Connecting ServiceTitan to Snowflake
ServiceTitan is the system of record for mid-to-large home services operations, and it knows the revenue on every job. Snowflake's distinguishing feature is that two organisations can join data without either copying it to the other, which turns measurement between partners from a legal negotiation into a query. The join between them succeeds or fails on one thing: whether whatever join key both parties agree on, usually a hashed identifier reaches the ServiceTitan job before the conversion happens, because Snowflake can only match a conversion it can tie back to a click it recognizes.
Free30 minutesNo obligation
Specific to this pair
ServiceTitan in Snowflake: jobs, invoices and spend in one model
Load ServiceTitan customers, locations, jobs, appointments, invoices, estimates and calls into Snowflake via a warehouse export or scheduled API pulls, alongside ad spend. Merge invoices on invoice ID with the modified timestamp instead of appending, so adjustments don't inflate revenue, and keep raw payloads. Fix one revenue-date definition and check API time zones before grouping by day.
Snowflake is a natural home for ServiceTitan data at a multi-location operator, because the questions owners actually ask cross systems: cost per invoiced job by campaign, by business unit, by branch, against ad spend pulled from four platforms. Check whether your ServiceTitan plan includes a warehouse export or data-sharing option; otherwise load from the API on a schedule into raw tables for customers, locations, jobs, appointments, invoices, estimates and calls.
Model the grain carefully. A job has many appointments and can have more than one invoice, and invoices get adjusted after the fact when a technician adds a part or the office issues a credit. Load invoices keyed on invoice ID with the modified timestamp and merge rather than append, or a single adjusted invoice shows up twice and revenue drifts upward month over month. Keep the raw API payloads alongside the modeled tables so you can rebuild when definitions change.
Most disagreements with ServiceTitan's own reports come from definitions, not data. ServiceTitan reports can count revenue by completion date, invoice date or business unit of the job versus the invoice. Pick one, write it into the model, and check which time zone the API timestamps use before grouping by day, or late-evening jobs land on the wrong date.
ServiceTitan → Snowflake · what breaks
Why this join usually fails — ServiceTitan → Snowflake
- 01
Whatever join key both parties agree on, usually a hashed identifier never reaches ServiceTitan
The identifier has to reach the customer record, not just the call. ServiceTitan's own call tracking captures the call; connecting it to the job that eventually invoices is the step that gets missed.
- 02
Rich data that never leaves the building
ServiceTitan knows the job revenue and its own attribution reporting is good — and it sends none of it to your ad platforms, so bidding never learns which jobs actually paid.
- 03
A clean room that answers a question nobody agreed on
The technology makes the join possible; it does not make the definitions match. Two organisations measuring the same campaign will disagree about what a conversion is, which window applies and whose attribution model governs — and the query will return a confident number for whichever set of assumptions got written into the SQL.
- 04
Nothing reconciles afterwards
Snowflake will not tell you anything is wrong. Verify by querying against each party's own totals and checking for duplicate rows on the join key before trusting any aggregate.
ServiceTitan → Snowflake · the fix
What working looks like — ServiceTitan → Snowflake
Whatever join key both parties agree on, usually a hashed identifier on every job
Conversions arriving in Snowflake
The row lands in a table you own, joinable to your own data and — through a share or clean room — to a partner's, without either party moving raw records.
Reconciled against the source
ServiceTitan → Snowflake · architecture
How ServiceTitan and Snowflake actually connect
- 1
ServiceTitan holds the truth
ServiceTitan distinguishes booked, dispatched, completed and invoiced. Only the last is money — a dispatched truck that ended in a no-charge callback cost you and earned nothing.
- 2
Snowflake needs a click to match
A schema both sides agree on, consistent types, and identifiers normalised identically before hashing. The failure mode is not rejection but silent divergence — a timestamp in the wrong zone or a trimmed-versus-untrimmed email produces answers that look plausible.
- 3
The identifier is the bridge
Whatever join key both parties agree on, usually a hashed identifier, captured at the visit and carried into the ServiceTitan job.
- 4
The event source decides reliability
Poll or subscribe for invoiced jobs and send with the invoiced value, job type and business unit. One job with three visits is one conversion, not three.
- 5
Deduplication at the destination
Nothing deduplicates for you. Loads are at-least-once, so a retried pipeline creates duplicate rows and the only protection is a deduplication step you write and actually schedule.
ServiceTitan → Snowflake · normal or broken
Working vs. broken, side by side — ServiceTitan → Snowflake
Normal, not a fault
Whatever join key both parties agree on, usually a hashed identifier present on the job
Sent from the ServiceTitan API on job invoiced
Deduplicated correctly
Nothing deduplicates for you. Loads are at-least-once, so a retried pipeline creates duplicate rows and the only protection is a deduplication step you write and actually schedule.
Reconciles with the source
Actually broken
Whatever join key both parties agree on, usually a hashed identifier missing on most records
The identifier has to reach the customer record, not just the call. ServiceTitan's own call tracking captures the call; connecting it to the job that eventually invoices is the step that gets missed.
Sent from the browser
ServiceTitan knows the job revenue and its own attribution reporting is good — and it sends none of it to your ad platforms, so bidding never learns which jobs actually paid.
