Activating systems, from the inside out

My first attempt stalled among the many campaigns and other projects we were running at the time. When the organisation shifted its focus to B2B, we decided to finally implement a CRM, and I was a key member of the group that selected HubSpot and the implementation partner.

My part in the implementation was the technical side and the data, which I knew well. I connected Google Analytics, Facebook and LinkedIn, and coordinated the DNS changes with the people who had access to the domain. Then I migrated the data out of Mailchimp.

For the cleanup, I defined lifecycle stages and marketing statuses and removed duplicates across 80,000 contacts and 12,000 companies. The aim was to show marketing and sales where their time is worth spending.

The last step was enrichment. Phone numbers and HubSpot’s standard enrichment filled in the basics. More useful for sales was what we already had without seeing it. Course bookings came in from Eventbrite, but the integration alone didn’t show whether a company’s employees had followed a course with us, or which types of course, a first impression of the size of a possible deal. I added both to the company records, along with which education platforms the companies were connected to.

About 2,500 companies were missing an industry, and a similar number a country, both of which count in the lead scoring. I designed the rules that mapped them onto a fixed set of labels and the checks on the results, and AI tooling carried out the research in stages. Where the data couldn’t support a label, the field stayed empty or got a label that flagged the unknown. Companies whose country couldn’t be determined now work from the Vatican. A clear flag, and easy for the team to remember.

Tracking came first, and it wasn’t simple. Five partners run the same website, and the tag manager isn’t ours, so I had no access to it. Even so, we got the buy-now buttons and Calendly calls tracked.

Reporting started in Google Analytics, and I managed the move from Universal Analytics to GA4. Excel came next, because it made the insights tangible for colleagues who didn’t naturally gravitate towards hard data. Every week I imported data from Google Ads, from GA4 (traffic, conversions and the channel they came from) and from Eventbrite (the bookings). Each import needed cleaning before it would show up in the pivot tables and graphs.

From there I could estimate the return on ad spend per course quite precisely, and those figures went back to the course developers. The data also showed a clear split: Google Ads worked best for the Professional courses, while Meta sold the Instagram-worthy online programmes very well.

Over time the separate reports, now with Meta’s ad data alongside, grew into dashboards, first in Excel and now built with AI tooling. Updating them takes much less time, and with everything in one place, colleagues started using them. The same numbers have since supported budget decisions, tooling choices and several negotiations.

The latest step goes deeper. I export the HubSpot and Eventbrite data, anonymise it, link the two by ID, and analyse the result with Claude to follow the customer journey in detail.

I laid the groundwork for this thorough approach in earlier research.

Writing brought me into marketing. I had written more than 200 blog articles when I started as a copywriter at DTG. There, I wrote website texts for over 300 clients, and the work went well beyond the words: search term analysis, meta texts, heading structure and the navigation itself. Later I built the ads for around a hundred more, and when DTG became Youvia, I managed the move of the website’s content.

At The School of Life Amsterdam I first sharpened my writing. I took a course in science journalism at SCW, taught by science journalists from NRC, de Volkskrant and Quest. We were a horizontal team at the time, so I convinced colleagues to meet as an editorial group, get the content calendar filled and edit each other’s texts. Over the years I wrote dozens of articles and newsletters, mostly around our courses. One of them, on the Ben Franklin effect, ranks in Google’s top three and is cited in its AI Overview.

I also initiated the current ads strategy and carried it out. With proper attention it scaled tenfold, with results to show for it: around 11 times return on non-brand search. What people search for also goes back to the team developing the courses.

At The School of Life Amsterdam I’m the point of contact for AI. I started with the criteria: a list for choosing AI systems that works for any vendor, which I then used to compare Claude and Copilot. The comparison ended in a short rule of thumb: Claude for creation, Copilot for coordination.

The rules followed. I wrote the governance documentation and the usage guidelines. One of the clearest decisions was to keep Claude outside our Microsoft 365 environment, so it only works on documents someone hands it. Customer data we analyse with Claude is anonymised first.

Training runs alongside as a small internal campaign: short AI tips in the all-staff meetings, a tips thread in Teams, and regular messages there on where the company stands with AI. Colleagues can also book a one-to-one session with me to work out how AI fits their own work.

To see whether any of it lands, we measure. A survey in January, another in August and a third in December follow how colleagues use AI over the year. The strategy team has agreed with my approach to official adoption.