What Do Consumers Really See When They Look at Your Brand? One Company Found the Answer in 7 Million Documents a Day

A while back, I came across a number that genuinely stunned me.

Seven million documents. Every single day.

Not a year. Not a month. A day.

The company processing all this data is called Launchmetrics. They specialize in marketing analytics for fashion, beauty, and lifestyle brands. Their job boils down to one thing: figuring out how consumers actually see your brand.

Just think about how hard that is.

The Brand You Think You Are vs. The Brand Consumers See

What exactly is brand perception?

Let me give you an analogy. You make running shoes. You believe your biggest selling point is the technology — the cushioning system. You pour tens of millions into ad spend telling that story.

But what do consumers say about you online?

They talk about the color of the shoelaces.

They talk about whether the soles wore flat after three months.

They talk about how fast your customer service processes returns.

You think you're selling technology. Consumers think you're selling looks.

That gap — between brand intent and consumer perception — is where fortunes are lost. For a brand pulling in billions in annual revenue, that gap can mean tens of millions in marketing budget going straight down the drain.

What Launchmetrics does is measure that gap.

Measuring With Data

How did brand perception analysis used to work? Surveys. Focus groups. Manually reading comments. Small sample sizes, slow turnaround — by the time your report was finished, public opinion had already shifted.

Launchmetrics takes a different approach. They scrape every online conversation about a given brand and pipe it all into an Amazon S3 data lake. At the scale of seven million documents per day. Then they run machine learning to identify keywords — to see what people are actually talking about.

But keywords alone aren't enough.

Say you notice everyone's talking about "quality." Is that good quality or bad? How much does it relate to your brand image? How do you stack up against competitors? These deeper questions — traditional machine learning models simply can't answer them.

Keywords get you in the door. Understanding is the real answer.

Large Language Models Filled the Most Critical Gap

When generative AI came online, Launchmetrics connected to Amazon Bedrock at the first opportunity. Their CTO, Pau Montero Parés, said something that really stuck with me.

He said that concepts they could never get their hands on before — now they could actually analyze them.

What kind of concepts?

Things like "femininity."

Like "creativity."

Like "masculinity."

How do you even define these words? How do you teach an algorithm to understand what "creative" means? You could write a hundred lines of if-else statements and still never get there. But large language models just inherently get it. They've been trained on massive amounts of text. They know that when people call a brand "creative," the surrounding context tends to contain certain types of expressions.

Launchmetrics built a two-step pipeline: first, machine learning extracts keywords and trends from tens of millions of text passages. Then, large language models correlate those trends with various brand dimensions and assign scores.

Finally, the LLM translates the analysis into language that marketers can actually understand.

You don't need to know data science. You don't need to write SQL. What you get is plain English: "This week, consumers felt your brand improved on these dimensions and slipped on those ones. Here's why."

From Five Months to a Few Weeks

And the results?

Prototype development cycles compressed from a maximum of five months down to a few weeks.

The difference between five months and a few weeks is the difference between surviving and getting wiped out.

But something else caught my eye even more.

You know what the hardest thing is for a new brand? It's not lack of ad budget. It's lack of data.

A designer brand that's only two years old might have just a few thousand online conversations about it. Try training a model on that — the sample size is nowhere near enough. The conclusions you'd draw would be barely better than a coin flip.

Launchmetrics used generative AI to do something quite clever: generate synthetic datasets for brands that don't have enough data.

What does this mean?

It means the number of brands they can serve expanded from the hundreds to the thousands. Those brands that were previously too small, with too little data — they can now access the same brand perception analytics as the big players.

Parés said something to the effect of: this is a massive shift.

I think he was being conservative. This isn't a shift. It's the barrier collapsing entirely.

One Last Thing

After reading this story, you might think it's just a tech case study.

But I see something else.

What is a brand? A brand is the collective impression a group of people holds about you. That impression is changing every second, across billions of phone screens.

Before, you could only guess by instinct. Now, someone is reading seven million documents a day, using large language models to translate the inner thoughts of consumers into a language you can actually understand.

Ten years ago, this would have been unthinkable.

So every time someone asks me what practical use generative AI actually has, I want to tell them this story. Nothing sci-fi about it, nothing world-changing in the dramatic sense. One company. One new tool. One thing that couldn't be done before — now done.

And done beautifully.