Counted twice, or not at all
The technology makes the join possible; it does not make the definitions match. Two organisations measuring the same campaign will disagree about what a conversion is, which window applies and whose attribution model governs — and the query will return a confident number for whichever set of assumptions got written into the SQL.
Nobody has reconciled it
ServiceTitan → Snowflake · diagnose
Diagnostic order — ServiceTitan → Snowflake
- 01
Count jobs carrying whatever join key both parties agree on, usually a hashed identifier
- 02
Follow one real conversion end to end
- 03
Open a recent invoiced job and look for the identifier
On the customer record. This one check is the whole build in miniature.
- 04
Write down both parties' conversion definitions before writing any query
Definition mismatch, not technology, is what makes most partner measurement projects produce numbers neither side trusts.
- 05
Read the match rate, not the success message
Snowflake will not tell you anything is wrong. Verify by querying against each party's own totals and checking for duplicate rows on the join key before trusting any aggregate.
ServiceTitan → Snowflake · prove
How it gets proven — ServiceTitan → Snowflake
Source of truth
ServiceTitan's completed and invoiced jobs by completion date, at their invoiced values.
- Whatever join key both parties agree on, usually a hashed identifier confirmed present on a live job.
- Row counts and revenue reconciled against each source system for a closed window, with duplicates checked on the join key and both definitions documented.
- A multi-visit job confirmed to produce exactly one conversion.
ServiceTitan → Snowflake · free audit
What we check in the free audit
- Count jobs carrying whatever join key both parties agree on, usually a hashed identifier
- Follow one real conversion end to end
- Open a recent invoiced job and look for the identifier
- Write down both parties' conversion definitions before writing any query
- Read the match rate, not the success message
Checked against
ServiceTitan's completed and invoiced jobs by completion date, at their invoiced values.
Free30 minutesNo obligation
ServiceTitan → Snowflake · when not to buy
When you don't need this — ServiceTitan → Snowflake
- Measure identifier coverage yourself first — pull your last hundred jobs and count how many carry whatever join key both parties agree on, usually a hashed identifier.
- If your business units and job types are applied inconsistently, fix that first — every conversion value depends on them.
- If you have no partner to share data with and one cloud, a warehouse you already run does the same job. The sharing capability is what justifies this one.
ServiceTitan → Snowflake · what we deliver
What gets built — ServiceTitan → Snowflake
Identifier capture into ServiceTitan
Store the click identifier in a custom field on the ServiceTitan customer record, so repeat jobs from the same customer trace back to the original acquisition.
Event source: the ServiceTitan API on job invoiced
Poll or subscribe for invoiced jobs and send with the invoiced value, job type and business unit. One job with three visits is one conversion, not three.
Dispatch to Snowflake
Batch or streaming loads from your order system, with the raw event retained alongside anything you derive from it.
Reversals where they exist
Warranty callbacks and credits should not count as new revenue, and excluding them keeps the number honest.
Written guarantee
95%+ tracking accuracy
Measured against ServiceTitan's completed and invoiced jobs by completion date, at their invoiced values.. Miss it and we keep working at no cost.
From $400 · See packages
FAQ
ServiceTitan → Snowflake questions, answered
Check whether your ServiceTitan plan includes a data export or share. If not, pull from the API on a schedule into raw tables and merge on record ID and modified date.
Usually a definition difference: completion date versus invoice date, job versus invoice business unit, or adjusted invoices loaded twice. Align the definition before checking the pipeline.
Almost always because whatever join key both parties agree on, usually a hashed identifier never reached the ServiceTitan job. The identifier has to reach the customer record, not just the call. ServiceTitan's own call tracking captures the call; connecting it to the job that eventually invoices is the step that gets missed. Measure coverage before debugging the dispatch.
There is no vendor identifier. In a data share or clean room the key is negotiated: hashed email, order ID, a partner-supplied ID. Normalisation before hashing is the entire match rate, and a formatting difference between two organisations looks exactly like a poor-performing partnership.
The ServiceTitan API on job invoiced. Poll or subscribe for invoiced jobs and send with the invoiced value, job type and business unit. One job with three visits is one conversion, not three.
Snowflake will not tell you anything is wrong. Verify by querying against each party's own totals and checking for duplicate rows on the join key before trusting any aggregate.
More on ServiceTitan → Snowflake
- ServiceTitan → Google AdsConnecting ServiceTitan to Google Ads
- ServiceTitan → MetaConnecting ServiceTitan to Meta Conversions API
- ServiceTitan → GA4Connecting ServiceTitan to GA4
Using ServiceTitan for bookings? How ServiceTitan bookings get tracked end to end
RELATED GUIDES
Where ServiceTitan → Snowflake fits in the rest of your tracking
- FieldEdge → SnowflakeConnecting FieldEdge to Snowflake
- Jobber → SnowflakeConnecting Jobber to Snowflake
- ServiceM8 → SnowflakeConnecting ServiceM8 to Snowflake
- ServiceTitan → SegmentConnecting ServiceTitan to Segment
- ServiceTitan → BigQueryConnecting ServiceTitan to BigQuery
- RoofingConversion tracking for roofing companies, from inspection to signed job
- Housecall Pro → SnowflakeConnecting Housecall Pro to Snowflake
- Workiz → SnowflakeConnecting Workiz to Snowflake
Free · 30 min
Get ServiceTitan and Snowflake joined properly
Free30 minutesNo obligation
