Vidyard

A video platform of long standing with a generative layer added on top in the last two years. What the company has sold since the start is recording, hosting, delivery and measurement: a representative records their screen or camera, the file is hosted, a link is shared, and the platform reports who opened it, how long they watched and where they stopped.

Around that sit call to action overlays, branded sharing pages, folder management, permissions and connections into the systems a seller already uses. All of it is deterministic and all of it predates the model work. Strip every model and the platform that the vendor states one hundred and sixty thousand companies use remains intact.

The generative additions are real and prominent, covering avatars generated from a short training clip, an agent that composes and dispatches personalised videos on buyer signals, and a script generator, and the vendor's marketing now leads with them.

They are also, structurally, features on a mature product rather than the product itself: the agent is a paid addition at every paid tier rather than the thing being bought, and the free plan's fifteen generated videos sit alongside a recording tool that works without them. Ask which capabilities depend on a model, and what proportion of videos sent are generated rather than recorded.

Detailed control over who may watch a video, and nothing describing what governs the agent that makes and sends them. The access controls are published per tier and are more thorough than most: password protection, restriction of playback to named domains, address based access control, an explicit request access flow, scheduling and expiry so a video stops being viewable after a date, custom team permissions and an account history log.

For a product whose output is a link that can be forwarded anywhere, that is the right control set and it is documented rather than implied. The gap is on the other side of the process. The video agent is described as automatically creating and delivering personalised messages when a buyer takes an action, which is composition and dispatch without a person in the loop, and nothing published describes an approval step before a generated video goes out, a volume ceiling, a confidence threshold, or what halts a workflow already running.

The distinction matters because the two control sets solve different problems: one stops the wrong person watching a video, and none of them stops the wrong video being made in a named employee's likeness and sent. Ask whether generated videos can be held for review before dispatch, and what limits agent sending volume.

Three generative capabilities named and described in operational detail, with nothing published about the models behind them. The vendor states that a custom avatar is trained from a user supplied video of roughly ninety seconds, that the result looks and sounds like that person, and that it delivers scripts in more than twenty five languages, which is a clearer account of the input and the output than most vendors give.

What is absent is everything in between: no model provider, family or version for the avatar generation, the agent or the script generator, no model card, and no accuracy, quality or failure rate for any of them. The gap is sharpest on quality, because independent reviewers report specific and consistent limitations, describing lip synchronisation as unnatural and facial expressions as stiff compared with a real recording, while the vendor's own material describes the output as hyper realistic.

Those two characterisations cannot both be right, and only one of them is published by the party that can measure it. Multilingual delivery raises a second unmeasured question, since a script rendered in a language the represented person does not speak has no baseline for a viewer to judge it against. Ask which models generate avatars and scripts, what quality measurement exists, and how multilingual output is validated.

Named customers, linked case studies, and multiples with no baselines behind any of them. The vendor publishes outcome figures attributed to identifiable companies and links each to its own case study, which is better attribution than a wall of unsourced statistics: a doubling of response rates at one, four times more meetings booked at another, five times response rates at a third, and an eight times improvement in click through with four times better replies at a fourth.

Company scale is disclosed openly at more than twelve million people using the platform, one hundred and sixty thousand companies, and roughly ninety million dollars raised. Review platform standing is cited with the count. None of the outcome figures states a baseline, a period, a population or what the comparison was against, and one published claim, an eleven hundred percent increase in demonstration conversion rate at a named marketing platform, is extraordinary enough that its absence of method is conspicuous rather than routine.

Two further points belong here. The vendor publishes a category buyer's guide on its own blog assessing the classes of product it competes in, which is not independent evidence wherever a reader encounters it. And independent reviewers report a quality limitation the vendor does not acknowledge, describing generated avatar lip synchronisation as unnatural and expressions as stiff against a real recording. Ask what baseline and period each customer figure covers.

The platform generates and hosts, and much of the sending happens through somebody else's infrastructure, which lightens this axis without emptying it. Videos reach prospects mainly through connected sales engagement platforms, record systems and marketing automation tools, so the sending obligations, suppression lists and consent records largely sit in those systems and with the customer operating them. That is a genuinely lighter position than a product with its own campaign engine.

