Return Signals

The agent takes authenticated action across a published catalog that is grouped and specific: messaging on iMessage, RCS and SMS, WhatsApp and Instagram; commerce on Shopify; returns and subscriptions with Loop, Redo, Stay.ai and Subscribfy; CX and helpdesk with Gorgias, Zendesk and Gladly; marketing and loyalty with Listrak, Postscript and Rivo; and fulfillment and operations with Order Desk, JayGroup, Manhattan Associates, EasyPost, Sanity, Google Sheets and Slack.

That reaches commerce, returns, support, marketing, fulfillment and messaging. The actions are writes: the agent turns a return into the right exchange, sends the product link with the correct size selected, and on subscriptions detects early churn signals and proactively adjusts delivery frequency or offers alternatives, which changes state in the merchant's subscription system.

The security page confirms the mechanism, describing validated credentials and authorization when connecting to commerce, messaging, support, fulfillment, return and analytics services, signed webhook validation for the Shopify app, and encrypted expiring access credentials that are refreshed or invalidated.

Sourcereturnsignals.com homepage and security pageread 2026-09-05

Multi-step work across a lifecycle, with no runtime or builder documented.

The agent runs sequences rather than single replies: on subscriptions it detects an early churn signal such as product buildup or a skipped order, then adjusts delivery frequency, offers an alternative or resolves the issue before a cancellation; on returns it reads intent, recommends the right exchange, and sends the product link with the correct size preselected; on delivery it checks in, diagnoses a fit or quality problem, and either resolves it or escalates in the same thread.

Scheduled follow ups run alongside live conversation, and playbooks are configured per merchant at onboarding. Nothing describes a runtime, a step model, branching, retries or failure handling, and there is no multi-agent or multi-participant execution. The merchant does not build or edit these flows; Signals configures them.

Sourcereturnsignals.com homepage and FAQread 2026-09-05

Rich context, assembled from live systems rather than held as an index. Signals grounds each conversation in a documented context set it calls customer context, covering size and fit, past orders, preferences, returns and support history, and answers product, care and order questions from the merchant's catalog, with the worked example citing stock in a specific size and a previously viewed product. Recommendations and Products are sections of the operating console, so the catalog is a first-class object.

No index, graph or embeddings layer that persists over the customer's knowledge is documented. What the pages describe is context drawn per conversation from Shopify, the helpdesk, the returns portal and the fulfillment feed, which is assembly from live systems, and no ingestion, indexing, chunking, retrieval or refresh surface is documented. The vendor's engineering blog describes agentic search over conversation history, and whether that rests on a maintained index is not stated.

Sourcereturnsignals.com homepageread 2026-09-05

A person can take the thread over; nothing gates what the agent sends before it goes. Most inbound and post-delivery conversations resolve with the AI, and when something unusual appears the conversation escalates to a human in the same thread so the team steps in without losing context.

The console makes that operational, showing escalated status, claimed and unclaimed states with the claiming agent named, a needs-reply activity flag, an AI against Human toggle on the thread, and a pause control, with worked examples including a photo uploaded for a quality issue and a reminder scheduled where a customer has not responded.

No pre-send queue, draft-review step, confidence threshold or per-action sign-off is documented, and the agent messages customers directly, including proactively, without a person releasing each message.

Sourcereturnsignals.com homepage operating surfaceread 2026-09-05

An access surface exists and no attestation does. Signals states it uses authenticated sessions and organization roles to control platform access, scopes application queries and authorization checks to the relevant organization, verifies Shopify App Bridge session tokens before returning merchant data, and requires fresh verification from the Shopify account owner before sensitive privacy exports.

Supporting control documentation is detailed: Google Cloud infrastructure with encryption at rest, application-level encryption of integration credentials and access tokens, secrets held in Secret Manager rather than source, separated production and non-production boundaries, signed webhook validation, idempotent redelivery, retention and deletion honoring Shopify's 30-day windows with de-identified suppression records preserving opt-outs, a bar on AI providers training general models on customer data, and a published vulnerability disclosure address.

There is no attestation, and Signals says so itself, opening the page by stating it describes controls without making claims about certifications it has not obtained, and the homepage describes an active SOC 2 program, meaning no certificate exists yet. The depth here is in infrastructure and data handling; the access side is a single sentence naming roles, with no permission model, role set or identity integration described.

Sourcereturnsignals.com security page, homepage FAQread 2026-09-05

A full record of what was said, without a reconstruction of why. The operating console lists every conversation with real state: status including escalated, activity such as needs reply, who claimed it and when, start and last-activity timestamps, applied tags such as intent and VIP, an AI against Human attribution on each thread, and a pause control, with search and filters across the set and an Analytics section beside it.

A team can therefore open any customer thread, read the whole exchange, and see whether the agent or a person was speaking. Nothing documents which product, order or policy context the agent retrieved for a given reply, which connected system it called, what it wrote back, or why it recommended one exchange over another, and no tool-call trace, decision log or action audit surface is documented. That matters because the agent writes into Shopify, returns and subscription systems.

Sourcereturnsignals.com homepage operating surfaceread 2026-09-05

Retained state is the product's central claim and its lifetime is never stated. Scope is clear and per customer: Signals keeps one iMessage conversation open across before order, order, delivery and next purchase, holds a customer context set of size and fit, past orders, preferences, returns and support history, and sells itself on knowing each customer personally, remembering what they like, remembering occasions that are coming up and following up on them, and telling them when something is back in stock.

Continuity across weeks is demonstrated rather than asserted, with a published thread where an exchange is resolved and the agent returns nine days later when the replacement arrives. Customers is a section of the operating console. No lifetime is stated. The security page says only that conversation and media retention can vary based on the customer's configuration and operational requirements, which defers the question rather than answering it, and nothing describes inspecting, editing or clearing what the agent remembers about one buyer.

