SellerFolio

The back office
TikTok Live sellers
actually run on.

One console that turns a TikTok Shop's raw order, settlement and live-stream data into costed profit, packed parcels and answered returns — with AI doing the reading and humans keeping the pen.

14
screens, one login
5
AI assistants, human-approved
$-level
TikTok settlement reconciliation
Roles
owner · manager · packer
Why this exists

Live selling moves fast. TikTok's tools don't keep the books.

A TikTok Live seller can move hundreds of orders in a two-hour broadcast — then spends the next two days asking questions Seller Center never answers:

  • Did that show actually make money? GMV is loud; costed profit is nowhere.
  • Which return do I have to answer by 3pm before TikTok auto-refunds it?
  • Does this label cost less than what the buyer paid for shipping?
  • Which bin did those units come from — and did the count survive the week?
  • TikTok paid me $X. Of what, exactly? The settlement is a wall of fee codes.
  • Who is this buyer — VIP, repeat, or one return away from a loss?
Home · /dashboard

Dashboard — the 30-second brief

Today's deadlines, period sales and cash position on open — plus how much of the window is settled fact vs. still moving.

  • Action strip: returns to answer, parcels past ship-by, items missing cost, reserve releases.
  • Five KPI tiles — orders, GMV, net payout, net profit, refunds — each a doorway to the screen that owns the number.
  • Trust strip: % of orders settled, % of revenue costed — the console refuses to let half-costed data pose as final profit.
  • Three labelled clocks (order date, statement date, live) so "contradicting" figures reconcile on sight.
Dashboard overview
Dashboard: action strip, KPI tiles, cash position
Sell · /shows

Shows — every broadcast, ranked by $/hour

Every TikTok room discovered from orders, ranked by dollars per hour — and matched to the video receipts of what was held up on camera.

  • Show list: orders, GMV, $/hr, audience, net payout per room; name and compare broadcasts.
  • Show page: money-pace curve, selling-intensity bars, hour-by-hour table, fees & payout.
  • Receipts & AI: one-click capture of video-receipt links via the Chrome extension; server-side transcription turns each clip into brand / item / colour / size.
  • Capture is free and re-runnable; transcription is the metered step, queued with progress and cancellation.
Shows list
Shows: one row per room — receipts, GMV, $/hr, net payout
Sell · /orders

Orders — the ledger, drilled to the dollar

Every order searchable by any field, filterable by stage, show or settlement — and one drawer deep per order.

  • Ledger: time window, settlement filter, show picker, field-scoped search, stage strip.
  • Order drawer: status history, money, package & tracking, items/splitting, AI video receipt.
  • Settlement receipt on demand: itemised by posting leg with TikTok's own field names — the order's financial birth certificate.
  • Label write-back: warns when TikTok hasn't received tracking and writes carrier + tracking back, preventing auto-cancellation.
Orders ledger
Orders: search, four figures, stage strip, ledger
Sell · /customers

Customers — a CRM TikTok never built

Buyers keyed on TikTok's stable buyer ID — segmented, flagged, and mapped — with coverage meters that admit when the data is thin.

  • Buyers grid: segments for VIP, repeat, lapsing, return-watch, one-and-done — with exact published thresholds.
  • Buyer drawer: lifetime value, shows bought from, returns, recent orders.
  • AI "usually buys" tags: size/brand/colour projected from show transcripts, each with an explicit confidence band.
  • Demographics tab: US choropleth, ZIP3/city demand, when-they-buy, loyalty, audience by show.
Customers grid
Buyers: segments, lifetime value, flags
Demographics
Demographics: demand by state, ZIP3 and hour
Deliver · /fulfillment

Fulfillment — queue → labels → proof of delivery

From "to ship" to "delivered" in one screen: buy labels, consolidate parcels, track carriers, and see where the days actually go.

