Prepare your data and choose the right upload

1. Choose where to upload

• Transactions: a sales, refund or credit export whose columns you want to map yourself.
• Customs imports: a table of customs entry lines and their declared duty components.
• Products → Import catalog CSV: a product list. The direct Import button creates/updates products immediately; there is no sales-style mapping/Commit step. Use Inbox when you want destination review first.
• Scoutly Inbox: several business files to inspect individually, including supported PDFs, tables and product catalogs, before choosing a destination.

Check the active organization before uploading. A name or organization ID written inside a file does not switch your workspace or give access to another organization. Separate files containing records for different entities before proceeding.

Use the signed-in website for uploads and review actions. The customer API guide documents a narrower set of integration endpoints; an API key does not grant access to every action used by these screens.

Start with fictional sample files

Use these small examples to see the expected structure before touching a real export. They contain fictional records, download locally, and do not create records or consume an AI or research allowance.

Before you upload: prepare a clean spreadsheet

1. Keep an untouched source copy. Export from the shop, payment processor, accounting system or customs broker. Work on a copy, not your only record.
2. Choose a useful time period. For a first threshold review, include the complete previous calendar year and this year to date when available. Some rules need earlier history, fiscal years or other windows. Check the measurement period on each result; no fixed number of months guarantees complete coverage.
3. Use one table, one header row, one kind of record. Put headings on the first row and data underneath. No report titles, merged cells, subtotals, grand-total rows, charts, blank separator rows or multiple tables. A CSV is a plain-text spreadsheet export, not a screenshot.
4. Use values, not spreadsheet formulas. In Excel or Google Sheets, copy the relevant table to a new sheet and use Paste special → Values only. Then export CSV UTF-8 or a supported XLSX. Do not just rename an XLS, PDF or image to CSV.
5. Protect identifiers. Treat order IDs, SKUs, postal codes, entry numbers and HTS codes as text so leading zeros survive. An identifier must stay the same in later exports. Keep separate real IDs for a sale and its refund.
6. Use clear dates and amounts. Prefer 2026-01-15, 1234.56 and a separate currency column such as USD. No currency symbols, thousands separators or prose in the recommended template. Write the decimal amount, not cents: 12 dollars is 12.00, not 1200.
7. Do not fill unknown facts with guesses. Leave unknown optional values blank and inspect missing-fact notices. Use zero only for a known zero. For yes/no flags use yes or no only when supported by your records.
8. Start with a small, genuine sample. Review 10–20 real records from one export, inspect the result, then import the rest using the same source platform and unchanged identifiers. Check duplicate reasons when later exports overlap; changing IDs or source platforms can defeat identity-based matching.

If an XLSX contains several worksheets, export the intended one for the simplest experience. Direct sales/customs review can select a sheet; Inbox holds multi-sheet files for review. It does not silently combine them.

Do not upload card numbers, passwords, API keys, identity documents or unrelated personal information. Use a customer reference instead of a customer's name/email when it meets your evidence needs.

Sales and refunds: one economic event per row

Go to Transactions → Upload a sales export. This is for completed sales and separate refund events—not bank payouts, monthly totals, open quotes, inventory movements or a file listing every item with the full order total repeated.

Required to accept a row: a valid transaction date, gross amount and currency (or a deliberately chosen default currency). Needed for a useful location result: destination country and, for U.S. state results, destination state. Supply stable event identifiers even though some exports can be accepted without them.

Optional mapped fields also include discount amount, refunded amount, quantity, SKU, description, buyer tax ID, channel, product tax category, ship-to postal/city, ship-from country/region and billing location. Different headings are allowed: map them deliberately. Unmapped columns do not become recognized facts just because they are in the file.

Refund example: sale EXAMPLE-SALE-001 for 120.00, then refund EXAMPLE-REFUND-001 for -24.00, both linked to EXAMPLE-ORDER-001. Use the refunded portion, not the entire original sale, on a partial refund. Refund gross is stored negative. A cumulative “Refunded amount” on the sale row does not create a separate refund event by itself. Do not both reduce the original gross and add the same refund again.

Important location/flag behavior: missing destination fields may fall back to billing location or a chosen country default. Check that this is defensible for your sales. If a mapped Marketplace column is blank and the facilitator flag is absent, the current importer treats that row as direct/non-facilitated. If marketplace status is genuinely unknown, do not use a mapped empty name column to imply direct sales; leave the marketplace mapping unset and inspect the resulting unknown facts. Organization-level business answers can also supply missing event flags; keep those answers accurate.

