MeasureCamp London 2024 Takeaways

Last weekend, I had the pleasure of attending MeasureCamp London again. It's good that I am wearing the event's t-shirt as I am typing this, so I only have to look at it to remember it was the 17th time. The event started with pre-event drinks on Friday and continued at the hotel's rooftop bar, where many others and I were also staying, with the London Shard in the background. The next day was the main event, with 351 attendees. When the London MeasureCamp moved from Pimlico to Fenchurch after the pandemic, the new venue felt jam-packed, and many would have felt the event overwhelming. It can only be with a fantastic team of volunteers and venue staff that the venue was less busy despite having about 50 more than last year, but close to 100 more attendees since the venue change. Thanks Keely Jacob , Anna Lewis , ?? Charles Meaden , Guillaume Lombard , Nicolas Malo , Andrew . and the other volunteers, whom I shamefully forgot. Thank you to the sponsors, without whom the event would not have existed. Without further ado, here are my extensive notes and comments on the sessions I have attended.

The first time slot on the session board is for the sponsors. I attended Adobe's session on Customer Journey Analytics (CJA) as we are adopting this at my employer, Legal & General. I attended two more Adobe sessions on CJA. Therefore, I am starting my takeaway notes with a medley of these three sessions.

Customer Intelligence Redefined: Breaking Down Barriers in the Omnichannel Odyssey by Alexandru Ivana, Product Manager, Customer Journey Analytics, Adobe

Graph it right: unlock customer secrets by Mansi Sharma, Adobe

AA -> CJA 10 things to know by Max Lagacé, Betway Group

CJA centralises all the data into a single place. The benefits are the elimination of redundancy and the risks that multiple data points for the same thing go out of sync, leading to confusion about which one is the correct and latest value. CJA is a significant shift. Instead of collecting data points, aggregating them, and extracting some general trends, CJA focuses on individuals. A core feature of CJA is the so-called identity graph. Adobe calls this old chestnut cross-device tracking. You need to tell CJA which data points work as unique identifiers for people. The email address would be an obvious example. CJA can't select these identifiers for you automatically; you have to tell which ones have individual identification potential. You can nominate one as a primary identifier, too.

It would be ideal if we had a single identifier for people across all devices, but we have a collection of IDs spanning two or more devices, and not always the same devices. For example, let's imagine three devices and three IDs. The work laptop has IDs A and B, the mobile phone has IDs A and C, and the home desktop has IDs B and C. Although we have no single, we can link all activity containing these IDs to a single individual. You might argue that the desktop is the family computer and wonder what happens to the spouse's IDs or the children. The browsing activity from the desktop would span two or more IDs, so identity resolution would still work.

There have been reports about how many screens and, by extension, browsers you have at home. The obvious question should be about how scalable graph resolution is. The answer depends on the timeframe of your analysis. The longer the period, the more reliable identity resolution and CJA will be. However, over short periods, the propensity of having browsing activity data we can't stitch back to a single individual can be a challenge. All the gangsters and terrorists already know this: they change phones often, maybe even every day. If you use the email address as an individual identifier, CJA offers only two lookback windows: one day and seven days.

A best practice when implementing CJA is tracking it in parallel to your existing implementation. You want to achieve as much parity for confidence and peace of mind. That, according to Max, will be no picnic. One issue he faced was how Adobe handled his company's details, i.e. the organisation. In Adobe's vocabulary, the organisation is a repository of everything your company needs in Adobe. That means all your dashboards, segments (filters), the definitions for your calculated metrics, and classifications (mappings). Do everything you can to keep the exact organisation details for your CJA parallel implementation to save you a lot of replication work.

Imagine you are migrating from the Adobe AEP WebSDK, which is server-side tracking, to CJA. Both will use completely separate XDM schemas. The extensible data model (XDM) is a data layer where all the data is grouped in one place, as mentioned earlier. The schemas refer to the mappings between the leaves in that data layer and the data points in the tools that need that data, i.e. eVars (dimensions describing the conversion such as your customer tier group), props (dimensions describing the page view such as a site section name). Various tools in the Adobe Experience Cloud leverage the XDM schemas, but other tools can work with the XDM and XDM schemas, even Google. With a parallel implementation, you will have two distinct XDM data layers and distinct XDM schemas, and the device data looks different. That's one example of inconsistent reporting that will make that quest for parity more elusive than it should be.

