Sendoso

A logistics business with a recommendation layer added on top nine years in. What this company actually operates is warehouses, sourcing relationships, packaging lines, carrier accounts and customs handling, and the software around them selects a recipient, captures an address, triggers a send and tracks a parcel. None of that requires a model and none of it would stop working if every model were removed.

The generative layer arrived in September 2025 as a named suite of three capabilities: one recommends which gift to send from signals including recorded sales conversations, recipient interest profiles and customer record activity, one writes the note copy, and one verifies postal addresses before dispatch. The last of those is address validation rather than inference.

All three sit on the upper tiers only, so the entry tier customer is buying the platform without any of it, which by itself settles where the product's substance lies. The vendor's own positioning is a sending management platform, and its stated scale is measured in sends processed and customers served rather than in anything the models do. Ask which capabilities depend on a model and what the entry tier does without them.

Sends can fire from campaigns and record events, so many go out without a person choosing each one. Sendoso sells a Policy Center and anti bribery controls for governing who may send what, with Policy Center included on Enterprise and available as an add on below it, alongside IP allow listing, an admin audit log and SCIM user management. Advanced and Enterprise add the AI agent Oso and the full SmartSuite. What the Policy Center enforces, whether value ceilings, approval routing or restricted recipients, and what Oso may send unattended, is not described.

The inputs are named with unusual specificity and nothing else is. The recommendation capability is described as drawing on recorded sales conversation data, recipient interest profiles inherited from an acquired company, and customer record activity, which is a more concrete account of what goes into a model than most vendors give and lets a buyer see immediately which of their systems would be feeding it. Naming the inputs is the useful half.

The other half is missing entirely: no model provider, family or version is stated for the recommendation engine or the copy generator, no model card exists, and no accuracy, acceptance or quality figure is published for either. For the recommendation engine that gap has a measurable consequence the vendor could report and does not, since a gift recommendation either is or is not accepted and the platform sees the outcome.

The copy generator produces the message that accompanies a physical gift, and nothing describes what constrains its tone or content before it is printed and dispatched, which is a one way operation in a way that an email draft is not. Ask which models power recommendation and copy generation, what proportion of recommended gifts are accepted, and what review sits between generated copy and a printed card.

Substantial scale figures, recognisable customers, and no published measure of the one thing the product is judged on. The vendor states more than fifteen million sends processed, more than eight hundred customers, revenue above one hundred million dollars annually and named brands across media, software and finance, which is a real account of size and more than most private companies disclose. What is absent is delivery performance.

This is a business whose deliverable is a physical parcel arriving at a stranger's desk on a particular day, and no on time rate, damage rate, undelivered rate or customisation accuracy figure is published anywhere. That absence is conspicuous rather than ordinary for two reasons. Independent reviewer commentary converges consistently on exactly that weakness, describing inconsistent shipping times, uneven packaging quality and missed customisation details across orders.

And the vendor ships a capability whose stated purpose is to reduce undelivered packages, which means it measures the undelivered rate internally and does not publish it. A return on campaign investment figure circulates on third party listings with no population, period or method attached and was not adopted. Ask for the on time delivery rate, the undelivered rate and the damage rate across a stated period and geography.

A physical send needs a recipient's postal address, often a home address, held and used without that person asking for anything. Sendoso offers anti bribery controls and a Policy Center aimed at gifts to regulated recipients, plus a data subject request portal and an opt out form for the sale of personal information.

Its policies leave out the consent basis for holding residential addresses, how a recipient stops future sends, and how unsolicited goods rules and gift limits for public officials apply across its international fulfillment.

Sendoso publishes a privacy policy, a data processing addendum, a subprocessor list, a CCPA notice, a data subject request portal and an opt out form for the sale of personal information, with a trust center on its own subdomain. The platform holds a good deal about people outside its customer base: names, employers, titles, postal addresses that are often residential, send and acceptance history and, through AI gifting, interests inferred to choose a gift. How long recipient records are kept, and how a recipient sees or contests an inferred interest profile, is not covered in Sendoso's product material.

The platform holds inferred personal information about individuals and no account of where it comes from was located. Recipient data arrives from two directions. The customer supplies names and addresses from their own systems, which is unremarkable and carries the customer's own provenance.

