The warehouse export puts your raw experiment data into your own Google Cloud Storage bucket every night. From there you can load it into BigQuery, Snowflake or any tool that reads JSON files, and join it with your own data: orders, CRM records, product usage and so on.
It is available on all paid plans.
What you get
Every night, shortly after midnight UTC, Mida adds the previous day's data for your project. There are four tables:
Table | One row per | Use it for |
exposures | Visitor put into a variant of a test, the first time it happens | Who saw which variant, and when |
events | Goal conversion or event you track (purchases, sign-ups, your own events) | What visitors did after they saw a variant |
results_daily | Variant of each live test that had visitors in the last 30 days, as the report shows it that day | The same totals as your Mida report, day by day |
tests | Test in the project, with its variants and goals | Test, variant and goal names for your queries |
Exposures and events share the same visitor ID, so you can join them. Page views and engagement pings are left out of events, to keep the files small. Goal conversions are included as events named goal_ followed by the goal ID.
Before you start
A paid Mida plan.
A Google Cloud Storage bucket. If you load the data into BigQuery, create the bucket in the same location as your BigQuery dataset (for example US or EU).
Permission to change the bucket's access in Google Cloud.
Owner or Admin role in Mida.
Set it up
In Google Cloud
Open Cloud Storage, then Buckets, and create a bucket or pick an existing one.
Open the bucket, choose Permissions, then Grant access.
Add the Mida service account shown on the setup screen in Mida, and give it the Storage Object User role. Save.
In Mida
Open Integrations and choose BigQuery or Snowflake under Data Warehouse. Both set up the same export.
Enter the bucket name and, if you like, a folder inside it.
Click Check access and turn on. Mida writes a small test file (
mida/v1/_manifests/connection_check.json) into the bucket first. If it cannot, it tells you what is missing and nothing is turned on.
Mida only writes into the bucket. It does not read or delete your other files.
When the data arrives
The first delivery arrives shortly after midnight UTC, with the data from the moment you turned the export on until the end of that day.
The first delivery also includes everyone who was already in your live tests (those that had visitors in the last 30 days), so a test that was running before you turned the export on can still be analyzed in full. Events from before that day are not included.
After that, each day arrives the following night.
If a delivery fails, for example because access was removed, Mida tries again every hour and keeps up to 7 days of data while it does. The setup screen shows the error.
Where the files are
One folder per table, one folder per day, one file per project:
gs://your-bucket/your-folder/mida/v1/exposures/dt=2026-11-18/project-1234.ndjson.gz
gs://your-bucket/your-folder/mida/v1/events/dt=2026-11-18/project-1234.ndjson.gz
gs://your-bucket/your-folder/mida/v1/results_daily/dt=2026-11-18/project-1234.ndjson.gz
gs://your-bucket/your-folder/mida/v1/tests/dt=2026-11-18/project-1234.ndjson.gz
gs://your-bucket/your-folder/mida/v1/_manifests/dt=2026-11-18/project-1234.json
Files are gzipped JSON, one row per line. dt is the day Mida delivered the data for. You can send several projects to the same bucket and folder; each has its own file.
The manifest file lists how many rows each table has for that day. When it is there, that day is complete.
Columns
exposures
Column | Meaning |
| Your Mida project ID |
| The test ID, as shown in Mida |
|
|
| Control or the variant name |
| Mida's visitor ID |
| Another ID Mida has for this visitor, for example the user ID your server-side SDK sends. Empty when there is none. |
| When the visitor was first put into the variant (UTC) |
|
|
events
Column | Meaning |
| Your Mida project ID |
| The event, for example |
| Same as in exposures |
| When Mida received the event (UTC) |
| The page, when the event came with one |
| The event's properties as JSON text, for example revenue and currency |
results_daily
One row per variant of each live test that had visitors in the last 30 days: result_date, experiment_id, experiment_name, goal_id, goal_name (the primary goal), variant_id, variant_name, is_control, visitors, conversions, conversion_rate (percent), revenue, orders, currency and stats_engine. The numbers are totals since the test started, as your report showed them that night.
tests
One row per test that is live, paused or completed (drafts are left out): experiment_id, name, url, type, status, stats_engine, created_at, completed_at, a list of variants (ID, name, traffic weight) and a list of goals (ID, name, whether it is the primary goal, and the event_name to look for in events). Metrics Mida calculates itself, such as bounce rate and time on site, have the type built_in and no event name, because they come from page views, which the events table leaves out.