It is not the whole picture. The vendor markets the generative capability explicitly for cold outreach, publishing guidance that representatives can send personalised introduction videos to prospects who have not engaged, with scripts assembled from the prospect's name, company and an insight drawn from enrichment data, and the agent dispatches on buyer signals automatically.

Nothing published describes what consent basis the vendor considers necessary for a first video to a cold contact, how a recipient stops receiving them, or what obligations transfer to the customer. Ask what the vendor considers the customer responsible for on cold video outreach, and how a recipient opts out.

A privacy notice, a state privacy opt out control and a dedicated trust site that was not retrieved on this pass, so this records the limit of the pass rather than an absence. Two categories of personal data sit in this platform and only one is ordinary. The ordinary one is viewer analytics: the vendor reports who watched a video, for how long and where they stopped, which is individual level behavioural data about recipients who received a link and may not know it is instrumented.

The unusual one is the avatar training material. A custom avatar is built from a video of a named employee's face and voice, and the resulting model is a synthetic likeness of a specific living person. Nothing published states how long that training footage is retained, whether the avatar model can be deleted on request, what happens to it when the employee leaves the company, whether the employee or the employer controls it, or what consent the employee gave beyond agreeing to record the clip.

Those are questions about biometric material belonging to an individual rather than to the customer that pays the invoice, and they are unaddressed. Ask who controls an avatar model, whether it is deleted when the person leaves, and what retention applies to training footage and viewer analytics.

One rights disclosure that is genuinely good, and a second one entirely absent. The good one concerns the stock avatars available on every plan including the free one: the vendor states they are created using paid actors and licensed for commercial use. That is an explicit account of who the likeness belongs to and on what basis a customer may use it commercially, and almost nothing else in this index makes an equivalent statement about generated likenesses.

It answers the question a careful buyer would otherwise have to ask. The absent one concerns custom avatars, where the likeness belongs to the customer's own employee rather than to a contracted actor. Nothing published describes what rights that person grants when they record the training clip, whether the grant is revocable, whether it survives their departure, who may generate videos in their likeness once it exists, or what happens if they object to a script attributed to them.

The stock case shows the vendor understands likeness rights need stating; the custom case is where the exposure actually sits, because the person is an employee rather than a contractor and may not have negotiated anything. Ask what rights an employee grants over a custom avatar, whether the grant is revocable, and who may generate in their likeness.

Official routes throughout, with a stated integration count and named connections into the systems that matter. The vendor publishes fifty one integrations covering record platforms, marketing automation, sales engagement, conversation intelligence and an automation service, each reached through that product's own documented interface, and the video agent is described as working through those same connections rather than around them.

A browser extension is distributed through the official store rather than as a download, which places it under that store's review and policy regime. Nothing scrapes a professional network, automates a social account or captures data the platforms do not expose, so the exposures that dominate this axis elsewhere do not arise at all. Two things hold it below the top band.

No conformance position is stated in the vendor's own words for any connected platform, and nothing describes what happens to deployed workflows if a connector's terms change. And the programmatic surface is the most restricted part of the product, with the interface and ingestion based integrations available on the top tier alone and data integrations capped at two on the tier below, so a customer wanting to work around a missing connector cannot unless they buy the largest plan. Ask for the stated position on each connected platform and what the browser extension accesses.

The platform manufactures convincing synthetic likenesses of real named people and publishes no governance around them. A custom avatar is a model of a specific employee's face and voice, capable of delivering any script in more than twenty five languages, and three slots are available on every plan including the free one.

Nothing published states who may generate video using a given avatar once it exists, whether the person depicted can review or refuse a script attributed to them, whether any approval sits between a written script and a published video in someone's likeness, what prevents an avatar being used after that person leaves, or whether any marker distinguishes generated output from recorded footage. Those are the controls that matter for this technology and none of them appears.

