Your data team already has customer events in the warehouse. Product wants the Amplitude SDK for session replay, experiment exposure, and deeper behavioral analysis. Engineering reasonably asks why the company needs another collection path.

Both positions can be right. The decision is not simply where you want to store events. It is whether the evidence needed to answer a future question will exist at all. Use the SDK for evidence created inside the product experience, the warehouse for reconciled business facts, and a hybrid when you need both.

Choose the evidence before you choose the ingestion path

An ingestion architecture should start with decisions, not plumbing. Ask what your product, growth, marketing, finance, and data teams must be able to decide after the implementation is live.

Some questions depend on interaction evidence: What did the user click? Where did the interface stop responding as expected? Which variant did the user actually encounter? What happened before an anonymous visitor signed in? Other questions depend on authoritative business state: Was an order paid? Was it refunded? Is the subscription active? Does the account still have an entitlement?

Question you need to answer	Best default	Reason
Where did users struggle inside a page or flow?	Amplitude SDK	The answer may require clicks, page state, Web Vitals, heatmaps, or session replay.
Which acquisition path led to activation?	Amplitude SDK, with deliberate identity handling	Referrers, UTM parameters, deep links, and activity before login originate at the client experience.
Which experiment experience did a user receive?	Amplitude SDK	Exposure must be recorded where the experience is delivered.
What is the current state of an order, payment, refund, subscription, or entitlement?	Warehouse ingestion	These records are normally reconciled in backend systems and the warehouse.
Do analysts need governed, correctable, lineage-tracked records?	Warehouse ingestion	The warehouse may already enforce canonical schemas, lineage, access controls, and correction workflows.
Why did behavior change, and what was the resulting business impact?	Hybrid	The diagnosis needs interaction evidence while the impact calculation needs authoritative business facts.

This distinction prevents a common category error. A warehouse can contain events without containing enough evidence to reconstruct an experience. An SDK can capture detailed behavior without becoming the authoritative ledger for money or entitlements.

The SDK captures evidence created inside the experience

The Amplitude SDK operates close to the browser, mobile app, or product surface where behavior occurs. That position lets it observe context that may never reach a transactional table. Amplitude identifies Web Vitals, rage clicks, dead clicks, acquisition context, pre-login behavior, and experiment exposure as capabilities the SDK captures directly while warehouse-only implementations require manual collection or cannot provide the corresponding product feature.

The SDK is the stronger default when your roadmap depends on any of the following:

• Session Replay that lets a team inspect what happened during a user journey.
• Heatmaps and Zoning Insights that connect interaction patterns to areas of a page.
• Automatic collection of signals such as rage clicks and dead clicks.
• Web Experimentation exposure tracking tied to the delivered experience.
• Delivery of in-product Guides and Surveys.
• Acquisition context such as referrers, UTM parameters, and deep links.
• Linking activity before login to the user who later authenticates.

The practical issue is irreversibility. You can import an old order record later because the order still exists in a backend system. You cannot retroactively create a replay, a DOM snapshot, or a reliable experiment exposure if nothing captured it during the session. If a likely product question depends on moment-of-interaction context, instrument that context before you need the answer.

That does not justify collecting everything. Autocapture and replay increase the amount and sensitivity of the data you manage. Decide which surfaces are in scope, mask or exclude sensitive fields, respect applicable consent choices, and give every retained signal a decision it supports. The SDK creates collection capability; it does not replace a tracking plan or privacy review.

Before committing, test the exact product capabilities available in your Amplitude environment. Product-specific availability and configuration can change, while the architectural principle remains stable: evidence that only exists in the live experience has to be collected there.

Warehouse ingestion protects business truth and governance

The warehouse is usually the better authority for facts created and reconciled by backend systems. Orders, payments, subscriptions, refunds, and entitlement changes may be updated after the client interaction. They can also exist even when client-side analytics is unavailable.

This makes warehouse ingestion valuable for three different reasons:

• Completeness: Reconciled revenue records are not lost merely because a browser blocked client-side collection or a user declined client-side analytics collection. That does not remove your obligation to apply the correct privacy, legal, and governance rules when using those records.
• Correction: A refunded payment or changed entitlement can be represented using the corrected backend state instead of the state the client observed earlier.
• Control: The warehouse may already provide approved schemas, lineage, access controls, and correction workflows. In a regulated environment, adding an independent collection path can create a governance problem that outweighs the extra behavioral detail.

For teams that need only these authoritative business states, warehouse import can be the appropriate ingestion path. You can analyze the imported facts in Amplitude, or export product data and join it to governed records in the warehouse.

Be precise about what “the data is already in the warehouse” means. If an existing pipeline such as an internal event system captured the required interaction signal, you may be able to reuse it. If the pipeline captured only page names, transactions, or coarse lifecycle events, moving those records into Amplitude will not manufacture rage-click evidence, session replay, DOM state, or trustworthy exposure events.

Before choosing a warehouse-only design, ask product and growth leaders to name the questions they expect to answer. If they need conversion totals but not an explanation of the experience behind them, warehouse ingestion may be enough. If they expect visual diagnosis, automatic interaction signals, in-product delivery, or experiment exposure, warehouse-only ingestion creates a capability gap.

A hybrid architecture keeps collection and authority separate

SDK and warehouse ingestion are not rival attempts to collect the same dataset. In a well-designed hybrid setup, each path has a distinct job:

<!– wp:list {

---