May: We turned two, we're still here and we love data

We have a new website. We know it's only exciting for us, and maybe Alex our graphic designer, but we love it and that’s enough. If you’re feeling curious, have a look here www.fridayinitiatives.com (or as you're reading this, you're on it and hopefully like it - congrats!)

For our London-based Fri-Up readers, we're hosting our first ever Friday Breakfast Club in Late June. Charlie Green (founder of RFI) is joining us to share how he built one of the smartest data positions in financial services research over the years, and what he's got wrong along the way.

In classic Friday style we’re aiming for informal but substantive. We promise that there will be excellent coffee, delicious toasted sandwiches and no slides. We're keeping it to ten or so people so please reply if you'd like a seat, or if you know anyone we should invite.

And, somehow, Friday Initiatives turned two. How exciting is that? Hopefully the terrific twos, not the terribles ones. But any which way, the Emma and Lauren of pre biz owner lyf would be very proud of themselves, their team and the wider Friday community. We launched with a lot of pluck, no understanding of VAT/GST and a hunch that most organisations didn't actually know what data they held. Luckily that paid off. But seriously, if you’ve supported our journey of the last few years in any way, thank you.

And to celebrate here’s a photo of baby Emma and Lauren from so long ago, Friday wasn't even a twinkle in their eyes.

Born to Smug

In essential news, firstborns (aka Emma) have long suspected they're the smart ones, and a new study has handed her the proof. Eldest siblings really do come out ahead on education and income, with researchers tracing it to a mix of healthier infancies and 20–30 extra minutes of parental attention per day. Thank you Economist for covering this essential news. So go ahead, fellow firstborns, frame it.

Lost in Transit

It turns out that many large UK companies are feeding sensitive data into AI systems with roughly the same diligence as posting a parcel without an address. A survey found that 61% of senior tech and data leaders at UK firms with over £100mn in revenue don't fully understand how their data is handled once it leaves the country, even though three-quarters say data flows out of the UK via AI tools at least weekly and a third report daily transfers.

The risks aren't theoretical: regulatory breaches and GDPR fines for unlawful international transfers; confidential strategy, M&A discussions or customer data quietly absorbed into someone else's model; intellectual property leaking to competitors via training data; and a near-total loss of audit trail when an employee can ship sensitive information to a foreign jurisdiction with a single copy-paste. Because of this, analysts expect a real shift toward region-specific AI systems by 2027 as regulatory pressure intensifies. Or, you know, you could implement actual data governance.

Bills, Bots and Bicameralism

The Economist has written about the rise of chatbots in US state legislatures, and it's an interesting read. The headline stat is that 44% of US state legislative staff used AI in their work last year, up from 20% in 2024. The driver isn't novelty, it's necessity.

State lawmakers are typically part-timers juggling day jobs, sitting on tiny staffs (South Dakota has just 60 staffers across 70 House members), and are expected to be fluent across an absurd range of policy areas. AI compresses hours of research into minutes and helps them push back against better-resourced lobbyists, one Vermont rep says she now fact-checks lobbyists in real time during hearings and has caught them lying.

The standout quote comes from South Dakota's Kent Roe, who runs draft bills through Grok for what he calls a 'constitutional stress test': "It's an accepted tool, same as a calculator is to do math, or a cell phone is to do phone calls. We're not in the world of buggy whips anymore."

The concerns are the obvious ones. Hallucinations (especially on case law), a looming wave of "slop laws" overwhelming the lawyers who review them, and the deeper question of what happens when the people writing legislation outsource the thinking. As a Kansas Republican put it: "Your constituents aren't electing Claude or ChatGPT. They're electing you." But still, fascinating.

EU Turn

In a pointed lament, The Economist argues that Europe's growing dependence on American firms, Apple and Google in your pocket, Visa and Mastercard at the till, Microsoft and Amazon in the cloud, is largely a wound it inflicted on itself. Decades of well-intentioned regulation imposed costs that Silicon Valley giants could absorb with their armies of compliance staff, while smaller European rivals, already hemmed in by a fragmented single market, simply couldn't scale to compete. This ultimately resulted in barriers to entry that quietly entrenched American incumbents.

