Offline Conversions from Google Sheets to Meta: Columns, Hashing and Deduplication

Plenty of shops log part of their sales in a Google Sheet: orders taken by phone, sales in a physical shop, or business orders paid by bank transfer. Someone adds a row and, once the money arrives, marks it "paid". Meta never hears about it, because the Pixel sees your website, not your spreadsheet. The Conversions API lets you pass those changes to Meta as offline conversions, meaning events that happened away from your site. This guide shows how to lay out the columns, how the data is hashed, how double counting is avoided, and where this approach stops working.

What are offline conversions in Meta?

They're events your business knows about and the browser doesn't: a phone sale, a purchase in a physical shop, an order paid by bank transfer a few days later. Meta now takes them through the same Conversions API as website events. It describes its older, separate offline conversions interface as legacy and recommends the Conversions API for new integrations.

An event from a sheet uses action_source = system_generated, i.e. raised by a business system. The key rules from Meta's server event parameters:

• event_time is when the event really happened, at most 7 days before sending
• one event that's too old makes Meta reject the whole request
• each event needs at least one customer identifier

Which columns does the sheet need?

Column names are up to you, because you map them in the panel. What matters is the content.

The minimum is record ID, status, time, consent and at least one person identifier: email or phone.

Why does the status change time matter so much?

Because Meta ties a conversion to the moment it happened. If a tool stamped events with the sync time, a Friday sale would look like a Monday sale. So Basketly never substitutes the read time for a missing date: a row without one is rejected, and the message points to the first failing row number without showing any customer data.

Date cells are read in the sheet's own time zone. Text dates must include a zone, e.g. 2026-04-14T10:30:00-05:00.

How is data from the sheet hashed?

Meta requires email and phone to be normalised and SHA-256 hashed. Following Meta's customer information parameters:

1. Email: trim spaces, lowercase, then SHA-256. Jane.Doe@Example.com and jane.doe@example.com produce the same hash.
2. Phone: strip spaces, dashes and brackets; the country code is mandatory; hash the digits only. +1 (415) 555-0123 becomes the hash of 14155550123.

Basketly does this on the server. A phone number without a country code is rejected rather than guessed, because a wrong prefix produces a hash that matches nobody. Hashes are still personal data under laws like the GDPR; see GDPR and hashing in the Conversions API (not legal advice).

How do you map statuses to Meta events?

A flow defines which source status becomes which Meta event. For example:

Pick event names that match what you'll later use in a campaign or custom conversion. One source can feed several destinations (two Pixels, say), and each destination has its own delivery status.

How does deduplication work with a sheet?

A sheet read every minute means the same row is seen hundreds of times. If every read sent an event, Meta would get a flood of duplicates. There are two layers of protection:

• Business event: a record ID plus status from a source is a one-off conversion. Re-entering the same status doesn't create a new one.
• Delivery: each event has a stable event_id derived from the record ID and status, and the queue won't send the same event to the same Meta dataset in the same mode twice, even from two different connections.

Retries keep the same event_id, so if Meta's response is lost in transit, Meta can still deduplicate the repeat. The Pixel never sends sheet events, so there's nothing to reconcile with the browser. More in Pixel and CAPI event deduplication.

What can't a sync reconstruct?

A sheet shows a row's current state, not its history. If someone moves a row from RECEIVED to PAID within a minute, a read might only see the second status. If statuses can change faster than the sheet is read, log events in a separate tab: one row per change, each with its own ID.

How do you check that events arrive?

1. Create a Meta destination in test mode with the code from Events Manager.
2. Change the status of one row containing test data.
3. Check the delivery in the panel: "Accepted by Meta" means a response with events_received: 1.
4. In Events Manager's test events tab, confirm the event arrived with the right name and parameters.
5. Only then create a production destination and flow. Test and production keep separate deduplication.

Acceptance says nothing about how well Meta matched the event to a person. Events Manager shows that as Event Match Quality; see how to improve Event Match Quality.

Common errors and their causes

When is a spreadsheet the wrong source?

• You update statuses once a week. Some events will pass the 7-day limit and be lost. The sheet has to be kept current.
• Your orders come from an online store. Then it's better to connect the shop itself (Shoper, IdoSell, BaseLinker): Basketly stores the order with its lines, the buyer and the purchase without copying anything into a sheet. How that works is covered in server-side purchase events for online stores.
• You have tens of thousands of active rows. At 500 rows per minute a full pass takes a while; a server-side integration suits that volume better.
• The sheet has no person identifier at all. Without email or phone, Meta has nothing to match against.

Next step

Add the columns from the table, connect the sheet in test mode and change one row's status. Read more about the sheet as a source on the Integrations page; plans are on the pricing page.