MeasureCamp 2024 Prague Report

The key takeaways from several presentations that we joined.

Last weekend we attended Measurecamp in Prague. For those unfamiliar with the event, MeasureCamp is an unconference focused on digital analytics, open to not just analysts but anyone interested in data, marketing, or tech. The agenda is set by the attendees on the day, with sessions covering the latest trends and developments in the analytics field.

AI for (coding) productivity (by Daniel Dominko)

One of the central topics was artificial intelligence (so unexpected 🤯). The speaker emphasized the importance of foundational knowledge and continuous learning (= keeping up with evolving technologies are key).

The speaker went from the basics related to AI (the process of tokenization and embeddings) and RAG (Retrieval-augmented generation) which works by first retrieving relevant documents or data from a large corpus using a retrieval model. Then, a generation model (like a language model) uses the retrieved information to generate more accurate and contextually relevant responses. This approach enhances the model’s ability to provide better answers by grounding its outputs in real-time or stored data.

The introduction was followed by the practical use of AI in individual areas such as coding (refactoring, testing, performance prompts), brainstorming (ideation, risk analysis, use cases, design verification prompts).

Personally we liked the thought that if something lasts for 5 minutes to do it, you don't need to prompt it for the same 5 minutes. We should still need to use our brain. In case that you don't want to think - maybe it would be helpful to check for example Promptbase for some prompt inspiration/database.

And if you want to go through some level up stuff, take a look at the tools/areas the speaker mentioned.

Analyst turned developer (by Michal Blazek)

Session explained on real use cases how to implement some measurement in case that you don't have a developer involved in the project, your developer does not have capacity or he/she only doesn't want to collaborate with you. 

So what can you manage in GTM so you don't need a developer:

• measuring clicking on filters, labels, actions connected to product (wishlist, comparison, favourite)

• A/B tests (eg. hidden or visible pricelist)

• Events related to the forms (and capturing the email, phone from the form)

• scroll measurement according to page sections (not only percents)

• And many more...slides you can find here. 

Whether you put this or other specialties in GTM, always test it and make documentation.

Scalable data warehouse (by Jakub Kříž)

This session took twice as long as a normal session, but it was definitely effectively used time. Jakub builded data warehouse from scratch on graph in this hour and while that he showed how it's it builded/connected and which components it needs to have and also from which tools should be builded in Google Cloud Platform.

First important point there - at the beginning of every project, you need to ask yourself basic questions:

• Do we understand data needs? 

• Do we have a business case? 

• It is really crucial to know what data is needed and why. Only by asking this questions at the beginning, it is possible to avoid collecting unnecessary data and prevent wasting resources.

• It is not (cost) effective to measure everything just in case, if we know that we won't use it. You can store all the raw data (such as a large dump database) somewhere, but it is important to have a second layer where you work only with the data you have selected and you will need them.

Maybe you listen first about Data Warehouse – technically it is a place somewhere in the cloud where your company has stored and processed data. So it connects all your GA / GSC / internal data sources and makes advanced graphs – it is not for everyone but if your company is data-dependent, so that's something you already need…

Quite simple data warehouse contains only basic layers of data

…we started “from end” so firstly we connected “data-sources” with “presenting layer” after that we added middle “bronze - silver - gold” layer that cleans and preparing data and at the end we make a lot of other “tuning” things that expand schema to “data warehouse”.

One of the key successful data warehouses are those “middle layers” that transform data. You start with a “tidy” bronze layer in which are just downloaded data from that sources and in the gold layer you have calculated outputs as you need them in reporting. So the rest of your company could work with a “trusted” gold dataset in which you have right data calculated and curated by analytics.

Final architecture

Also Jakub showed  a little about the security of that data warehouse. Like every third-party tool need to have own import into that data warehouse because it will have maybe an external person who takes care of it and to whom you don't want to show your "golden" datasets. Similarly, developers who are dumping some internal data, you might not want to show them everything.

BIG last point there: it's not about having the architecture like this from the beginning, but about having it prepared like this and gradually expanding it and simply counting with it that the final architecture would be like the one the speaker shared with us. 

Money in GSC and how to get it (by Martin Žatkovič)

Martin gave a quick talk about Google Search Console and its data, showing that in connection with BigQuery, there is a lot more data available than through the API – so it is important to connect it – but you know about that, right?

Also he showed us how to “win over” sampling in Google Search Console when you have a large website like Zbozi.cz or Mapy.cz and create multiple properties for each directory of that website – if you have great URL structure. 😅

Also he recommended to use DataForm in BigQuery and connect data with custom imported Google Sheets where you should have stored labels / pagetypes and you should have much more detailed graphs that aren't available in GSC now because they will be clustered.

And also he showed that GPT-4o has great performance of tasks to build queries to BigQuery and based screens it could connect two tables by inner join and deliver data that you want to use. 🙂

And where are that money in GSC? You start getting that money from GSC when you start working with that data and using it in your strategy. 

Cursor AI (by Peter Šutarík)

Cursor is an AI-powered code editor that enhances developer productivity with features like code autocompletion, error detection, and real-time suggestions. The speaker showed us how this tool helped him to develop a simple extension “CSS selector viewer” and cool app from scratch just only thanks to intelligent recommendations and real-time insights which the tool provided him.

• It is definitely a tool which is worthy to try and see how it can help you, but be careful what data you send to it.

Leverage of Google Ads export to BigQuery (by Anna Horáková and Vašek Jelen)

Main aim of this session was to use Google BigQuery to create custom analyses and reports from Google Ads data because of overcoming limitations in UI and of course because of the need for advanced data manipulation, merging, and querying (BQ we ♥️ you). In GA4 reports you have aggregated and modeled data (not set issues).

Speaker presented a really straightforward comparison between using GA4 Interface/API and using GA4 + Ads data in BigQuery for a specific campaign. You can see notable differences in the number of visits recorded. We should mention Google Click Identifier (GCLID) - for sessions delivered using Google Ads a GCLID, you can assign the source/medium/campaign.

The export Google Ads to BQ is created as a data transfer. The dataset itself has about 200 tables and views, which need to be linked together in order to have the relevant data (I can NOT find the campaign name and ID in one table (using the pairing key campaign id join)

Link to query which simplifies Ads export and create one base (L1) table which you can use for following operations: https://github.com/MeasureDesign/Public_repository/tree/main 

There was also a preview of an upcoming feature in BQ that will simplify the integration and querying process. It is called “Custom Query” and the release date will be probably at the end of Q4 2024 or Q1 2025. 

Slides you can find here. 

Data-visual minimalism (by Dominik Jirotka)

The session was focused on minimalist data visualisation, which is crucial for effective client communication.

• We should not complicate graphs and reports - let's use basics like pie charts, line charts, bar charts and for some nice visualisation for example sparklines.

• Definitely don't use any 3D models, abstract shapes, shadows.

• It is always good to stay consistent with the colour scheme. Best practice can be limiting the use of colours to three brand-specific shades/colours to maintain a clean and professional look.

Analytics Therapy

Program-filled day ended with the session, which was an open discussion about what exhausts us the most in the field of analytics and work with users/customers. You all probably know…

The last part of the program was of course the afterparty - yay 🥳!

Are any of you going to MeasureCamp in Bratislava?