Several items on one order? Ask for an order/transaction-level export. Keep item details as supporting evidence. The importer does not automatically reconstruct orders, allocate shared tax/shipping or join separate tables. ORDER_REVIEW_REQUIRED is a request to resolve that ambiguity, not an invitation to rename IDs.

Customs: keep entry, merchandise line and duty component separate

Go to Imports → Upload customs entry lines. Ask your broker for a line-level U.S. entry export rather than a scanned entry-summary PDF. An HTS code is the customs classification code recorded for the goods; importing one does not validate it.

For a base line and Chapter 99 attachment, keep the same entry number and line number, with the actual declared code on each row. Do not put a whole-line total duty on every component or duplicate the entered value as if each attachment were another shipment. Preserve AD/CVD case references and MPF/HMF fee amounts when known; unresolved scope, units or interactions can leave the calculation partial.

After Commit, check inserted merchandise lines, not only accepted source rows. Open an entry to inspect its attached components and calculation. The template's code illustrates formatting, not a recommended classification or a current duty rate.

Products: a factual catalog, not a sales ledger

Use Products → Import catalog CSV for a ready-to-import catalog. Use Scoutly Inbox to inspect a proposed product destination first. For predictable hosted uploads, keep catalogs below 3.5 MiB and 5,000 rows; direct catalog import accepts CSV, TSV or TXT, not XLSX. Export the sheet as CSV first.

What Import changes: new SKUs create products; existing SKUs can update product details and create new fact versions. Duplicate SKUs within one file keep the first occurrence and skip later rows. Check the results and ignored-column list. There is no sales-style final Commit screen or one-click catalog-batch undo; keep a copy and inspect existing SKUs before using direct Import.

Open a product to add fuller facts, sourcing profiles and evidence. Uploading the catalog does not automatically run new paid research. Starting research is a separate permission, allowance and provider-availability check.

Business information belongs in Settings, not in sales totals

• Settings → Entities: record the legal businesses involved. A company name in a sales file does not automatically route rows among legal entities. Keep entity-specific files separate and confirm the intended entity handling before combining them.
• Settings → Presence: offices, inventory, warehouses, employees and other activity, with jurisdiction and real start/end dates. A shipping destination is not physical presence; a presence fact is not an automatic legal conclusion.
• Settings → Registrations: record registrations you actually hold, including regime, jurisdiction, status and relevant dates. Do not mark yourself registered just because a threshold appears crossed.
• Settings → Organization: review establishment, reporting currency and business answers. Blanket answers can affect missing transaction flags; use them only when true across the relevant activity.

Inbox may hold organization, membership, plan, presence, registration, correction/history, research-request and support-message files for manual review. Uploading a file cannot grant permissions, change a tier, create a support conversation or overwrite legal history. Use the corresponding screen or ask support.

For PDF/text evidence, use Inbox. PDF tables are not automatically turned into ledger rows; scanned PDFs have no OCR. Obtain a CSV/XLSX from the source for calculations. After the app confirms that each file was saved, background processing can continue when you leave the page. Keep it open until uploads finish; this is not a resumable browser file transfer.

After importing: check the numbers before relying on the result

1. Check the batch: Uploaded is not Committed. Compare source rows, accepted, rejected, duplicates, skipped and finally inserted records. Read warnings even when a row was accepted.
2. Reconcile totals: on Transactions, filter to the batch and compare sales/refunds by currency with your original export. Do not sum unrelated currencies. On Imports compare merchandise lines and recorded duty with the broker's file.
3. Review locations: resolve missing destinations and confirm that billing fallback/defaults did not put sales in the wrong state or country. Confirm marketplace, exempt, resale and business facts.
4. Open Global Obligations: allow the evaluation job to finish, then open a jurisdiction. Inspect its measurement period, included sales/counts, rule version, source dates, missing facts and first-crossing history. No supported result for a location does not mean no obligation there.
5. Read the status: “approaching” is a monitoring signal; “appears crossed” means the available data meets a supported test; “insufficient information” needs more evidence. “Registered” reflects your recorded registration, not independent government verification. “Clear” is not a guarantee that every possible obligation was tested.
6. Open Findings and Reports: investigate flagged records and export an evidence packet. A projection is an estimate based on observed activity, not a promised date. Review material decisions with a qualified tax advisor or customs broker.

Customer glossary explains terms without exposing internal Ops procedures. You can always return to Documentation; reading documentation does not modify records, confirm a legal result or consume AI allowance.

2. File sizes and previews have different limits

Before uploading, use the column-by-column preparation reference above. Blank templates contain only headers; examples are fictional and must not be imported into your real business ledger.

