Discover how MarketingIntelligence.io drives cutting-edge marketing analytics to help their clients scale and grow
MarketingIntelligence.io empowers companies to leverage their in-house data with advanced data-driven insights to transform and optimize marketing efforts towards scalable growth.
We talked to Marek, a battle-proven data scientist and co-founder of MarketingIntelligence.io.
Marek started his career as a solo data scientist in a marketing team, grew into the Head of Marketing, and decided to marry the two fields into the successful company we nowadays know as MarketingIntelligence.io.
We hopped on a call with Marek, who has used Keboola since its first days, to get a first-person account of how Marketing Intelligence provides value to its clients and how Keboola fits into the picture.
MarketingIntelligence.io enables companies - from hyper-growth startups to large global enterprises - to understand their marketing efforts better. It provides strategic insights into how to successfully grow marketing.
In a post-cookie world, privacy-first laws and the dispersion of data across multiple marketing platforms impede companies from fully understanding their marketing.
MarketingIntelligence.io navigates these treacherous waters to help their clients understand, prove, and scale the value marketing provides to their companies.
They use data and technology to:
But the road to marketing insights excellence was paved with unique data challenges.
When Marek first started as a data scientist, there was no Keboola to help extract data needed to perform advanced analytics.
The community referred to that time as “Excel hell”:
“If I wanted some cost data other than from Facebook or Google ads, I had to ask someone, and what I got was an Excel file. The data was aggregated, either by months or something else, and I had to make it work by breaking it down to the session level and things like that. It was very difficult. And there were no versions, it was completely unsustainable.”
Sending the Excel file back-and-forth with all the breaking changes was painful enough that Marek and his team developed custom extraction scripts in Google Sheets.
But even those were a maintenance nightmare. Scripts kept on breaking or not running at all, and precious time was spent debugging the ETL pipeline.
Until they migrated their data pipelines to Keboola:
“When you’re running enterprise ETL solutions such as Keboola, you only look into one place to understand what is happening. And you can find the problem there. So debugging and solving these issues is much, much easier. One of the first things I realized when we adopted Keboola was that I no longer have to use five different platforms to do one thing. I simply had to get acquainted with Keboola and everything was in one place.”
We asked Marek what the first impressions with Keboola were and how Keboola compared to other platforms on the market.
“My first impression was that I was really happy that everything was in one place. I was lucky enough that the people in the Business Intelligence department taught me how to use Keboola and to work with it.”
That was in the time when other competitive platforms started to offer similar solutions to drive the entire ETL data pipelines.
But the competition did not speak the language of the data scientists. As Marek put it:
“The problem was that you had to code in Java to achieve some things and I don't know Java. I used R from my first day as a data scientist, when my boss, our current CEO, told me that ‘you should learn this because this is what statisticians use. And this is maybe one of the biggest reasons why I've been faithful and trustful to Keboola over the years. Keboola uses the R backend. There is a way to do R transformations. There is one package - data.table - in R which is simply the fastest in the world to manipulate huge datasets.”
The main advantage with Keboola, Marek points out, is that Keboola allows you to run your R transformations in the cloud:
“I have an 18 core processor at home. But I have to run a virtual machine with Linux to get to utilize these multi-core capabilities to their fullest. That’s not the case with the Keboola cloud and our transformations. In Keboola, if there is a four-core or eight-core virtual environment, we can use all those cores, which is great. I can use what works at home and just run it in the cloud.”
So how is Keboola helping Marek and his team drive advanced data analytics today?
When MarketingIntelligence.io develops data science models in the backend and visualizations in Tableau, they start with smaller use-cases.
As they prove the value of their insights for marketing, their clients add new requests. From one customer lifetime feature to 10. From one website being analyzed to 30. Expanding on brands, channels, regions, and everything in between.
“The problem is that in the past, this would mean changing every single code for every domain and every country by hand. Yeah, it's most often copy-paste when it's working, but still very painful even for one client running tens of domains.”
But the manual work is tedious, prone to human errors, and does not scale well. Keboola addressed this common data science pain point with shared code.
Shared code is a feature that allows you to reuse the same transformation logic across multiple implementations.
“The shared codes feature helps us include more and more countries and domains for the same client. It is much easier to maintain, debug and implement new features, simply thanks to the shared codes. So there is literally one specific shared code right now that is used in 50 or more transformations with slightly different variables as parameters.”
Brands often spend a considerable amount of their marketing budgets on display and brand advertising to increase brand recognition and drive purchases down the line.
But the reality is, it is hard to evaluate the value of this wide-targeting marketing approach. With companies on long sales cycles and multiple touchpoints to prove the product value, it is even harder.
Did the customer convert because they saw an ad a month ago, or did the ad with the discount push them towards a same-day impulse buy?
Advertising platforms often make this attribution harder by not divulging customer journeys from first-click onwards. Last-click attribution is the default, crediting the last customer touchpoint for all the marketing heavy-lifting.
Marek and his team helped multiple clients to stitch their data from various sources and build coherent customer journey paths.
“For one of our clients who sells a CRM solution, we stitched all the data in Keboola from the trials, from converting to paid version, from website visits, the cost data from all the advertising platforms … We were able to reconstruct all these paths from the beginning. And we showed the client that in some cases, there were 150 website visits (paid and unpaid) before converting from the trial to the paid tier during those three to six months.”
Before understanding how the different customer paths lead from the first visit to final conversions, their client
“... had no idea if it's too much they're spending, if it's too little, if they could double, or even triple their advertising budgets, maybe cut them in half. It was like throwing darts on a target, but with your eyes closed.”
With careful analysis of how different channels and pathways interact with each other and what customer segment tends to convert, Marketing Intelligence helped save their client 30% on marketing costs while increasing acquisition by informing them what conversion paths work best.
MarketingIntelligence.io recently tackled a hard problem - how to establish the customer lifetime value for a massive eCommerce client.
The competition for the share of digital shoppers skyrocketed in the last two years. eCommerce brands are trying to beat their rivals by acquiring customers before the competition does.
But with multiple touchpoints (and dispersed data), buy-and-never-return customers, and irregular repurchasing behavior, it is hard for online shops to determine how much they can spend on acquiring a new customer. Especially if they are willing to spend some more to get a new customer and then wait on the long-term profits trickling in with repurchases.
This is where Marek and his team came to the rescue. By building a segmented customer lifetime model, MarketingIntelligence.io was able to showcase the type of customers their eCommerce client could spend more on acquiring while still turning a profit:
“The hard numbers show that some customers are up to four to five times more valuable than other customers, even among those very active ones. We proved that they [the eCommerce client] can triple or even quintuple the customer acquisition cost in the right segments, because the return on investment is there.”
By using Keboola, Marek and his team were able to stitch data across different data sources much faster and build a customer lifetime model in just 3 weeks instead of months.
Keboola helped MarketingIntelligence.io bring all the data together from disparate sources faster, more transparently, and with fewer errors. Scaling their ETL pipelines in Keboola delivered the data in record times. Accelerating the time MarketingIntelligence.io delivered value to their clients.
It also empowered them to do their best analytical work in R, the language of data science, by liberating them from spending resources on data operations.
As Marek said:
“If it wasn't for Keboola, I would be half the data scientist I am.”