  • To-ship queue with ship-by urgency bands and same-buyer consolidation into one parcel.
  • Manifest dock: ready-to-buy vs. needs-attention, box & weight source, printable manifest.
  • Shipping-margin preview before you buy — buyer-paid shipping vs. label cost, so money-losing labels never ship by accident.
  • In-transit tracking: background scan poller, five-stop journey rail, stuck / failed / SLA-breach verdicts.
  • Delivery tab: stage-by-stage dwell chart with median & p90, by destination state.
Delivery performance
Delivery: where a parcel's days go, stage by stage
Deliver · /packing

Packing — the warehouse bench

A bench loop built for speed and forgiveness: tick parcels, print labels in batch, pull cancelled items — nothing irreversible.

  • Fast tick for the common case; careful pick drawer for multi-item orders; "mark all packed" for the run.
  • Batch label printing: label or label+slip, merged into one PDF for the printer.
  • Cancelled-item pull list so refunded items never leave the building.
  • Honest packing timer: idle threshold adapts to each packer's own median pace — fast workers aren't nagged, slow runs aren't flattered.
  • "Not collected" filter surfaces parcels sitting ready-to-ship for 2+ days.
Packing bench
Packing: run strip, progress tiles, bench queue
Deliver · /returns

Returns — beat the auto-refund clock

TikTok auto-approves unanswered returns. This screen is built around never missing that deadline — then around what returns cost and who drives them.

Work QueueAnalyticsBuyersCancellations
  • Deadline colour-coding: overdue / <12h / 12–24h / 1–2d / safe — the queue sorts itself by risk.
  • Bulk approve re-checks live eligibility with TikTok one return at a time — no stale approvals.
  • Analytics: return rate (with its denominator stated), refunds vs. label cost, priority brand, margin kept — by brand and by show.
  • Buyer review: seller-side issues checked first; every observation labelled fact vs. hypothesis.
Returns work queue
Returns queue: pipeline, cost-by-reason, claims worth filing
Products · /inventory

Inventory — what you have, what it's worth

A real stock ledger: record buys as bins, attach them to listings, count what survived, and write off what didn't — with the P&L booked correctly.

  • Record a bin: total, units, landed costs — freight and fees spread into unit cost.
  • Stock tab & lot card: on-hand, value at cost, sell-through, full movement ledger per lot.
  • FIFO or weighted-average binding with a preview that says exactly how many lines get costed, from which bins, before anything is written.
  • Counts & write-offs: damaged / lost / sample book to the P&L; a miscount doesn't — no money was lost.
  • Re-cost when a supplier credit changes a purchase total.
Stock tab
Stock: lots with on-hand, value at cost, sell-through
Lot card
Lot card: depletion and the movement ledger
Products · /merchandise

Catalog — one identity per product

The master that normalizes what AI heard on the stream into clean brand and product names — so every other screen groups the same item together.

  • Brand-grouped list with filter chips: missing weight, weight suggestions, name suggestions.
  • AI weight lookup per item or as a capped sweep — but labels only ever use weights a person approved. Wrong AI answers can't reach a shipment.
  • AI name clustering can settle on the correct real-world spelling even when it appears nowhere in the raw transcripts.
  • Manual rename & merge for the cases a human settles in two seconds.
Catalog list
Catalog: brand groups, orders, missing weights
Products · /costing

Costing — COGS, with a memory

The workbench that turns sold lines into costed profit: AI identity, cost memory, bulk actions and outlier checks.

UncostedCostedMargin outliersBy listingBy identity
  • Coverage bar + profit strip always say how much of the book is costed.
  • Inline costing with keyboard shortcuts and a focus queue — built for hundreds of lines a day.
  • Cost memory applies a past cost only when it's consistent (≥80% agreement or five consecutive matches) — otherwise it says "history unclear" instead of guessing.
  • ✦ Identify transcribes the video receipt into brand/item/colour/size; fixing one brand can instantly cost a whole family of lines.
  • Margin-outlier tab flags the costs worth a second look.
Costing workbench
Costing: coverage bar, profit strip, inline cost field
Products · /performance

Performance — analytics that show their working

What sold, how fast, what came back, and whether shipping and ads actually pay — every metric with its formula and denominator stated.

