How We Built a Scalable Analytics System for Mentorshow and Cut Costs by 2.25x
6 minutes
151
Getting a real analytics system in place, with interactive dashboards and a proper database instead of messy Google Sheets, feels like putting on glasses for the first time. Suddenly everything comes into focus. You notice every detail and wonder how you ever managed without it.
But sometimes the magic just doesn’t happen. The numbers stay blurry. And instead of clarity, analytics becomes just another headache.
Across 40+ projects, we’ve seen it all. In one case, a client tried to build their own analytics system. But instead of insights, they ended up with clunky dashboards and messy, unreliable data. Things got so bad that they had to recheck KPIs in Excel, manually, and pay extra for it.
Spoiler: the story ends well. We cleaned up the chaos and built a system they could finally trust. Here’s what happened, and what you can learn from it.
When One Database Isn’t Enough
Our client, Mentorshow, is a French edtech platform similar to MasterClass. They offer video courses from experts across various industries. By the time they came to us, they were a mature company with a fairly developed analytics stack.
Their request was clear. Move everything to a new database, rebuild several core dashboards, and make sure the numbers were accurate. They had serious doubts. So serious, in fact, that they were manually recalculating performance metrics in Excel every month. That alone cost them hundreds of euros.
We started digging into the setup. What did we find?
The usual suspects. Data from social media, their website, ad platforms, student activity, and financial operations. Data collection was automated using a few custom Python scripts. Data was stored in not one but two databases—Postgres and Redshift. That’s where things got messy.
Visualization was handled in Tableau. Seven dashboards covered the essentials: revenue, ad spend, sales, user activity, and so on. Tableau and Redshift are both paid tools. Mentorshow was spending money on them, yet still not trusting the data. Worse, they were paying again to recalculate it.
They had multiple data sources all feeding into two databases, often duplicating or overwriting each other. The dashboards were inconsistent, and it was hard to tell what data was accurate or whether the metrics were being calculated properly.
It was like getting different reports in Slack and by email, sometimes with conflicting numbers and sometimes in Wingdings. How are you supposed to find insights in that?
The original architect of the system had already left the company. There was no one to ask why things were built this way, and no documentation either.
So how did it get to this point?
We think the analytics just couldn’t keep up with the company’s growth. In the early days, a couple of scripts were enough. But as they grew—adding courses, instructors, marketing channels, a sales team, a website, an app, payments—the system didn’t scale with them. New scripts were added on top of old ones. Things got tangled.
The dual database setup may have started as a migration plan. But something went wrong. Without proper expertise, they couldn’t build a scalable system.
Eventually it all became too complicated to fix. Starting from scratch was easier.
How We Cut Their Analytics Costs in Half (Actually, by 2.25x)
Mentorshow was paying for Redshift and Tableau but not getting value from either. Scaling or improving the system was out of the question.
We proposed a few key changes.
First, clean up the data. We migrated everything to a single database to eliminate duplicates and errors. They wanted to switch to ClickHouse, which is one of our specialties. That’s one of the reasons they reached out to us.
Second, make the system more adaptable. We implemented Apache Airflow as the orchestrator. It collects, processes, and monitors data centrally, and sends alerts if something breaks. It is much easier to manage than a bunch of scattered scripts.
If you’re not sure what a data orchestrator does, think of it like an executive assistant for your data. It gathers reports from different departments, organizes them, and sends a clean summary to the boss.
But back to Mentorshow.
It took us about four months to rebuild the system and migrate the data. But it was worth it.
Mentorshow cut their analytics costs by 2.25x. We did the math.
And now, those costs actually pay off. The client trusts their data. They use the dashboards. Analytics became what it’s supposed to be: a tool that helps the business grow.
We also built seven new dashboards in Superset, an open-source tool. That alone made future updates cheaper than continuing with Tableau.
This project showed, once again, that well-designed analytics is an investment. Not a cost center. Done right, it pays off. It scales with your business.
Why They Ended Up With a Frankenstein Analytics Monster
It’s easy to wonder how a setup like this even happens. Good tools. Automation. But in the end, the numbers were still off and they went back to Excel.
Here’s the reality. Mentorshow grew from a startup into a solid mid-sized business. But their analytics stayed stuck at the startup level.
That happens more often than you think. Sometimes teams don’t notice the tipping point when the old setup stops working. They’re too deep in the day-to-day to zoom out. They don’t have the expertise to assess the system objectively.
So how do you avoid this?
It depends on your stage.
- If you’re a young startup, build the foundation early. Set data standards and policies. Decide where data lives, how metrics are calculated, who has access. If you have a tech person, make them document everything. It will save your future team a lot of pain.
- If you’re a scale-up or mid-sized company, don’t reinvent the wheel. You need expertise to build something solid. That means either hiring a full data team or bringing in an expert partner.
Building a team takes time and money. You’ll need to manage it closely. A consultancy brings ready-made experience and tools. It gives you a predictable outcome, on time and on budget.
