SUGARBEAR’s leadership deck used to be rebuilt by hand every period — exports, paste, reconcile, write the sentences, hope. We built the data layer that assembles it instead: every number with a definition, a source, and a refresh time attached.
SUGARBEAR is a consumer brand. The numbers leadership sees every period live in four places that don’t talk to each other: the storefront, marketplace settlement reports, a 3PL portal, and the accounting system. The deck that summarises them was assembled by a person — a capable person, which made it worse, because assembling it ate the day they were hired for.
The ritual: export everything, paste it into last period’s file, chase the rows that don’t match, rebuild the charts, then write the sentence under each chart. Deck due Thursday morning meant Wednesday was gone. Every period. It had been running this way long enough that nobody called it a problem. It was just how Wednesdays worked.
The charts were never the hard part. Charts take an afternoon. Two things were actually broken, and neither shows up in a screenshot.
First: definitions. “Active customer” meant three different things depending on who pulled the number — trailing 90 days for one person, calendar quarter for another, and a third version that quietly included influencer seeding orders at 100% discount. All three were defensible. All three appeared in decks. When two slides disagree by 8%, the meeting stops being about the business and starts being about the spreadsheet, and that meeting costs more than the deck ever did.
Second: trust. The rebuild-by-hand ritual survived because a human was implicitly re-checking everything as they pasted. Automate the assembly and you remove that human. Nobody double-checks a system they trust. Which means an automated deck has to be right in a way a manual one never had to be. That constraint shaped everything we built.
A reporting layer that sits between the four source systems and the slides. Source rows land in a warehouse, pass through versioned metric definitions — written down, numbered, argued about once instead of monthly — and come out the other side as the five slides leadership actually reads, commentary included. The sentence under the chart is the part executives read; it’s also the part that used to take a human an hour per slide. It’s generated from the same numbers now, so it can’t drift from the chart above it.
Below is the mechanism, live. Step through the deck and watch each slide assemble — rows in, aggregation, chart, commentary. Then flip any slide over. The back of the slide is the actual product: where the number came from, which transforms touched it, what was checked before it was allowed into the deck.
All figures below are representative, not SUGARBEAR’s actuals. The mechanism is the real thing.
Arrow keys work. Hover any mark for the exact figure. Flip any slide to see its lineage.
Settling “active customer” took three meetings and produced definition v3. v1 died because it counted the seeding orders; v2 died because finance and marketing used different windows. v3 is written at the bottom of the slide it feeds, with its version number, which turned out to be the thing that ended the argument — not the definition itself, but the fact that there visibly is one.
The build refuses to run on stale data. If a source hasn’t refreshed inside its window, the deck doesn’t assemble with a quiet asterisk — it doesn’t assemble. A late deck gets asked about; a wrong one gets believed. And every build reconciles against the ledger before it ships: revenue has to tie to the GL within tolerance, channel totals have to sum to the revenue slide exactly, cohort counts have to add up to the deduped customer table. Boring checks. They caught a duplicated settlement file in the second month that a human paster would have missed at 11pm.
The most annoying single fact in the whole system: marketplace fees settle on a two-week lag, so the freshest two weeks of margin are estimates. The wrong move is hiding that. The slide flags the estimated window instead, because an executive who finds one silently-wrong number stops trusting all of them. Then you’re back to rebuilding decks by hand.
What we’d do differently: version the definitions from day one. We started in week six, after the second definition argument, and migrating the early metrics onto the versioning scheme was a week of unglamorous backfill that a day of foresight would have avoided.
Assembly takes minutes, unattended, checks included. Wednesday belongs to the analyst again. They review the deck now instead of manufacturing it, which is the job description they were hired under.
The quieter change is what disagreements look like. Nobody argues about whose number is right anymore, because there’s one number with a definition version printed under it. The arguments moved to whether the definition is right — which is a better fight. That one actually improves the business when somebody wins it.
A deck nobody has to build is nice. A deck nobody has to doubt is the product.
Tell us which deck eats your Wednesday. We’ll tell you which parts of it a pipeline should be writing.
contact@gopivotedge.comOne email. We reply with questions about your sources, not a pitch.
Net revenue hit $4.29M in July, closing an $11.7M quarter — up 11.4% on the quarter before. Two thirds of the growth came from repeat orders, not new customers; acquisition spend was flat.