The second source is the one that needs explaining: recipient interest profiles, inherited from an acquired company and described in independent accounts as derived from job title, company size, industry and inferred personal interests, are used to select what a person receives. Nothing published names a data supplier behind that inference, states whether it is purchased, licensed or derived, or describes what basis supports holding it. No indemnification position was located.

A separate provenance question sits on the merchandise side, where the vendor sources physical goods and operates a company store capability, and nothing published addresses supplier standards, product authenticity or origin, which for branded goods a customer's logo appears on is a reputational exposure of a kind software vendors do not carry. Ask what the recipient interest profiles are derived from and under what licence, and what supplier and sourcing standards apply to merchandise.

Connections run through sanctioned interfaces and the vendor operates its own fulfilment rather than borrowing anyone's. Integration is a stated strength of the product with a claimed library in the high double digits, and the named connections are to customer record platforms, sales engagement tools, marketing automation and conversation intelligence systems, each through that product's own documented route.

Nothing scrapes a professional network, drives a browser extension against one, or automates a user's social account, so the exposures that dominate this axis elsewhere do not arise. Held below the top band on two grounds. No conformance position is stated in the vendor's own words for any connected platform.

And the dependency that actually threatens continuity here is not a software platform at all: the product depends on postal and courier carriers, customs regimes and import rules across every country it ships to, and nothing published describes what happens to delivery commitments when a carrier changes terms, a customs regime changes, or a route becomes unavailable. That is the platform risk for this vendor and it is undocumented. Ask what carrier and customs dependencies underpin international delivery and what contingency exists when one changes.

One published data flow here deserves a buyer's attention more than anything else on this axis. The recommendation capability is described as drawing on recorded sales conversation data, which means the content of a customer's calls with their own prospects moves from a conversation intelligence system into this vendor's recommendation model in order to choose a gift.

Sales calls contain pricing discussions, competitive information, personal remarks and everything else people say when they think only two parties are listening, and nothing published states what portion is transmitted, what is retained, whether it trains any model, which provider processes it or whether anything crosses a customer boundary. The same questions apply to the recipient interest profiles and to generated note copy, and none is answered.

The trust centre that might address them renders client side and returned no content to this pass, so material may exist behind it and nothing here asserts otherwise. No governance document, evaluation record or audited management standard for artificial intelligence was located. Ask exactly what conversation data the recommendation engine receives and retains, whether it trains any model, and which provider processes it.

The gift is honest about who sent it and the note attached to it may not be honest about who wrote it. There is no persona, no synthetic voice and no agent conducting a conversation, and the recipient knows exactly which company the parcel came from, which places this well above the practices this axis usually records. The complication is the message.

The vendor ships a capability that generates the note copy accompanying a gift, and separately offers packaging options including handwritten copy on the card, so a recipient can receive what presents as a personal note written by the named sender whose words were produced by a model. A handwritten card carries an implicit claim about effort that a printed email signature does not, and that claim is the thing being automated.

Nothing published states whether generated copy is reviewed before printing, whether a sender must approve it, or whether anything indicates to the recipient that the message was generated. The recipient interest profiling compounds it, since a note referencing a personal interest the recipient never disclosed to the sender may read as thoughtful attention rather than as inference. Ask whether generated note copy is approved before printing and whether any indication reaches the recipient.

Sendoso connects to CRM, sales engagement, marketing automation and conversation intelligence tools, with one platform integration on Start With Gifting, two on Core and unlimited integrations on Advanced and Enterprise, and Shopify on the merchandise plan. A public REST API is documented, and the Sendoso MCP server comes with the gifting plans and is listed on the Databricks Marketplace.

Underneath the software sit Sendoso's own fulfillment centers, including Phoenix and Ireland, storage for customers' branded stock, and sourcing and carrier arrangements across several countries. Sync behavior and limits for each connector are not documented in one place.

Residency has two dimensions for this vendor and neither is documented. The ordinary one concerns where recipient records are hosted, and nothing on provider, region, tenancy or recovery objectives was established, with the trust centre that would carry it rendering client side and returning no content to this pass. The second dimension is physical and is particular to this product.

