a crystal clear vision of for whom we’re creating that future.
And while there were many strong ideas about the audience definition, there was little to no data about them.
This is where we proposed a large scale, 360 mixed methods research project to produce one source-of-truth report. And in 3 months, we’ve done just that - at 10% of the cost of a comparable agency project.
How do you define an audience for a well-established, major brand without making it vague? We did 4 concrete things
Because of these 4 steps and lots of testing, we were no longer running blind. While I can’t share the finer details, we ended up with a 3 question screener that was a very strong predictor of brand-love for Polaroid, as well as long-term intention to buy. This was such a fundamentally necessary base to have for this project, as it allowed us to measure “Polaroid attunment” and correlate with other variables.
Speaking of those other variables, based on both internal and external workshops, we chose 10 key areas to focus on - from hobbies and creativity, to device usage to activism and IRL events. We run multiple short tests for each scale to make sure it was internally valid, balanced and easy to understand.
With all the groundwork, we’ve got to work on the actual research.
Launching surveys in 2020’s is actually extremely easy - there are dozens of amazing sampling agencies out there. All you have to do is ensure some basic systems to detect completions, deter bots, uncover trolls and stratify the sample among key segments to follow demographic distributions. We’ve set for sampling a total of 3000 participants from 2 various sources - mostly to see which effects might be due to sample quality and not actual population behaviors. It was just big enough to detect small clusters or unique user groups, while staying budget-conscious.
One week later we had our responses and it was time for me to sit down and spend next couple days glued to SPSS, NLP software, and excel.
Alongside the surveys, we’ve also launched 10 high-quality interviews with people who were a 100% fit for our target audience. While these were not really critical for findings - we’ve run way more interviews pre-survey and had a strong qualitative knowledge base already stablished - they gave us some wonderful quotes and faces to use. This made the findings much more digestible and believable towards senior executives and ultimate made them 10x more impactful and successful.
Analyzing wide, 100+ column data sets can be a real pain. While I had plenty of concrete research questions, a big part of this research was also exploration - and so I’ve run well over 300 statistical tests. Sure, some of them are bound to be false-positives, even at 99% significance. But with the research designed we’ve opted in - combining interviews with multi-wave research and blending open-ended questions with closed-ended scales, hypothesis testing was not the only way to go.
I like to think about data analysis like a detective case. You’re collecting clues/proof of varying strength and quality. Yeah, that 99.99% sig test is REALLY impactful (unless the question design was bad). But a few questions pointing to the effect at 90% sig + a lot of clues in open-ended questions + majority of interviewees confirming the effect is also something worth considering in the report. Statistically it’s as unlikely to have a 99% sig false positives as A LOT of overlapping weaker false-positive proofs.
With this approach, we were able to find many more qualitative trends that could be levered by the brilliant minds across product, marketing and sales. It wasn’t just about rigorously testing each variable for an academic paper - it was about sifting through the numbers to uncover most likely and plausible stories, verify them with statistics and simplify into a few strong words.
Another key part of analysis was processing open-ended answers. I really wish I had had chat-GPT back then - but alas, I had to do with NLP tools like Orange, relying on word clustering, word clouds and basic linguistic analytics. I’ve opted in to have a good balance between open-ended and multiple choice questions - primarily as open-ended questions can reveal much deeper and more robust insights, at the cost of being harder to quantify and analyze.
Once I had the preliminary findings, I’ve worked with a designer and a few select key stakeholders to turn them into a consumable report. Picking the findings to include was probably the hardest part of this research. At the end, we set on grouping findings into “themes” - different aspects of our audience’s lives. Somehow we’ve turned 200+ pages of notes into a 30 slide deck. (and then into a 60 page deck once we’ve included all the brand, explainers, personas, strategies and openings).