OverviewCatalogSell-throughDiscountsReturnsShippingAds & Ops
  • Overview: nine KPIs, trend chart, insight cards.
  • Show quadrant: every broadcast plotted GMV vs. costed margin — Winners / Scale these / Watch / Fix margin.
  • Shipping tab: margin curve answers "does consolidating parcels pay?", with sparse data drawn honestly (hollow and thin-line) instead of smoothed over.
  • Ads & Ops: spend, ROAS, packing and label costs against sales.
Performance overview
Performance: KPI tiles and daily trend
Money · /finance

Finance — reconciliation to the cent

TikTok's settlement legs, reconciled against the books, traced from gross sales to the payout that hit the bank.

OverviewProfitabilitySettlementsCashCosts & drivers
  • First line says "Reconciled" or "Off by $X" — and expands to the full equation ending at TikTok's own settlement total.
  • Where the money went: funnel view or signed ledger, TikTok fees itemised by fee key.
  • Cash tab: committed cash over the next 30 days, reserve releases, payout timeline.
  • Data dictionary: every label mapped to the exact TikTok Finance API field — disputes argued in TikTok's own language.
Finance overview
Finance: Reconciled line and the money funnel
Funnel expanded
Fully expanded: every fee, by TikTok fee key
Money · /expenses

Expenses — forward the receipt, keep the pen

The costs no order pays for — rent, ads, packaging — captured by email, read by AI, booked by a human.

  • Private per-workspace forwarding address turns emailed receipts into a review queue; the address strip stays disabled until MX record and webhook signing are verified — receipts are never silently swallowed.
  • AI reads vendor, date and amount — and only prefills the form. A reading never writes an expense.
  • "Already logged?" detection stops the same receipt being filed twice.
  • Categories & vendors are optional, archived not deleted; defaults get a seller started in minutes.
  • Everything lands somewhere honest: the operating stage in Finance, a show tag, or a shrinkage row from Inventory.
  • Keyboard-driven queue with AI/LAST marks showing what was read, remembered, or typed.
Account · /settings

Settings — roles, devices, integrations

Team access with teeth, capture pairing, and shipping configuration — the control room behind the console.

  • Three roles — Owner, Manager, Packer — over a granular permission matrix.
  • Global "Show financial figures" switch: a packer sees Orders, Fulfillment and Packing with every money column dropped, not dashed.
  • Permission changes sign the member out everywhere, immediately — no stale sessions holding old access.
  • Signed-in device list and password management per member.
  • Video-receipt capture pairing with revocable tokens.
  • Shippo integration with explicit TEST/LIVE mode, sender address and parcel presets — label buying runs against a configured account, not a hope.
Chrome · capture extension

The capture extension — receipts from the source

A small Chrome extension that pulls video-receipt links from the seller's own signed-in Seller Center session and pairs them to shows in one click.

  • Reads no TikTok cookies. It asks the signed-in page for receipt links and sends only order id, room id, video URL and timestamp.
  • Pair once with a workspace token; green CAPTURING pill when live; tokens revocable from Settings.
  • One-click per show: select the room, press Capture receipts, watch the chip turn green — e.g. 97 of 101 orders now carry a video receipt.
  • Those receipts are what feed ✦ transcription, Costing identity, Catalog names and buyer preferences.
Extension popup connected
Connected: workspace + Seller Center both green
Captured receipts
Receipts captured: chip reports coverage per show
The through-line

Built to be trusted with someone else's books

Four design commitments show up on every screen — and they are the moat.

  • AI proposes, humans dispose. Transcription, weight lookup, receipt reading, name clustering and cost memory all produce suggestions or prefills. Nothing silently writes a cost, a weight, an expense or a name.
  • Every number shows its working. Formulas, denominators and which clock a date sits on are stated on the screen — "return rate" admits it's three different ratios in three different places.
  • Coverage is a first-class figure. % settled, % costed, coverage bars and "est." tags mean a user can never quote half-baked data as final.
  • Speaks TikTok natively. Settlement receipts, fee keys and a full data dictionary map every label to TikTok's own API field — the console argues disputes in the platform's language.
SellerFolio

From "went live"
to "made money"
with nothing in between lost.

Shows → orders → costing → fulfillment → returns → reconciliation. One console, one login, every dollar accounted for.

Live
in production today
/deck
this walkthrough
/guide
per-screen documentation
1 / 19 navigate · Home start · End close