We just wrapped up the INMA Product and Data for Media Summit, organised by my fellow co-curator Jodie Hopperton and myself — a programme packed with insightful speakers and so-many-things-to-think-about that, well, it’s giving me lots to think about. If you couldn’t catch every module, remember all of it is available on demand — insights on tap, if you will.
We now head into the holidays and I have a bit of lull in events with INMA, which means … I want to chat with you. What awesome things are you working on? I live in the “reply” button.
Chat soon and all my best, Ariane
4 best practices for building strong data teams
At the Product and Data Summit we just wrapped up, one of the modules dealt with “the way we work.” In particular, we heard from José Meroño, the former head of data at Prisa Noticias of Spain, the publisher of El Pais.
I have mentioned José before when we originally connected a few months ago because he had a very interesting story to share about how Prisa spun up its data team. The headline is this: Prisa created a mature, “full-powered” team — including the architecting, the engineering, and hiring work — in 18 months.
This is … very speedy — more speedy than I would have imagined possible in the context of a reorganisation.
But there were some particularly useful insights in what José shared with us. And so, for the folks who didn’t join us at the summit — or maybe who want a handy blog post to share with their boss as they try to negotiate the budget of the data team next year (I’ve been there) — here’s what we learned:
1. Build a team with a specific mission, even when the team isn’t there yet.
José was hired as a seasoned data exec with an early mission that wasn’t so much “build the data team” as “build the data team to support our new foray into paid subscription.”
I think when we spin teams, those that find their footing faster are those that are built in response to a specific goal or task to accomplish. Even as many publishers think, “Hey, I know data is important, we need a bigger team to accomplish our goals,” I would ask that big boss: What is the key mission of data in the next year?
And mission isn’t: “Make us more customer centric.” That’s a company value — not the mission. But Prisa actually gave José a proper mission, which could underpin pretty much every decision in terms of what type of folks to hire or what architecture to develop for the data platform.
It also gave some time horizon to the enterprise. The subscription push was underway, so what were ways that the data team — even as it was being spun — could begin to make contributions to the mission.
2. Find ways to bring value as early as possible.
I speak here as someone with extensive product management experience in heavy-duty technology (analytics tools, CMS) — the kind of product that rarely is a light lil’ MVP. So I am particularly sympathetic to the travails of imagining the earliest version of what you have and how you’ll work with your earliest users.
Usually, that’s a pretty dismal affair. The products don’t do much, the users are living in some kind of super crude system, and the output of all this is usually pretty basic. Which is why I was quite impressed when José told us he went from 20 early testing users in the second quarter of their buildout to a 100 in the third quarter.
Reader, I wanted to reach across the screen and say, “Oh my god, don’t do that!” (It was like when you know where the killer is and you want to tell the blonde on screen, “No, don’t lock yourself into that closet!”)
In any event, goes to show that I was too conservative because José’s team made it work. And what they got was, of course, more customer feedback early — which is invaluable to course-correct before you’re overly invested but also something that every young data team eventually stumbles on …
3. Figure out where you’ll be bleeding money, as quickly as you can.
One of the things José and his team learned in that first phase of scaling is just where their data machine was going to eat them out of their house.
Now, in a previous job, I remember being on a very uncomfortable hot seat during an emergency meeting because my team’s data budget had unexpectedly exploded in the span of a month as we had onboarded some new customers.
The number by the way was astounding. But the mistake was very much mine: We had had a slower Alpha, and then, having established that all was ship shape, deeply accelerated for a Beta.
The problem is, of course, that data costs tend to be pretty linear with usage when you’re running a lot of uncached, custom queries, and other “rich” pulls. It’s the thing we often forget about because we’re focused on data availability, cleanliness, the experience of our users who interact with our data … and costs feels like a thing we can always get ahold of.
And we can get ahold of our costs. José killed a bunch of expensive reports and only gave certain power users the ability to pull the more expensive dashboards. But there is a real disruption to having to re-engineer potentially significant parts of your systems because you built them out without an early feedback loop into their true costs at scale.
So José had to take on the burden of more customers early, when his team was only six or seven people. But, on the other hand, the build was still early, and they could identify where they’d need to stop the bleeding — before they had fully committed to that bit.
4. Bring in contractors to speed up your time to market while you build the forever team.
I am not going to tell you anything you don’t know when I say that even in the context of recessionary times, hiring for data is a contact sport. So is there such a thing as efficient time-to-market when hiring is going to be either badly rushed, or, well, will get in the way of your timeline?
José had no qualms about hiring contractors to support the team in the build-out, but doing this gave him something else as well. Since he was hiring for well-defined needs, he didn’t have to overthink whether this person was the right mix of skills and experience for the team as it would evolve over time.
One reason early data team hires are often difficult is that you need a few key folks who are jacks-of-all-trades: data engineer, data architect, data scientist … and often a product manager and an executive whisperer. These people exist, but it will take interviewing to find them.
Meanwhile, finding folks who can build some of your early data pipes in your cloud of choice — and not really worry about whether this is someone who will be able to evolve into a different role in six months — that’s significantly easier.
And time is money, which is why time-to-market is a real thing even for a trade, like data, that usually has trouble assigning ROI to many parts of our daily contribution.
Since Prisa had articulated the mission as supporting the subscription effort, there was a way to articulate the value of bringing the data team online earlier rather than take more time to get there.
We come full circle, really, to the point that the transformative approach when it came to this team was that it always had a specific mission.
If you’re a publisher who is assessing how or where or how much to grow your data team, first find a core mission for this team. Not soft n’ squishy aspirations but an actual mission that will orient their choice and help choose way stations to start delivering value early and consistently.
Further afield on the wide, wide Web
- They are called the “Bias Buccaneers”, and it’s such a great name. I’m going to regret I have none of the skills to join this group because I certainly would have loved to introduce myself as a member of the Bias Buccaneers. But anyway, you’ve got two weeks to build a machine-learning model for face detection that is as free of bias as you can make it. It’s a bounty competition funded by various large corporations, the first of what will hopefully be a series of other “bias bounty” competitions.
- Thanksgiving is coming up in the U.S. — the best American holiday in my opinion since it’s mostly food-based. The New York Times took an interesting tack in their cooking coverage with this YouTube video that’s diving into AI-generated recipes for Thanksgiving. OK, look, I’m sharing this clip because having GPT-3 generate recipes is fun as a thought experiment, but know that I was yelling at my computer when the segment got to the stuffing recipe. I want this AI put in jail.
About this newsletter
Today’s newsletter is written by Ariane Bernard, a Paris- and New York-based consultant who focuses on publishing utilities and data products, and is the CEO of a young incubated company, Helio.cloud.