three recent engagements. clients anonymised by request — the problems and the outcomes are real. names available on a 20-min call.
engagement i
a london field-marketing agency · ~40 staff.
the problem.
monday mornings were a fire drill. account managers across three city teams pulled performance data from postgres, sales rosters from a spreadsheet, and visit logs from a separate crm — to send a single weekly to each client. by wednesday they were already behind on the next one.
what we built.
a client-portal dashboard cluster on top of a bigquery warehouse fed from their postgres replica. one surface per client, refreshed nightly, branded as theirs. account managers stopped pulling data — they answered questions on top of it.
outcome.
monday work down to ~30 min.
two clients renewed on the strength of the reporting alone. account managers spent the freed time on next-quarter pitching.
engagement ii
a uk shopify dtc brand · 8-figure run-rate.
the problem.
contribution margin was scattered across shopify (revenue, cogs), klaviyo (email-attributed revenue), meta and google (spend, attribution). the leadership team was reallocating media weekly off blended roas — a number that didn't reflect actual product profitability.
what we built.
a unified contribution-margin dashboard on bigquery. daily refresh. cogs joined to orders, returns netted, attribution credited per channel using their existing rules. one number per day per channel — real contribution per pound of spend.
outcome.
media reallocated weekly on real cm.
leadership stopped arguing about attribution methodology. the dashboard became the source of truth for the weekly media meeting.
engagement iii
a 12-person digital agency.
the problem.
the team ran paid search and paid social for ~20 clients. every monthly review was a 4-hour build — pulling gads, gads search-terms, meta ads, ga4. by the time the deck was ready, the data was already a week stale.
what we built.
a looker studio cluster — one branded surface per client — fed by a bigquery warehouse with daily fivetran loads from gads, ga4, and meta. account managers opened the surface, narrated the trends, sent the link.
outcome.
weekly client reviews from 4 hrs to 40 min. per client.
the agency took on three more clients without hiring an account manager. retention on weekly reviews went from 'occasionally we skip' to 'every monday, every client'.
v — how this works.
build in 2–3 weeks. weekly iteration. on-call for breakage. cancel any month. see what's managed →
vi — who we work with.