The insight exists. The fix is known. It just never gets installed properly. That's where I work: activation, retention, acquisition and the infrastructure underneath.
Erwin van Haren · Longines Global Champions"Keeps his promises. Involves you as a client in his thought process."
The channel engine running at The Cirqle today. Every stage has an owner, a metric and a written rule. The thresholds stay internal; the machine doesn't have to.
Paid social, LinkedIn, outbound and co-marketing with Meta's GTM team. Creative runs through a 12-cell matrix, tested against written kill and scale rules.
An ICP-scoring model steers lead quality before sales touches anything. Volume without fit is noise; the model filters it out.
Demo bookings feed the warehouse. Activation and CS-to-product loops carry the account after the close, so retention starts on day one.
CAC payback per channel, CPA thresholds per creative, spend floors before any verdict. Rules written before the test runs, not after the result disappoints.
Every signal lands in BigQuery, modeled in dbt, read by an agent each morning. Decisions come from the warehouse, not from platform screenshots.
The stack reads the warehouse overnight and reports what moved, why, and what to do about it. Three trajectories, three different machines, one way of working.
Paid works when it's an operating system: a strategy, then a multi-layer roadmap you can actually run. Creative volume, kill and scale rules, ad set structure, CPA multiples, which KPIs matter and which just look busy.
→ What do you do when 90% of budget piles into one creative and the other million assets starve?
→ How do you creative-test after splitting CBO into ABO without becoming an admin machine copy-pasting for Meta?
→ To hook rate or not to hook rate?
→ And when the test finally ran: can you actually read the results?
The read I run monthly on any acquisition mix. Leads come in, then the columns decide what happens next: CVR, CAC, payback, LTV:CAC, margin. A high lead count means nothing until it survives those columns.
| Channel | CVR | CAC (idx) | Payback (mo) | LTV:CAC | Margin |
|---|---|---|---|---|---|
| Paid social | 3–6% | 1.5× | 6–9 | 2.8× | 62% |
| Demand capture | 8–12% | 1.1× | 4–6 | 3.9× | 68% |
| Outbound | 2–4% | 1.3× | 5–8 | 3.1× | 65% |
| Referrals | 15–20% | 0.4× | 1–2 | 7.5× | 81% |
| Partnerships | 10–14% | 0.6× | 2–4 | 5.4× | 58% |
| Events | 12–18% | 0.9× | 3–5 | 4.6× | 60% |
Illustrative bands. CAC indexed to blended = 1.0×. Real thresholds stay internal.
Referrals and partnerships run LTV:CAC above 5× with payback inside a quarter. Paid sits at 2.8× and 6–9 months. Blended numbers hide exactly this spread — the read only works per channel, per cohort.
The channels with the best ratios are the hardest to scale with budget. Paid is the only line that absorbs spend on demand, which is why its funnel economics get fixed before its budget gets raised.
Budget flows to the columns, not the channel: anything under 3× LTV:CAC or over 6 months payback earns a fix, not more spend. Everything above earns volume.
Built solo, running in production. AI scales the execution — it isn't the story.
Audits an ad account the way a senior buyer would: pulls it apart, catalogues the failure modes, prescribes the fix. Run across 34 brands.
Every growth signal in one warehouse. An agent reads it each morning and reports what moved, why, and what to do about it.
Audience segments crossed with proof pillars from customer verbatims. Every creative lives or dies by a written rule.
A spec Alfa Romeo commercial produced without a camera. Static JPEGs in, Nano Banana Pro for the look, Kling 2.6 Pro for the motion — controlled with an 11-factor variable list instead of re-rolls. Built on Pletor, where I'm a partner.
Pletor is the production infrastructure this runs on: image, video and text models orchestrated in repeatable pipelines with brand context, not a prompt box. The exact template behind this ad is public.
Duplicate the workflowOne kb_view event, two cohorts, one retention delta. How a 58pp business case gets measured in practice.
Flux, Imagen and Sora 2 in a production content pipeline. The system, not the pixels, is the asset.
Argued it back when the market still bought follower counts. It took the industry years to get there. Working today at a company built on that thesis.
Long-form write-ups with the numbers left in. The full workflows and economics behind the AI creative section above live here.
A spec Alfa Romeo commercial from static JPEGs: Nano Banana Pro for the look, Kling 2.6 Pro for the motion. Controlled with an 11-factor variable list and hard negative constraints — engineered, not prompted and prayed for.
Read it → Economics$0.35 vs $5.00 per 5-second shot. A 20-second commercial at 75 generations: $26.25 on the decoupled image-plus-video stack vs $187.50 on Sora 2 Pro. Same brief, 7× the margin.
Read it → StrategyChurn is rarely a pricing problem, it's a proof-of-value problem. Users who hit the knowledge base in their first 7 days retained at 78% vs 20% — a 58pp gap, measured with one event and two cohorts.
Read it →I've always figured things out by taking them apart. Started at six with piano, no sheet music, just listening until it made sense.
Modeled for 20+ global brands during my studies. Liked numbers and complicated problems. The corporate path didn't appeal to me. I drifted.
Then I picked a direction and went fast.
Came at advertising backwards. Data management and CDPs first, ads second. Built a webshop, scaled it, watched it fail. Joined an agency, built and coached a team from scratch across paid, creative and strategy - while consulting for SaaS and enterprise clients at the top end of the market.
In 2017 I argued brands should stop chasing high-follower Instagrammers and bet on small, believable creators. The industry took years to agree. I now work at a company built on that thesis.
Tried learning Python. Failed. Somehow know my way around regex. Have two products in production. AI is part of how I scale execution, not the story.
That's still how I work. Just with growth systems now.
Most growth problems aren't strategic. They're operational. The insight exists. The fix is known. It just never gets installed properly.
That's where I work. Activation, retention, acquisition and the infrastructure underneath.
Activation, retention, acquisition or the infrastructure underneath. Bring the problem, I'll bring the mechanism.
No autoresponder. I reply myself, usually same day.
I haven’t set up automatic email responders that send the schedule to you. So as soon as I receive a notification of this request, I’ll send it over to you.
Talk soon.
B