All applicable limits must be met. A file can be below the byte limit and still have too many rows, columns, cells or unsupported workbook features.

The smaller Inbox file limit intentionally leaves room for the upload form inside its 4 MiB total request limit. It is not the same limit as a direct sales/customs file upload. The text-table parser also allows at most 100,001 records including headers, preliminary notes and blank records; those extras can reduce the usable data-row allowance. Individual cells are limited to 100,000 text characters.

The import review grid shows 25 source rows per page and keeps all parsed columns available by horizontal scrolling. Use Previous rows and Next rows to inspect more. The page size does not mean only 25 rows will be validated.

The Inbox preview is smaller: up to 8 rows and 24 columns, with cell text shortened to 120 characters. PDF previews show only a short extracted excerpt. These previews are not a complete display of the file; download the private original when reviewing omitted content.

Safe-fix suggestions scan only the first 5,000 rows. A review supports at most 500 cell changes and 512 KB of change data; accept at most 200 selected proposals at once. Larger cleanup belongs in a corrected source export that you can inspect and upload again.

Prefer a plain, values-only CSV for difficult spreadsheets. Archives, executables, HTML and SVG are not supported Inbox uploads. Renaming an unsupported file does not convert it. Workbook formulas, active content and unusually large compressed contents may cause rejection even when the file is small.

3. Review a sales or customs import step by step

1. Select a file. Drop one file onto the upload area or use the native file chooser. Check its name and size, select the source platform when offered, then choose Upload. Merely dropping a file does not upload it.
2. Wait for confirmation. Once the server has saved the file, the batch page reports parsing or other background work. An upload failure is different from a saved file that is still being processed.
3. Inspect the layout. Open Inspect file layout and headers. Check the header record, delimiter, worksheet and notices. A delimiter is the separator between cells, such as a comma, semicolon, tab or pipe. Header record numbers start at 1 among nonblank records; the control supports the first 100.
4. Correct the layout if necessary. Choose the intended sheet or separator and confirm the warning before re-reading. This rebuilds the parsed rows from the original file and clears cell changes, proposals and column mapping. Review them again.
5. Map the columns. Mapping means matching a file heading, such as “Total,” to a field such as “Gross amount.” Compare it with net amount, tax, shipping and refunds before choosing. A suggested or similarly named column is not proof that the meaning matches.
6. Choose formats and inspect changes. Set the date order and decimal convention below. Review proposals and any manual corrections. Confirm ambiguous or Scoutly column matches explicitly.
7. Validate rows. Validation checks the current reviewed values and marks accepted rows, rejected rows and duplicates with reasons. Read warnings and use the error CSV to investigate rows that will be left out.
8. Commit only after checking the result. Commit adds eligible records. Rejected rows and identified duplicates are skipped. Wait for the final result and compare accepted, inserted and skipped counts. An accepted-row count is not always the number of distinct customs merchandise lines.

Changing a cell or layout makes the earlier validation stale. Validate again before committing. If another tab changes the review, reload and inspect the latest values instead of repeatedly submitting an old selection.

4. Make number and date meanings explicit

Separators mean different things in different exports. For example, 1,234 can mean one thousand two hundred thirty-four or a number slightly larger than one. Select the convention used by the source, not the convention you hope the file uses.

For dates, 03/04/2026 means March 4 with month/day/year, or 3 April with day/month/year. Choose the export’s date order explicitly. A date such as 2026-04-03 uses year-month-day and is easier to check. Invalid dates and prose such as “early April” need correction; the importer does not invent a calendar date.

Preserve currency codes, transaction kinds and refund/credit meaning. A currency symbol is not enough to establish which country’s dollars an amount uses. Set a default currency or country only if it is correct for the rows where that field is blank. Missing values do not automatically mean zero, false or “not taxable.”

If the export mixes decimal or date conventions, split it or correct the inconsistent cells from reliable source records. Do not remove arbitrary words from a financial cell just to make it pass validation.

5. Keep the original, then accept or change only what you understand

The original file is not rewritten by a review correction. Within the selected layout, original parsed cell values are kept separately from the values used after you accept changes. The grid shows Original and Current for a changed cell, with a label and border as well as the reason.

