SurfaceNet — Business Overview

Methodology & caveats

Filter — matches Brightpearl

Every figure applies the canonical “Orders We Count” filter from queries/orders_we_count.sql: exactly the 8 statuses Brightpearl’s own order report uses — New web order, New phone order, Payment Pending, Review, Warehouse, Back order, Invoiced, On hold. Excluded: Cancelled, Quote sent, Draft / Quote, Quote Not Filled and Quote Filled.

Quote Filled is excluded because filling a quote creates a separate invoiced order for the real sale and leaves the original parked — so counting it would double-count. One further order (258180) is excluded as a confirmed duplicate of Invoiced 259048. Accrual basis: invoiced-but-unpaid orders do count.

Verified against the native Brightpearl report for June 2026: this filter reproduces it exactly — 386 orders / $437,886.98.

Definitions

• Revenue = Net Sales + S&H, which is Brightpearl’s “Net Revenue” (product + shipping, excluding tax). Bare Net Sales omits shipping and runs low. Gross Revenue = that plus tax = OrderTotalAmount.
• Gross margin is (Net Revenue − Extended COGS) / Net Revenue. Note this is Brightpearl’s own convention: shipping revenue is in the numerator but shipping cost is not in Extended COGS, which flatters margin by roughly 3–4 points. It ties to their reports; it is not a true product margin.
• The order date is NOT KeyedDate:. That column holds the UTC date, so orders placed 17:00–23:59 Phoenix are stored a day late — it matches Brightpearl for only 90.2% of 2026 orders. We derive the true Phoenix date from bp_orderlines.WarehouseDate: (which the export populates with Brightpearl’s createdOn — the very field their report filters on), shifted UTC−7. Accuracy: 99.92%, zero wrong derivations. Brightpearl’s account timezone is America/Phoenix (UTC−7, no DST), not Pacific.
• AOV is net sales ÷ order count.
• The margin decomposition is revenue-weighted: the counterfactual holds the 2020 channel mix and substitutes 2025 per-channel margins.

Things to know about this data

• 2026 is a partial year — through 16 Jul. It is charted but tagged “YTD”, and the hero comparison is cut to the same calendar day across years.
• The mirror effectively starts in 2020. Five orders sit in 2008 ($1,816 gross) and there are zero orders from 2009 to 2019. Those five are excluded from the charts (they remain in the all-time totals). Treat “all-time” as “2020 onward” plus a rounding error.
• Repeat rate is 25.3%, not the ~22.7% baseline. This is a real discrepancy and I could not reproduce 22.7%. Computed by distinct Customer ID: over the counted set, the rate is 25.25% (4,898 of 19,396 customers with more than one order). Variants tried: excluding partial 2026 → 25.20%; dropping the canonical filter → 24.80%; web-only → 23.98%. None land on 22.7%, so the baseline likely used a different customer-identity definition (email or company rather than Customer ID) or a cohort window. Worth reconciling before this number is quoted externally.
• An undocumented channel 3 exists — 5 orders, $1,816, all in 2008. CLAUDE.md documents only 1 = phone and 2 = web. Immaterial to every figure here, but real.
• IsCustomersFirstOrder: is empty on every counted row — it looks like a field the mirror never populated. Don't build on it.
• On bp_orderlines, UnitPrice:, Quantity, Net Sales and Extended COGS are varchar, not numeric. Line-level arithmetic needs CAST. The bp_orders money columns are proper decimals, and this dashboard is built entirely from the header table.
• Because the mirror is a faithful copy, it is corrected at the query layer, not by rewriting stored data — nothing downstream (the Laravel app, the BPDATA→RJMetrics export) is affected by these adjustments.

Source

Queries: queries/business_overview.sql. Read-only via the MySQL MCP connector. Generated 16 Jul 2026.