Load it into BigQuery
The quickest way is external tables: BigQuery reads the files where they are, and new days show up on their own. Run this once in the BigQuery console:
-- Replace your-bucket/your-folder with your bucket and folder (leave out the folder if you have none).
CREATE SCHEMA IF NOT EXISTS mida;
CREATE OR REPLACE EXTERNAL TABLE mida.exposures
WITH PARTITION COLUMNS
OPTIONS (
format = 'NEWLINE_DELIMITED_JSON',
compression = 'GZIP',
uris = ['gs://your-bucket/your-folder/mida/v1/exposures/*'],
hive_partition_uri_prefix = 'gs://your-bucket/your-folder/mida/v1/exposures'
);
CREATE OR REPLACE EXTERNAL TABLE mida.events
WITH PARTITION COLUMNS
OPTIONS (
format = 'NEWLINE_DELIMITED_JSON',
compression = 'GZIP',
uris = ['gs://your-bucket/your-folder/mida/v1/events/*'],
hive_partition_uri_prefix = 'gs://your-bucket/your-folder/mida/v1/events'
);
CREATE OR REPLACE EXTERNAL TABLE mida.results_daily
WITH PARTITION COLUMNS
OPTIONS (
format = 'NEWLINE_DELIMITED_JSON',
compression = 'GZIP',
uris = ['gs://your-bucket/your-folder/mida/v1/results_daily/*'],
hive_partition_uri_prefix = 'gs://your-bucket/your-folder/mida/v1/results_daily'
);
CREATE OR REPLACE EXTERNAL TABLE mida.tests
WITH PARTITION COLUMNS
OPTIONS (
format = 'NEWLINE_DELIMITED_JSON',
compression = 'GZIP',
uris = ['gs://your-bucket/your-folder/mida/v1/tests/*'],
hive_partition_uri_prefix = 'gs://your-bucket/your-folder/mida/v1/tests'
);
Each table gets a dt column from the folder names. Filter on it (for example WHERE dt >= '2026-11-01') so BigQuery only reads the days you need.
Example: conversion rate per variant
-- Visitors and conversions per variant for one test, counting a conversion only
-- when it happened after the visitor first saw the test.
DECLARE test_id INT64 DEFAULT 12345; -- your test ID
DECLARE goal_event STRING DEFAULT 'goal_678'; -- or your own event, such as 'Purchase'
SELECT
x.variant_name,
COUNT(DISTINCT x.visitor_id) AS visitors,
COUNT(DISTINCT IF(e.event_time >= x.exposed_at, x.visitor_id, NULL)) AS converted,
SAFE_DIVIDE(COUNT(DISTINCT IF(e.event_time >= x.exposed_at, x.visitor_id, NULL)),
COUNT(DISTINCT x.visitor_id)) AS conversion_rate
FROM mida.exposures AS x
LEFT JOIN mida.events AS e
ON e.project_id = x.project_id
AND e.visitor_id = x.visitor_id
AND e.event_name = goal_event
WHERE x.experiment_id = test_id
GROUP BY x.variant_name
ORDER BY x.variant_name;
Find a goal's event name in the goals list of the tests table, or use the name of your own event.
-- Event properties are a JSON text column. Read a value with JSON_VALUE.
SELECT event_time, visitor_id, CAST(JSON_VALUE(properties, '$.revenue') AS NUMERIC) AS revenue
FROM mida.events
WHERE dt >= '2026-11-01' AND event_name = 'Purchase';
If you query the data often, copy it into regular BigQuery tables on a schedule (for example with a scheduled query) to make queries faster and cheaper.
Load it into Snowflake
Snowflake reads the files from the bucket through a stage. Run this once:
-- 1. Let Snowflake read the bucket.
CREATE STORAGE INTEGRATION mida_gcs
TYPE = EXTERNAL_STAGE
STORAGE_PROVIDER = 'GCS'
ENABLED = TRUE
STORAGE_ALLOWED_LOCATIONS = ('gcs://your-bucket/your-folder/mida/v1/');
DESC STORAGE INTEGRATION mida_gcs;
-- Give the account shown as STORAGE_GCP_SERVICE_ACCOUNT the
-- Storage Object Viewer role on your bucket in Google Cloud.