• Suggest safe fixes: proposes supported whitespace, capitalization and number/date formatting changes. This tool does not use AI and does not apply proposals by itself. Check the selected format and both values before accepting.
• Accept: applies that proposal to the review. Accept selected applies only the checked proposals. A proposal still labelled Proposed has not changed the value used for import.
• Discard proposal: removes an unaccepted proposal. Revert removes an applied cell change so the original value is used again. Neither action deletes the original upload.
• Edit: opens a text field for the exact value you intend. Save cell edit records a manual correction; Cancel edit leaves it unchanged. Each edit is limited to 2,000 characters. Saving an edit does not commit a ledger record.

Example: a cell containing usd may be proposed as USD because the supplied code has surrounding spaces and lowercase letters. That proposal does not establish that USD is the correct currency for the transaction.

A layout reset is different from reverting one cell: it discards the current review and mapping and parses the original again. A cell’s Revert button is also different from reversing a committed batch. Committed batches are read-only in cell review; the existing sales-batch reversal control, when available, reverses committed events rather than erasing their history. Customs batch reversal is not currently offered.

“Original unchanged” does not mean “stored forever.” Keep your own source copy. Inbox originals have a displayed expiry and can be explicitly expired; ordinary import source-file retention is also limited. Re-reading a layout requires the original to remain available.

6. A row, an item and a duplicate are not the same thing

Start by deciding what one row represents in your export. Row totals, order counts, quantities, accepted records and customs components answer different questions.

Identical file

The content fingerprint can identify an upload already present in the relevant workflow. Reopening that file is not evidence that a second set of records was created. Changing only its filename does not change its content.

Already committed / duplicates within the file

Import validation compares record identities, not only how the rows look. Sales checks use source transaction identifiers and SKU when supplied, or a combination of mapped order/date/kind/amount/currency/SKU/location fields. Read the reason before changing identifiers: two different business events can look similar, and an overlapping export can look different while containing an existing record.

Order versus sale item

One order may have several item rows. A row with quantity 3 is not automatically three transactions. Repeating a whole order total on every item can overstate activity. The importer holds ambiguous repeated order/transaction rows for review, including multiple SKUs for the same order or a repeated financial amount on different rows. A “line total” heading or different item IDs does not by itself establish independent transaction meaning.

Merchandise line versus customs component

A merchandise line can have a base tariff row and additional duty/component rows, such as Chapter 99. Preserve entry number, line number, code and component kind. Supported customs processing groups related accepted rows into merchandise lines with declared components. Two accepted source rows can therefore produce one inserted merchandise line. The Inbox does not perform that grouping during classification.

If you see ORDER_REVIEW_REQUIRED: the batch cannot be committed yet. Prepare a reviewed transaction-level export with one transaction/order per row, using genuine source identifiers and event-level amounts. Keep the item-level detail as supporting evidence. The app does not automatically join item rows, aggregate them into an order total, allocate amounts or invent identifiers. Exact repeated source rows still follow the duplicate checks; an ambiguity hold is a separate issue.

Do not delete component or item rows simply because an entry or order number repeats, and do not change identifiers merely to bypass a duplicate warning. Compare the original source, mapping, reasons and final inserted/skipped counts. The current import screen’s record counts are not a certification of filing quantities or legal transaction-count rules.

7. Use Scoutly Inbox to inspect destinations

Open Scoutly Inbox, read the file-count and retention limits shown for your organization, then select files. Files upload one by one. Check each result: one successful upload does not mean every selected file was saved.

Keep the page open until upload confirmation. Queued processing happens on the server after a file has been saved, so it can continue when you leave the page. An unfinished transfer still depends on the browser. Returning later lets you inspect the status; it does not approve a destination automatically. Background work can still fail or be stopped by permission, account or resource limits.

Read Evidence, coverage and counterchecks, including reasons against the proposed destination and missing information. The available destinations have different effects:

• Sales, refunds and credits / Customs import review: confirmation sends a supported table to a new or existing import batch. Continue column mapping and validation, then separately confirm the ledger commit.
• Product catalog: confirmation can create new SKUs and update supported fields on matching products. Inspect the catalog preview, mapped/ignored columns and update notice first. This does not have the sales/customs ledger-commit step.
• Supporting documents: keeps the original as private evidence in the Inbox until its displayed expiry. It does not create sales, customs or catalog records.
• Unknown / missing information: means the evidence does not support a confident destination. Unknown or competing table signatures are held for manual review. Keep them as supporting evidence or upload a clarified table; choosing a destination does not override a review-only hold.

File text, formulas, links and apparent instructions are treated as data. They do not grant the file permission to run code, visit its links, switch organizations or change access.

8. A readable preview may still be partial

Inbox PDF extraction reads available text; it does not reliably turn a PDF table into sales, customs or product records. It inspects at most the first 30 pages and stops at 100,000 characters or the 20-second extraction limit. These are processing ceilings, not a promise that every page below them is readable.

