When fragmented commerce workflows become one operating system

Orders lived in five consoles (two marketplaces, D2C stack, WMS export, finance spreadsheet). Ops “swivel-chaired” to decide cancellations and address fixes. RTO hovered near 18% on high-ticket categories; finance could not reconcile contribution until mid-week.

Growth ran promos without inventory buffers tied to marketplace SLAs—twice in six months, hero SKUs oversold during a livestream, damaging seller health and CX.

OmniBrain pilot was politically sensitive: leadership needed explainability, not a “black box” block list.

Each outlet manager maintained partial shadow sheets for Swiggy/Zomato adjustments; chargebacks surfaced two weeks late in lump CSVs.

Kitchen display and modifier errors spiked cancellation rates during dinner rush; HQ lacked a single ticket timeline across brands.

Expansion to two new cities duplicated onboarding pain—no standard playbook for menu syndication and tax mapping.

BOPIS and ship-from-store ran on store managers’ initiative; ATP in ecommerce never matched backroom counts until end-of-day batch.

Transfers between stores were often “known informally”; HQ planning used DC snapshots that lagged floor reality by 2–3 days.

Peak seasons produced angry regional chats when same SKU showed “available” online but not on shelf.

SKU count crossed 4k; listing errors and fee changes created silent margin erosion on long-tail items.

Return reasons were inconsistently coded—demand planning and AI experiments lacked trustworthy labels.

Working capital swung wildly with payout timing; leadership had no forward view tied to actual remittance calendars.

Each account team built custom Looker + sheet stacks; client finance questioned ROAS that ignored returns and fees.

Experiment holdouts were documented inconsistently—hard to prove incrementality in QBRs.

When a client hit inventory walls mid-campaign, blame bounced between media and ops.