Another issue that will frustrate your quest for parity is how they report the environment differently. In the AEP WebSDK, the environment contains information that identifies your browser by its operating system, browser name, their respective versions and more. It sounds like some variant of your browser's user agent string. But on CJA, you have to choose from multiple environments. I assume that's because it seeks to stitch browsing activity from various devices, each with its user agent strings. Also, when comparing the XDM data layers, some are more descriptive than others, i.e. if you use the metaphor of a tree for data layers, one has more branches and leaves than the other, so there again, comparing both will be a challenge, and you will have to prune one and aim for the lowest common denominator, i.e. only compare what you can, and ignore the data points in one data layer without a match in the other.

The environment is not the only dimension in which CJA makes you choose between different possible values for the AEP WebSDK. Because CJS focuses on unique individuals rather than data points to aggregate, you will have not one but several people identifying IDs tied to ID maps. Which ID you choose will impact the traffic metrics, as some will help you stitch browsing traffic data more comprehensively than others.

CJA and the AEP WebSDK also differ significantly in exporting data. With the latter, you can still use the Adobe Datawarehouse, which has no max row limit and can take many columns. Alternatively, a workspace (Adobe's ad-hoc drag-and-drop analysis and dashboard design interface) has a max row limit of 400,000 rows but won't allow more than a few rows. With CJA, the workspace limit is 50,000 rows, with the same breakdown limitations. So, if you need more, you will need a separate database solution to roll up your own SQL. There is a query service capped at 7,000 rows, which may be plenty for most. If your tables contain more rows, there may be a way to handle that challenge by using SQL pagination functions such TOP, LIMIT, and OFFSET, one 7,000 slice at a time, but it's far from ideal.

The data view component shows about 3,000 columns, many of which have very obtuse names and limited business value. After manually pruning that, you will have between 100 and 200 useful columns. The more of that pruning you can do early, the better. The data view component can show you the AEP WebSDK and the CJA data on the same screen, but you might hear the call of the SaaS Pimp and write your own Javascript to make the user interface more helpful. Alternatively, using the Adobe APIs might be an option to get that data, but getting CSVs from the interface is easier.

Session on the dangers of marketing pixels

Sadly, I didn't note the speaker's name, and none of the sessions for that time slot were published on the session board. I will be more than happy to fix that after the article goes live because it was probably one of the best sessions I attended that day.

Did you know that Zoom, which we use for virtual meetings, is battling an $86 million lawsuit because it has been sharing unobfuscated IP addresses with Facebook through having marketing pixels? How many companies have Facebook pixels? Actually, no. How many Facebook pixels do we have on our websites? If you manage your tags with a traditional, client-side tag management system (TMS), all tags contain the visitor's IP address in the HTTP request headers. I am all ears if there are exceptions, but I think it's the norm rather than the exception.

Collecting IP addresses can be an issue, but it's rarely as sensitive as a phone number or an email address because your IP address is, except in rare cases, a fixed number. IP addresses change every few days due to the Dynamic Host Configuration Protocol (DHCP). If you use an Internet Service Provider (ISP), they lease you an IP address for 24 hours. It renews automatically but with a new IP address. Twenty-four hours is the default DHCP setting; it can be longer or shorter at the discretion of the ISP's network administrators.

Is there a way to obfuscate the IP address? Yes, it's server-side tracking! By funnelling all that traffic through your server, you can rewrite anything you like, such as the cookie expiry date that Safari may have overwritten. It seems that obfuscating the IP address is no longer an option but a must. I could also inspect every HTTP request header for anything that could look like a unique user ID and obfuscate that. The marketing campaigns will likely get more expensive as they will require more data to continue accurately targeting.

Another old chestnut of tags is their impact on the website speed. One metric, particularly the time to interactive (TTI), typically increases by 40%. Here we already know the solution: server-side tracking again! By offloading the firing of the pixels to the server, your websites will be faster. Guess who likes fast websites? Apart from your customers and prospects, search engines. Faster websites rank better on the search engine results pages. It's only one of many metrics that determine your page rank, but for most companies, even those with a household brand name, people still use the search engine to find the website when guessing it would have been just as effective.