-- 2. Point a stage at Mida's folder.
CREATE STAGE mida_stage
URL = 'gcs://your-bucket/your-folder/mida/v1/'
STORAGE_INTEGRATION = mida_gcs
FILE_FORMAT = (TYPE = JSON COMPRESSION = GZIP);
-- 3. Tables.
CREATE TABLE mida_exposures (
project_id NUMBER, experiment_id NUMBER, variant_id STRING, variant_name STRING,
visitor_id STRING, user_id STRING, exposed_at TIMESTAMP_TZ, source STRING
);
CREATE TABLE mida_events (
project_id NUMBER, event_name STRING, visitor_id STRING, user_id STRING,
event_time TIMESTAMP_TZ, url STRING, properties STRING
);
CREATE TABLE mida_results_daily (
result_date DATE, project_id NUMBER, experiment_id NUMBER, experiment_name STRING,
goal_id NUMBER, goal_name STRING, variant_id STRING, variant_name STRING, is_control BOOLEAN,
visitors NUMBER, conversions NUMBER, conversion_rate FLOAT, revenue FLOAT, orders NUMBER,
currency STRING, stats_engine STRING, computed_at TIMESTAMP_TZ
);
CREATE TABLE mida_tests (
snapshot_date DATE, project_id NUMBER, experiment_id NUMBER, name STRING, url STRING,
type STRING, status STRING, stats_engine STRING, is_bandit BOOLEAN, traffic_allocation NUMBER,
serving_variant_id STRING, created_at TIMESTAMP_TZ, completed_at TIMESTAMP_TZ,
variants VARIANT, goals VARIANT
);
Then load the files every day:
-- Run every day (for example as a Snowflake task after 01:00 UTC).
-- COPY remembers which files it has loaded, so each run only adds new days.
COPY INTO mida_exposures FROM @mida_stage/exposures/ MATCH_BY_COLUMN_NAME = CASE_INSENSITIVE;
COPY INTO mida_events FROM @mida_stage/events/ MATCH_BY_COLUMN_NAME = CASE_INSENSITIVE;
COPY INTO mida_results_daily FROM @mida_stage/results_daily/ MATCH_BY_COLUMN_NAME = CASE_INSENSITIVE;
COPY INTO mida_tests FROM @mida_stage/tests/ MATCH_BY_COLUMN_NAME = CASE_INSENSITIVE;
Things to know
A visitor appears in exposures once per test: the first time they were put into a variant. When they come back, they see the same variant, so there is no new row.
Results you calculate yourself may differ a little from the Mida report. The report also removes bots it detects later, applies the filters you set and counts revenue goals in its own way. results_daily has the report's own numbers.
Mida removes visitors it identifies as bots from exposures before delivery.
All times are in UTC.
Visitor IDs are Mida's own. When Mida has another ID for the visitor, for example from the server-side SDK, it is in
user_id.Turning the export off stops it straight away. Turning it on again starts a new collection from that moment. The data from the days in between is not added later.
Turn it off
Open the BigQuery or Snowflake card in Integrations and click Turn off export. Mida stops collecting at once and sends nothing more. Files already in your bucket stay there. You can also remove the Mida service account from the bucket.
Troubleshooting
What you see | What to check |
"Mida does not have permission to write" | The service account shown in Mida needs the Storage Object User role on that bucket. A role on a different bucket or project does not count. |
"The bucket was not found" | Check the bucket name. Enter just the name, for example |
No files yet | The first files arrive shortly after midnight UTC on the night after you turn the export on. |
A day has no events file | No events were tracked that day, apart from page views. The manifest still lists the day with 0 rows. |
BigQuery says the location does not match | Create the BigQuery dataset in the same location as the bucket. |
FAQ
Can I export to Amazon S3 or Azure?
Not yet. The export goes to Google Cloud Storage. Snowflake, Databricks and most warehouses can read from there.
Can I get data from before I turned the export on?
Only the visitors already in your live tests, in the first delivery. For anything older, use the CSV export on each test report.
Does it slow down my site?
No. The export runs on Mida's servers. Nothing changes on your site or in the Mida tag.
