Lab · 7 tools · updated October 2026
Seven tools you can run.
They come from my work on AI agents and marketing measurement. Each one runs in your browser on sample data, and three also take your own CSV. Change an input and the results update.
Media mix model & budget optimizer
Fits a media mix model to two years of weekly spend and revenue, then moves budget to the channels where the next euro earns most.
Default run: shift €12k/wk toward Paid Social, +€6.0k revenue/wk
Open →
Geo-lift incrementality test
Tests whether a campaign caused its sales lift by comparing test cities with a control built from the rest. It also sizes the test before you run it.
Default run: +8.9% lift (90% interval +1.7% to +15.4%), iROAS 1.67×
Open →
Lead scorer with product-usage signals
Ranks accounts on fit and product usage, shows the rule behind every point, and lets real conversions re-weight the model.
Default run: 10 of 50 accounts clear Tier A; learned weights lift AUC 0.72 → 0.77
Open →
Brand & claims gate for AI-written copy
Checks AI-written ad copy against brand voice, approved claims, competitor mentions and regulated wording, then passes, holds or blocks each draft.
Default run: needs review, 3 warnings add up to 35 against a limit of 20
Open →
CRM dedupe & standardiser
Turns a messy contact export into one clean record per person and explains every merge. Your file stays in your browser.
Default run: 328 rows → 201 records, 7 pairs left for review
Open →
Multi-agent swarm with no one in charge
Replays a real run in which agents agree on work without a coordinator. See why each proposal passed, then add an objection and watch the outcome change.
Default view: an objection at event 15 reopens a proposal that had already passed
Open →
Governance layer for marketing agents
Logs and checks every action a marketing agent takes, then records whether it would auto-approve, need review or be blocked. Decide the held ones and the weekly report updates.
Default run: 22 of 27 actions auto-approve, 3 need review, 2 are blocked
Open →