During that talk, I mentioned a product that is still in beta and open-source called PartyTown, by a company called Builder.io. Javascript is single-threaded, meaning it can only execute one piece of code at a time instead of two or more simultaneously. Back at JS Conf 2017, some Javascript geniuses unveiled WorkerDOM, a way to make Javascript multi-threaded. I suspect that PartyTown is leveraging this. To use a metaphor, it's like going from a swimming pool with a single lane to multiple lanes. With the same number of swimmers, the queue for the first pool moves slowly. With more lanes, the queue is faster.

PartyTown supports Google Tag Manager. There are no public cases of working with Adobe Launch yet. In a server-side tracking context, most tags fire on the server, thus making PartyTowns USP less attractive. But server-side tracking really means a hybrid solution blending traditional client-side Javascript code and server-side processes. It should be possible to push whatever little TMS code remains client-side to a separate browser thread and squeeze the last performance drop. How big will that drop be compared to the time savings of migrating the tags server-side? I have no answer until I tried it, but I can access Builder.io's Discord.

Once you consider DHCP, what are the chances of the class action being in the favour of the plaintiff? However, tackling server-side tracking and possibly using PartyTown may be more expensive. Would you rather pay $86 million and endure reputational damage on top of that?

LEAP turbocharge adoption with data by Kevin Mithani, Adobe

Kevin shared with us a scorecard by one of his client services colleagues at Adobe. Albert Einstein famously said that "everything should be made as simple as possible, but not simpler," thus suggesting that simplicity is a sweet spot and that the scorecard made it too simple. Kevin developed a framework called the Logs Enabled Adoption Programme (LEAP).

The output of that framework is a PDF deck that keeps topline metrics, ensuring it remains quickly digestible for people who are short on time while offering the option to drill down further for curious people. The deck still requires client feedback, typically three months' worth of data.

The deck can provide details such as a split of the users into five buckets: one for the Adobe products user champions and one for the beginners. The deck also surfaces which dashboards people use the most and tracks how dashboards wax and wane in popularity over time.

CJA, which I covered earlier, is currently out of scope for such LEAP decks, but this will change soon.

Running a session? Moi? Tips for a first session, yours truly

The sessions I give typically fall into one of two buckets: cutting-edge technical demos using tools few have ever heard of before or leveraging psychology/neuroscience to have more impact as a data analyst. This session falls into the latter category but focuses on ourselves rather than the stakeholders or clients. Although I have attended close to 60 MeasureCamps now and present at most at least once, I have given about 100 presentations. I had anywhere between 0 and about 80 attendees, but I did not dare to present at my first MeasureCamp. By the time I attended, I had missed four MeasureCamps. Presenting was the last thing on my mind, but by the end of my second one, I felt like I was leeching off the people who made the effort to present. I decided to present at my third MeasureCamp, and I have continued ever since.

For my first talk, I covered single page applications, which are a pain for all digital analytics developers. This special kind of website loads the first page like a traditional website would, but for all subsequent pages, it loads only so-called page fragments, the smallest chunk of code required to update the page. Because the update does not require a page load, you get no page view unless you understand how to tag such websites. I learned a Javascript framework called Angular, got myself a web server and a domain name, and created two demos, my first session ever. About 25 people came. Still, about ten years after my first session last year, people kept searching for new ways to address this challenge.

For those who don't know me well, my background is not computer science, even if I code and like experimenting with cutting-edge tech stuff. I am a linguist, and while doing an internship in Germany, I used to mingle with other French interns like me and Germans our age after work as part of the local French-German social club activities. One of the French was in her final year of her Master's, studying German as her major. She would mock us for the mistakes we made when we spoke, the grammar, the pronunciation, everything. But she was surprisingly silent among these discussions. She was quick to judge others but was too afraid of putting herself in the spotlight. I realised that self-consciousness is the biggest barrier you face when trying to become fluent in a language. Only when you stop caring about making mistakes do you grow, and the same applies to everything, not just languages. The muscles you produce are an overreaction to microscopic muscle tears you cause by pushing yourself. It does not just repair; it makes more muscle.

Prior to my first MeasureCamp, I was used to attending smaller digital analytics events, where the same people came event after event. However, MeasureCamp London attracts people from all over Europe and beyond. The number of people I didn't know was staggering. I stayed until the end of the event and for the after-party, but I felt I needed to recharge my batteries. That is a typical reaction of an introvert.