Sourcereturnsignals.com homepage and security pageread 2026-09-05

One infrastructure, disclosed and not selectable. Signals states plainly that it runs on Google Cloud Platform, with Google Cloud managed services providing encryption at rest and production secrets in Google Secret Manager.

No region is named, no region selector exists, and no VPC, single-tenant, on-premises or customer-tenant option is documented; the security page's separation of production from non-production is an internal boundary rather than a customer deployment choice. The product's shape makes that unsurprising, since it operates as a Shopify app messaging end customers over carrier and Apple messaging channels, none of which a merchant could host. Retention and deletion are documented in detail, including Shopify's 30-day erasure windows, but that is data handling rather than deployment.

Sourcereturnsignals.com security pageread 2026-09-05

Named assets exist and the vendor assembles them. Signals ships recognizable lifecycle plays rather than a blank canvas, naming them on its own pages: Text Us capture from the storefront, post-delivery outreach, review workflows, back-in-stock notification, occasion follow-ups, fit and exchange handling, and subscription churn saves, with an Outfitting demo and a Recommendations section in the operating console.

Onboarding is described as configuring the first playbooks, and setup can take ten to twenty minutes because Signals connects to an existing stack rather than replacing it. Nothing is published as a set a buyer browses and adopts. There is no template gallery, no library, no named starter pack and no entitlement, and the playbooks are configured by Signals during a scoped pilot rather than selected by the merchant. The product spans pre-purchase questions, order changes, delivery, exchanges, subscriptions and repeat purchase.

Sourcereturnsignals.com homepage and FAQread 2026-09-05

The agent starts the conversation on an event, and the events are named. Delivery is the primary trigger: Signals checks in when items are delivered, which requires the shipment state that its EasyPost and fulfillment integrations supply, and the console carries a Scheduled follow ups section for timed outreach. Other documented starts are back-in-stock notification, occasion-based follow-ups the agent remembers and returns to, and subscription churn signals such as product buildup or skipped orders that prompt outreach before a cancellation.

Inbound arrives too, through Text Us capture on the storefront and customers replying into the open thread. The channels are distinct: iMessage, RCS and SMS, WhatsApp and Instagram, each an independent route on which a customer can reply, share photos and resume later. Reach is the differentiator the vendor sells on, claiming roughly 90 percent of buyers are reachable against a 10 to 20 percent marketing SMS opt-in, because these are service conversations rather than broadcast marketing.

Sourcereturnsignals.com homepage and FAQread 2026-09-05

The provider is disclosed and the merchant has no say in it. Signals is open about what it runs, publishing engineering posts that name its production configuration as Gemini 3.5 Flash and score candidate successors, reporting that Gemini 3.7 Flash High matched production across 35 customer-agent tasks at 52 percent lower cost per run and that 3.8 Flash Medium tied the previous winner. The security page adds that AI providers are not permitted to train general models on customer data.

So the model is known, and it changes when Signals decides it should. No customer or admin control is documented: there is no model selector, no routing rule, no per-workflow assignment and no bring-your-own-key path, and the vendor does not describe run-time routing across providers either; it selects one production model and re-evaluates it periodically, which is an internal engineering decision.

Sourcereturnsignals.com engineering blog and security pageread 2026-09-05

No API, SDK, developer documentation or MCP endpoint for Signals' own platform is published. The security page mentions public web and API endpoints served over HTTPS so merchants, Shopify, messaging providers and other connected services can reach them, but that describes the platform's own internal surface and its integration callbacks rather than a documented interface a customer builds against, and no reference, authentication model, endpoint list or client library is published.

Signals validates credentials when connecting out to commerce, messaging, support, fulfillment, return and analytics services, and validates signed webhooks arriving from Shopify. Both are the platform integrating rather than being integrated with. Google Sheets, Sanity and Slack are connectors in its integration catalog. The product is positioned away from a developer surface, sold on connecting to an existing stack in ten to twenty minutes with Signals handling the rest.

Sourcereturnsignals.com security page, homepage and FAQread 2026-09-05

A controlled experiment about the agent, run per customer as a matter of course. Signals states that it runs as an A/B test against a holdout so the lift a buyer reads is measured rather than modeled, that ROI is measured against a randomized holdout in the buyer's own Shopify order data, and that onboarding proceeds to a controlled rollout measured against that holdout.

Experiments is a named section of the operating console alongside Conversations, Analytics and Customers, so this is a product surface rather than a one-off study.

The published case studies show the mechanism: at Quaker Marine, from 18 February 2026 every new order was randomly assigned 50/50 to treatment or control across 1,910 customers, with treatment receiving a post-delivery check-in, yielding 16 percent more repeat purchases at three weeks; American Trench ran the same design against a holdout that received nothing.

The assignment is deterministic and measures incrementality against a stable control, with the counterfactual about the agent rather than about spend. Signals also publishes an internal model harness, scoring candidate models across 35 customer-agent tasks before a production change; that evaluates its own stack rather than giving the customer a capability.

Sourcereturnsignals.com homepage, Quaker Marine and American Trench case studiesread 2026-09-05

No browser, desktop or remote computer control is documented. Signals reaches every system it touches through credentialed integrations, which its security page describes as validating credentials and authorization when connecting to commerce, messaging, support, fulfillment, return and analytics services, with signed webhooks arriving from Shopify. Its customer-facing surface is a messaging thread on iMessage, RCS, SMS, WhatsApp and Instagram, which is a channel the agent speaks through rather than an interface it operates, and its merchant-facing surface is Signals' own console.

Sourcereturnsignals.com homepage and security pageread 2026-09-05