Fulfilling an international send means a recipient's name and address move to a warehouse and a carrier in the destination country, and the vendor operates its own facilities across regions, so recipient personal data crosses borders as an operational necessity rather than as an architectural choice. Nothing published describes which entities receive that data, under what arrangements, or whether a customer can restrict fulfilment to a region.

For a customer sending into Europe from a United States platform that is the first question a data protection review will ask and there is no published answer. Ask where recipient records are hosted, which fulfilment entities receive recipient data for international sends, and whether regional fulfilment or storage can be specified.

A dedicated trust centre exists on its own subdomain, built on a recognised compliance automation platform, and it would not render to a machine on this pass. The page returned descriptive metadata and no body, which is a retrieval limitation rather than a finding, and portals of that kind ordinarily carry certification status, policy documents and a request route, so material almost certainly sits behind it.

Independent commentary refers readers to it for the company's compliance standards and certifications, which corroborates that it holds something. What could be established from the surfaces reached is nothing: no certification is named, no audit period, auditor or report appears, no penetration testing statement was found, and no vulnerability disclosure route was located. Nothing here should be read as asserting the vendor holds no certification.

The practical point for a buyer is the same one this pattern always produces: a security posture that only a browser can read is unavailable to procurement tooling and to automated vendor risk assessment, which for a vendor selling into enterprise revenue organisations is a friction the vendor is choosing. Ask for the current attestation with its audit period, scope and auditor, and for a document set that does not require a rendered portal.

Sendoso publishes its plan structure and its fulfillment rates but no license price. Five plans run from Start With Merch and Start With Gifting, each with five complimentary licenses and no spend commitment, through Core with 10 licenses and unlimited sending, Advanced with 50 and Enterprise with custom licensing, and each plan's sends, fulfillment regions, Campaign Studio projects, integrations and AI tools are listed.

Shipping and handling for most inventoried sends follow published rate tables, and handling for a send above 10 items is $10 plus $0.85 for each extra item. Priority shipping, heavy ground parcels, Amazon sends and aged inventory fall outside those tables. Every plan is priced through sales.

Two exit exposures here have nothing to do with data and both are more consequential than the usual export question. The first is money. Gift spend runs against a prepaid balance the customer funds ahead of use, and nothing published states what happens to an unspent balance when a contract ends, whether it is refundable, whether it expires, or how quickly it is returned.

A customer who prefunds a quarter and cancels mid quarter has an amount of their own money sitting with the vendor on undisclosed terms. The second is physical. Customers store their own branded merchandise in the vendor's warehouses, and nothing published states who owns that stock at termination, whether it is returned, what return shipping costs, or what happens to it if it is not collected.

Both are ordinary commercial terms in logistics and neither is visible before a sales conversation. On the data side the position is the usual one: no export scope, format, retention or deletion timeline was located for send history, recipient records or campaign analytics. Ask whether unspent prepaid balance is refundable and on what timeline, what happens to stored merchandise at termination and at whose cost, and what send and recipient history exports.

This vendor operates the most literal form of deliverability in the index and publishes no measure of it. Everywhere else on this axis the question is whether an electronic message reaches an inbox; here it is whether a physical parcel reaches a named person at a correct address, intact, on time and customised as instructed.

One real control is published: an address verification capability that checks and corrects addresses before dispatch, with the stated purpose of reducing undelivered packages, which is the right control at the right point in the process and directly addresses the most common failure. Against it sits everything that is not published.

No on time delivery rate, undelivered rate, damage rate or customisation accuracy figure appears anywhere, no service level commitment is stated for any destination, and no account is given of what happens when a send fails, who bears the cost of a reship, or how long international delivery takes by region. Independent reviewer commentary converges on inconsistency in exactly these areas, describing delays, uneven packaging quality and missed customisation. Ask for delivery success and on time rates by region, the service level for international sends, and who bears the cost when a send fails or arrives damaged.

Sendoso sells to marketing, sales, customer experience and people teams, from small businesses to global enterprises. The two starter plans carry five complimentary licenses and no spend commitment, Core and Advanced serve teams running gifting as a repeatable program, and Enterprise covers global programs across every fulfillment region. Fulfillment runs from its own centers in several countries, with phone lines in the United States, Canada, the United Kingdom and Australia. Named use cases span account based marketing, demand generation, field marketing, event fulfillment, sales development, retention and employee recognition. Supported languages go unlisted.