Now, people tend to confuse introversion with shyness. I would agree that shy people are introverted, but they are not strict synonyms. Extroverted people feel energised by events with loads of people, even if they know few to none of them. Introverted people find that solitude does this for them instead, and large crowds drain them. However, the more MeasureCamps I attended, the more people I knew and recognised. Some even started travelling to attend MeasureCamp abroad, like me. I noticed how the energy drain got slower. I still welcome a recharge break between the closing session and the after-party.

Susan Cain, the author of Quiet and presenter of very influential TED talk on introversion, has been trying to rehabilitate the concept of quiet, shy, introverted people. As an introvert herself, she demonstrates in her book how, despite living and working in a world that prefers extroverted people and promotes them ahead of introverts, being an introvert can be a superpower. Cain argues that this preference for extroversion is not universal. Even among the developed countries, Sweden and Finland see introversion as the ideal.

Music provides many examples of what seems to be a paradox: the most fearless frontwomen and frontmen we have known are introverted, shy, quiet people when off the stage, at home. In the opening session, Keely showed us a picture of Siouxie Sioux, the singer of Siouxie and the Banshees. The band has been very influential in the punk and goth scene. Her startling looks and heavy make-up hide someone shy in real life. The late Freddie Mercury and Prince (thanks to Ton for mentioning Prince) were also shy and needed no introduction. Jim Morrison, singer of the sixties band The Doors, was too shy to face the audience, so he sang, showing them his back, facing his bandmates standing in a circle like when rehearsing instead. Jim's main aspiration was to become a poet, and he published books. Still, he found the gap between his rockstar persona and his quieter real self hard to cope with: "I think of myself as an intelligent, sensitive human being, but with a soul of a clown which forces me to blow it at the most important moments."

My blockers for speaking at MeasureCamp from day one were the lack of an idea for a session, even a discussion, at least one others would find worthy to attend. By the end of my second MeasureCamp, a sense of shame came over me. Here lies another key concept: eliminating the blockers won't turn you into a speaker; something else will. It resonates with another famous quote by Albert Einstein: "We cannot solve our problems with the same thinking we used when we created them."

Another session attendee, I didn't ask for her name, happy to fix that later, explained how she would prepare a lot before a talk, only to find that control slipping like sand through her fingers when she realises she skipped and forgot something. Ton Wesseling , the organiser of the Conference Hotel event in the Netherlands, an accomplished speaker, and even a keynote speaker, attended my talk. Ton explained how nobody else knows and notices when you forget something. But if you let that sense of frustration become visible, they will, so don't show it. It may be easier said than done, but it relates to self-talk. What's your tone when you are judging your work or performance? We are our worst critics, and nobody will criticise us as harshly as ourselves. Making mistakes makes you come across as more human and more relatable. We should seek to be more forgiving towards ourselves.

Another advice from Ton was to let the audience catch up with you and repeat some of your key points. Although repeating yourself may feel tedious, it's just right for your audience. Wait for two seconds for the audience to react to a joke; if they don't, move on to your next point. Also, starting with a personal anecdote helps you make it more relatable. Audiences love stories, but dry data and facts are tough to sell. Even if your audience may contain people who are experts, they view in you their younger self, where they were on their journey with your years of experience, not as the critic they imagine you to be. They know it's hard to speak in public.

Arnout Hellemans , also from the Netherlands, recommended having some slides just before your closing points that you can skip over if you are short on time. Trying to control your delivery is a fool's game; it is more important to come across as human, with all our imperfections. In an age when Gen AI generates all this unoriginal content, recycling old talking points, I believe there are blockers in you and a switch. Not everybody will turn into a rockstar of public speaking when they find that switch and flip it, but I bet that there are inside your presentations and discussions far more interesting than what Gen AI produces.

Bridging the gap: Transitioning into data leadership, Bhav Patel

Bhav, the creator of the CRAP Talks events, shared with us how he managed to break through the glass ceiling in a field notorious for its lack of career progression. Bhav sees a parallel between the role of data analyst and the consigliere in The Godfather films. The consigliere is an adviser to the top decision-maker. So when wanting to break into leadership, we must ask ourselves: Do we want to succeed through our actions or those of others? How well do we manage ourselves?