Scanned or image-only PDFs have no OCR. OCR means recognizing text from a picture; that is not performed here. Blank pages, unsupported features, missing character resources and damaged files can also leave text unavailable. Encrypted/password-protected PDFs are unsupported. Form values and attachments are not fully extracted.

“Pages inspected” is a coverage count, not necessarily the total document length or proof that every item on those pages was read. A partially read last page can be included. Partial coverage and notices identify omitted pages or text; if extraction fails or times out, do not assume AI has read the PDF.

Text extraction is also capped at 100,000 characters. AI receives a smaller bounded sample, not every extracted character, row or column. Read the reported sample and missing facts. A displayed excerpt, an AI opinion and a full-file review are different levels of coverage.

For complete manual review, download the original. For structured importing, obtain a values-only CSV/XLSX export from the source system, or carefully prepare a supported table and verify it against the original. An unreadable PDF is not automatically an empty or zero-value business record.

9. Optional AI needs permission and has its own limits

In an import batch, Suggest with Scoutly queues a request for column-mapping proposals. The screen refreshes while it runs. In the Inbox, approved optional AI can run as part of background inspection: it proposes a category and separately checks that opinion for contradictions. AI does not accept its own proposals or confirm placement.

AI access requires a valid signed-in session with write permission, the required verified Google sign-in, an eligible account, and acknowledgment of the current processing notice on the Scoutly page. Service-owner deployment approval, the approved processing connection, available plan allowance and budget checks must also pass. An account upgrade alone does not bypass these checks.

The import screen shows the current sample-row allowance and a reason when AI is unavailable. Import mapping assistance sees bounded headers and masked value shapes. Inbox assistance can receive bounded extracted text, including content from readable PDFs; it does not receive a guarantee of complete coverage. Review the processing notice and avoid uploading secrets or unnecessary personal data.

AI is metered work. File-assistance attempts have monthly allowances and organization/shared provider-budget limits. Inbox AI includes a separately budgeted second opinion. An attempt that starts can use allowance or incur provider work even if no usable proposal is returned. Upload counts, committed-record usage and AI attempts are separate measures; this guide does not promise a fixed per-file price, unlimited retries or free AI on every file.

Queued means waiting or running, not approved or complete. Check the final status, sample coverage, missing information and reasons. A confidence label is a qualitative opinion, not a guaranteed probability. If models disagree, the destination can remain unknown. Failure does not authorize a switch to another provider or an automatic business-record change.

If AI is unavailable, use the original file, header checks, layout controls, manual mapping, safe-fix proposals and cell editing. AI approval is not required for those manual import tools, although normal account permissions and import capacity still apply.

10. If something does not work

File too large or too many cells

Split the export into meaningful parts below the applicable byte and row limits, keeping headers and stable identifiers in each part. For workbooks, export the relevant sheet as static values. A small compressed XLSX can still exceed workbook safety limits.

Columns are shifted or numbers/dates are rejected

Inspect the separator, header record and selected sheet. Check the source’s decimal/date convention, required fields and each rejection reason. UTF-8 CSV is a useful common export format; Inbox text requires valid UTF-8. Do not guess missing financial facts.

Rows are missing from the preview

Check whether you are looking at the 25-row import page, a status-filtered view, the 8-row/24-column Inbox preview or partial extracted text. Scroll, change pages, clear the filter or open the original as appropriate.

An edit conflicts or Commit is unavailable

Reload the latest review, compare your draft with the current values and validate again. Read-only access cannot upload, accept changes or confirm placement. An already committed batch cannot be edited through cell review.

A job stays queued or reports failure

Use Refresh or reopen the saved file/batch. Check its actual status before uploading or confirming again. A request can reach the server even if your browser loses the response. A stuck product-placement attempt needs reconciliation before repeating a potentially completed update.

The original expired or cannot be read

Use your retained source copy and prepare a new supported export if needed. Expiring an Inbox original removes that Inbox copy and preview; it does not reverse records or delete independently retained destination copies.

When you need help, use Contact support with the batch/file identifier and the displayed error. Do not include passwords or API secrets. Sending a support conversation is a separate action; a file named “support tickets” does not create one.

This is a research estimate and not a legal, tax, or customs determination. It's a good idea to review the result with your accountant, tax advisor, or other qualified professional before registering, filing, or changing tax treatment. It's a good idea to have your customs broker review the classification before acting.

Read the calculation and source methodology for what happens after records are imported.