top of page
Writer's pictureOmer Chen

How does a data-driven company improve a qualitative objective using AI?

Updated: Aug 17, 2023

How does a data-driven company choose to pivot its core-product KPI and improve a qualitative objective? 🤔



👋 Context

As you might know, the last months were dedicated to putting the content in the center. We’d like to tell you more about the story behind the strategic pivot, work process, and results.



🔙 The pivot

6 months ago, Inplace was already well known as a highly effective exposure to dollar channel for influencers’ campaigns, and that was one of our initial product KPIs since we started.


But as a customer-centric company, we’re always obsessed with finding new ways to expand our value and offering to every customer — and that’s why we started a discovery process to understand the next “big thing” for us.


One of the unique components of influencer marketing, in differ to well-known marketing channels, is the ability to gain not only awareness and traction but also reusable, creative content which can be recycled into the brand’s other channels as well (Instagram, TikTok, YouTube, etc..).


That was the moment we realized that the content isn’t just a “nice-to-have” feature, but is at the core of the customer value.



👩‍💻 The process

The first step in any process of product improvement is to define our product KPI (if you can’t measure it, you can’t improve it), but the issue is that content is a “qualitative” value — and here was our first challenge, how to measure a qualitative improvement as a data-oriented company?


That’s when the AI entered the picture. We started by analyzing first whether or not there’s a mutual objective agreement about which content is good and which one isn’t, by rating our own content — it wasn’t a CSM task or R&D task — but a whole company mission. Most of our talents, live and breathe content from different perspectives as an Instagram user, as an Inplace contributors individuals, and also thanks to their individual and professional perspectives from their scope of work in the company. After achieving this, we found the first essential and nontrivial insight: Most of the time, there’s a mutual agreement about which content is better.


Then we started to analyze and look for patterns using any resource we could bring to the table — from big data we collected about +100K pieces of content made using Inplace, up to manual data tagging we did ourselves.


After building some hypotheses about shared patterns among great content, we started testing it using statistical models and machine learning techniques. Then, we built our first MVP model to predict content quality (and it’s already delivering very well, but we will get to that soon 🤭).


The thing is that a content quality formula is just not enough — some of the main features in the models were tagged manually on our data set and weren’t viable for our current scale… 😪


That’s why we had to find out an automated way to tag data like — edited videos, product appearance, and voiceover with AI-driven tools — the results were unbelievably good — and now, we have a valid, ready-to-scale KPI and we can start to improve it 📈.


The new AI model not only helped us to track our progress but also provided us with useful insights about the things that impact content quality the most. It started by optimizing campaign briefs with a focus on the most important stuff, and then we launched the “Inplace Premium 💎” feature which grants exclusive access to our top content creators for high-demand offers in the app. Finally, we were transparent about the KPI to the influencers themselves. Each influencer now has access to their own content quality score, along with some insights about how to improve their class and become premium members.




📈 The results

Today, after a 6-months product journey shaping and optimizing a new KPI that looks non-feasible to optimize at the start of the road, we’re proud to introduce the first-ever AI model to analyze influencer campaigns’ content quality with not only a zero-touch classification but also actionable insights about how to improve each piece of content to perfection.


Our content KPI was improved by 43% which also led to a dramatic improvement in our customers’ collaboration rating on their business app, as well as, yes, day-to-day qualitative feedback from customers that just felt that something good just happened to the content.


4,013 views1 comment

1 Comment


Guest
Aug 17, 2023

Wow!! amazing 🙌

Like
bottom of page