The concept of consigliere is interesting. I can recommend a book by Richard Hytner called "Consiglieri: Leading from the Shadows," in which he provides many examples of how the head honcho relied on a trusted number two to make important decisions. Although being number two feels like winning the silver medal, there is some appeal in being the brains behind the decisions while remaining discreet. The consigliere metaphor may work well and highlight the importance of developing influencing skills, but it does not apply to the people doing implementation work.

Succeeding through others reminds me of famous quotes by Johann von Goethe, the equivalent of Shakespeare in the German-speaking sphere: "If you want to be a leader, you must first become a servant."

"Treat a man as he is, and he will remain as he is. Treat him as he could be, and he will become what he should be."

The servant-leader vs the micromanager is another old chestnut, but according to Bhav, people will fall along several spectrums, and it is vital to know where you are on these:

1. Needing direction vs self-starter
2. No vision vs vision
3. Individual contributor vs people leadership

People leadership is hard when transitioning to a leadership role for the first time. You will spend a lot of time managing your reports and controlling the quality of their work. Few have support from their senior leadership to manage people, and according to Bhav, the middle manager role is the loneliest because the C-suite has assistants, but you do not, and you are no longer one of the individual contributors either. This feeling of solitude is a challenge for many and an unexpected one.

Data should be:

• available
• accurate
• accessible
• actionable

You also want to be accountable, but does this guarantee that your processes and output will align with the business needs?

Bhav recommended making an inventory of all the people in the business and categorising them based on whether you need to influence them. If they do, you need to influence them one by one and slowly expand your circle of influence and your internal network.

Bahv introduced a quadrant, i.e. two intersecting spectrums: unknown vs known, important vs not important. What you don't know, you should research, but what you know and also know is important is where you should focus your efforts.

Finally, Bhav explained how being a pushover won't get you through the glass ceiling. Being unpleasant and hard to work with begets respect. Bhav admitted how that may go against our nature, as always wanting to help.

How are data professionals adapting in AI advancements, Sairah Chaudry

AI was everywhere on the session board. I managed to avoid most of it, but Sairah's talk appealed to me as I thought I would discover new AI tools. Many are using AI to do more rather than delivering the same amount of work with less effort. One example is writing Python code, and although AI can help, what good is it if you can't explain how the code works? A manager in attendance admitted feeling nervous about relying on code that his direct reports generated with AI and an approximative understanding of the code.

My personal experience with AI is rather diverse. I am using Grammarly to write the content you are reading. It fixes my grammar, but also my tone of voice. The content, the ideas and the reasoning are still mine. I often use a large language model such as ChatGPT or Claude Sonnet to bounce off ideas and see if there's a dead angle in my reasoning. I tinkered heavily with a web version of Stable Diffusion to generate images last winter. For my Javascript Senpai monthly course, I use ChatGPT to create a first rough draft of the text content of the lesson, which I rework a lot. Then, I split the text into smallish chunks and passed them on to a text-to-speech AI. I trained it with my voice, I have to convert most acronyms and some words phonetically, but it works and is a real time saver for me.

In recent weeks, I relied heavily on ChatGPT to port Javascript from a personal automation project to TypeScript and noticed something: straight code translation did not work for me, so I used ChatGPT only to find some stubborn missing bracket, or how to declare variables in the correct data type. For anything else, I noticed how AI made me lazy, making it harder to get back into my rhythm. It felt similar to using a pocket calculator instead of calculating mentally. After a prolonged period of relying on a calculator exclusively, going back to calculating by head becomes harder than ever. AI will divide people between people doing less vs more and between people who can still work without AI and those who can't anymore. I now consciously rely on myself more when coding and calculating.

An attendee also noted how AI is making things more complicated in HR. We are in a period when people are losing their jobs, especially in tech, and young graduates are struggling to find their first jobs. These graduates, and a good chunk of more experienced applicants, have taken AI by storm to write their CVs and cover letters. HR is responding by training its staff to recognise what an AI-generated CV and cover letter look like. All these applications go straight to the bin now.

Gen AI is a fantastic tool if you leverage it to do more and fight to keep your ability to work without as a backup. However, there is a real risk of becoming complacent, and when that complacency becomes blatant, I suspect your job will be heading for the chopping block.