European governments and businesses are waking up to just how much of their critical data, AI infrastructure, and cloud computing sits on American rails, and how exposed that leaves them in a more weaponised geopolitical environment. Sovereignty requirements are only going to tighten from here, and as the article points out, complex data-protection compliance was absorbable for the Googles and OpenAIs of the world but punishing for European challengers. A reminder that thoughtful, well-resourced data governance isn't just a compliance line item, it's a competitive moat.

Perhaps in response, the EU has finally decided that after a decade of cheerfully heaping paperwork onto businesses, it's time to stop strangling itself. Ursula von der Leyen, once a red-tape enthusiast, is now leading a campaign of "omnibus" bills to slash reporting requirements, exempt smaller firms from environmental and human-rights tracking, and resurrect long-dormant ideas like services passports and a pan-European "28th regime" letting startups register digitally for under €100. We live in hope.

Trash, Treasure, Takeover

After two decades of being quietly elbowed aside by Amazon, Walmart and a swarm of niche resale platforms, eBay has staged an improbable comeback. Q1 sales are up 17%, shares up 130% since 2024, and a $56bn takeover offer from GameStop now on the table. The playbook combines authentication for high-value goods, acquisitions of adjacent marketplaces, the removal of seller fees, slick international shipping, and AI tools that turn a half-forgotten Beanie Baby into a listing in seconds. eBay's revival is, in no small part, a story about data finally being put to work, listing data, transaction histories, authentication signals and 30 years of pricing intelligence, and feeding it into AI to make the marketplace easier. It's a useful reminder that the data sitting quietly in your systems isn't just a compliance liability to be governed; properly stewarded, it can be the most monetisable asset on your balance sheet.

Running on Empty (Tokens)

Turns out the AI revolution has a deeply unromantic problem, it's running out of stuff. The Economist reports that weekly token consumption quadrupled between January and March, and the industry simply cannot keep up. Anthropic is rationing peak-hour usage, Amazon is blaming "capacity constraints," and OpenAI has quietly scrapped its video-generation model because it doesn't have the compute. The economic consequences are about to bite including the cheery assumption that inference prices will only ever fall.

The takeaway is that optionality is no longer a nice-to-have, it's essential. Organisations that hard-wire themselves to a single model, a single vendor, or a single deployment pattern are going to find their costs rising, their access throttled, and their roadmaps held hostage to someone else's capacity plan

Ads, Now With Extra AI

AI now runs the digital ad business, and a smart NYT piece lays out exactly how. Brands used to say "target women in New York between 24 and 35." Now Meta tells the brand who to target, what the ad should say, and decides on the fly which version performs best. 30% off campaign costs and 65% off creative with queries up, watch time up and irrelevant ads down 40%.

But the line that should stop a CMO mid-coffee is buried halfway down. "Advertisers have less control over how ads look, where they appear and how they perform." Sold as a feature, it’s a data governance nightmare. Because if a regulator, a board, or a journalist asks why your brand showed up next to that piece of content, or why a particular customer was targeted with a particular offer, the honest answer is increasingly "I don't know, the model decided."

The EU AI Act's automated decision-making provisions don't care that the model was good at conversion. The UK's Data (Use and Access) Act doesn't get easier to comply with because Gemini wrote the copy.

So three questions worth pinning above your marketing team's desk before next quarter:

• Which datasets are feeding our paid AI tools, and would we be comfortable explaining that list to a customer?

• When the system makes a targeting decision, can we reconstruct it six months later, or only the outcome? And;

• If the model gets something embarrassingly wrong, who at the company actually owns the answer?

None of this is an argument against AI advertising. The numbers in the article are real and the savings will keep flowing. It's an argument for knowing what you've handed over before you celebrate getting a discount on it.

Watercooler Chat

A section of the things we like that keep us sane while running a small business…

• Casa Ema in Houston. Some of the best TexMex food the other Emma has ever had.

• This apartment building in Sydney’s Double Bay as featured by the Design Files.

• Whitstable, and specifically oysters at The Forge.

• Stable Wines in our very own De Beauvoir. Go in winter and try not to pay. But tres chic.

• This charmingly indulgent ode to cooking for oneself from Nigella Lawson. Read and put on some pasta.