Data communications, Alice Moore

Managing the requests queue of a digital analytics team always comes with challenges despite having ticketing tools such as JIRA, Workday or Trello. Asking the stakeholders the following two questions can help make things more manageable:

1. What is the business purpose for that request?
2. What is the risk of not doing it at all?

With dashboard requests, stakeholders can tend to act like kids in a candy shop, where all the other kids are invisible to them. They can't understand why you can't serve them right now. It's important to remind them that their requests must go through triage. For the teams working in sprints, taking something in after the sprint planning has been completed is possible, but it will come at the expense of taking one or more to make capacity for it. The better approach for dashboards is to start simple and iterate. Gall's Law applies here, i.e. the Big Bang approach won't work, and patching the result won't. The solution was to start small and iterating from there. Should they stop using a dashboard, it is always a good idea to do a post-mortem to understand why so much effort went to waste.

Asking your stakeholders to formulate their requests as "I need X so I can do Y" works well for collecting requirements. With the recent migrations from Google Analytics 3 to GA4, many stakeholders have not upskilled and rely on the digital analytics department for data. They may believe it's quicker for them to ask you directly when it breaks your flow, and it will take time to find it back. If only they could see all the other invisible kids in the candy shop!

Being the bearer of bad news rarely works long-term. In Greek mythology, Cassandra had the power to see the future. When she turns down Apollo's romantic attentions, he punishes her, making it impossible for others to take her predictions seriously. In the nineteenth century, Ignaz Semmelweis, a physician working for the largest maternity ward in Vienna, Austria, found evidence that hygiene saves lives. But his colleagues and his boss took great pride in their bloodstained aprons because the dirtier the apron, the more it showed how hard you worked. But, unbeknownst to them, the dirty aprons were perfect breeding grounds for bacteria that killed mothers. The idea went against the widely accepted understanding of medicine. Semmelweis persisted and clashed with his boss until his boss decided that the problem was not hygiene, but Semmelweis himself and fired him. When bearing bad news, you better have allies to have your back.

Conversely, delivering insight that all stakeholders agree with makes them feel great. In between good and bad news lie findings about which there may not be much opinion or consensus in senior management. They will learn something new. The smart way to achieve this, according to Alice, is to blend findings that confirm that they are right with surprising but safe-to-share findings.

Not every stakeholder is the same. Some will be more numerate than others, especially if they have a STEM degree rather than a humanities degree. You should adapt your communications accordingly. Defining a Northern Star metric can help, but remember to define four more complementary key performance indicators (KPIs) to mitigate people's attempts to game the Northern Star KPI for personal gain.

And now the hardest part of this post. I apologise if missed anybody as I spoke with so many of you, any first time attendees, too. Some of you LinkedIn can't find!

Keely Jacob Anna Lewis ?? Charles Meaden Guillaume Lombard Nicolas Malo Kevin Mithani Trystan Colwyn-Thomas Dominic Woodman Titus over de Linden Pavel Petrinich Dion Jones Thuong-Le Phong Alice Crawley Chiara C. Enrico Pavan Piotr Gruszecki Jomar Reyes Rodney Perry Gavin Attard Stef Elliott ????? Adam Greco Daniel Perry-Reed Craig Sullivan Ellie Hughes Max Lagacé ?????Alexander Holman-Butt Dorian Potma Sarah Crooke Tim Ceuppens Matt Bentley Mickael Lucarelli Mansi Sharma Matt Gershoff Marcella Sullivan Mads-Emil Sykora Larsen Eugen Potlog Tanu Sharma Aymen Tabbakha Nino Weerman Jon Su Andrew Hood Sonia Charles Astrid Illum Omar Khalifeh Kevin Swelsen Tom Carper Camille Chaudet Gideon Delayahu Alun Lucas Phil Pearce Amrdeep Athwal Josele Perez Sébastien Monnier Steen Rasmussen Doug Hall Jon Su Martijn van Vreeden Johan Strand Valentina Gatto David Johnson David Vallejo Tom Robbins Michael Fong Barry Mann Simo Ahava Tim Stewart Alec Cochrane Mark Pinkerton Anisa Boumrifak Stephan Koch Andrew Gershoff Nicholas Redding Adrian Kingwell Samia ABARA Glenn Vanderlinden

#MeasureCamp #London #2024 #Takeaways #DigitalAnalytics #WAWCPH #CBUSDAW