Separately, nothing states whether training footage, generated videos or scripts train or tune any model, which provider processes generation, or what retention applies. A trust site exists on its own subdomain and was not retrieved here, so material may sit behind it. The exposure is not hypothetical: a likeness that can say anything, in languages its owner does not speak, is a capability whose misuse controls are the first thing a security review should ask about. Ask who can generate using an avatar, whether the depicted person can refuse a script, and what happens to their avatar when they leave.

The product's marketed benefit is that a prospect forms an impression of a person from a video that person did not make. A representative records one clip; the platform then generates any number of videos in their likeness, speaking scripts they did not write, in languages they may not speak, addressed by name to prospects who have never met them.

The vendor's own guidance describes this for cold outreach and gives the reason plainly, that the recipient can put a face to the name immediately. That sentence is the finding: the value being sold is the impression of personal effort, and the effort is what has been removed.

Nothing published requires or suggests telling the recipient that a video is synthetic, no marker or watermark on generated output is described, and nothing indicates what a prospect would be told if they asked whether the person in the video had recorded it. The disclosure that does exist runs the other way, covering the commercial licensing of stock actor likenesses, which protects the customer rather than informing the viewer. Two features sharpen it rather than softening it.

Multilingual delivery means a recipient may watch someone appear to address them fluently in their own language on no basis at all. And the agent dispatches automatically on buyer signals, so no person need see the video before the prospect does. Ask whether generated videos carry any indication of synthesis, and what a recipient is told.

A stated count with the categories that matter named, plus surfaces beyond the connector list. Fifty one integrations are published, covering record platforms, marketing automation, sales engagement tools, conversation intelligence and a general automation service, which is the right spread for a product whose output has to reach a buyer through whatever the seller already uses.

Around them sit a browser extension distributed through the official store, a knowledge base on its own subdomain, a public status page reporting availability, and a partner application framework available even on the free plan. The agent is designed to operate through those connections rather than as a separate motion, which is the correct architecture for this product. Held below the top band because the programmatic layer is reserved almost entirely for the largest buyers.

The application interface and ingestion based integrations appear on the top tier alone, custom metadata attributes likewise, and data integrations are capped at two on the tier below before becoming unlimited at the top. A customer on the self serve plan gets the connectors the vendor built and no route to build anything. Ask which integrations are available at the tier being quoted, and what the interface permits at the top tier.

Nothing on hosting, region, tenancy or recovery was established on this pass, and a dedicated trust site exists on its own subdomain that was not retrieved, so this records the limit rather than an absence. Two published facts bear on the question without answering it.

A public status page reports availability, which is the one operational commitment visible, and matters more here than for most products because the deliverable is a hosted video served to a prospect's browser during a live deal rather than a back office function. And an integration with a compliance archiving service is offered as an addition at the top tier, which signals that regulated buyers are served and tells a reader nothing about where anything is stored.

The residency question has real weight for this vendor. It is a Canadian company selling into Europe and the United States, and the material at stake includes video recordings of employees, synthetic likeness models built from them, and viewer level analytics on identified recipients. No region, residency election, tenancy model or recovery objective was located in any direction. Ask which provider and regions host video, avatar models and analytics, and whether any regional option exists.

Product level controls published in unusual detail, and the assurance behind the platform not reached. The plan comparison sets out security capability per tier rather than in general terms, covering password protection, standard and custom team permissions, restriction of playback to named domains, address based access control, a request access flow, and an account history log, with single sign on at the top tier and an integration with a compliance archiving service available there as an addition.

Publishing that as a per tier matrix lets a buyer see exactly which protections their budget reaches, which is more than most vendors offer, and the domain and address restrictions are the right controls for a product whose output is a forwardable link. What could not be established is the vendor's own posture.

A trust site exists at its own subdomain, described as the security programme, and was not retrieved on this pass, so no certification, audit period, auditor, report request route, penetration testing statement or vulnerability disclosure route could be confirmed and nothing here asserts that none exists. One gating decision is worth noting: single sign on is available on the top tier alone, so identity governance depends on spend. Ask which attestations are held with their audit periods and auditors, and what the trust site states.

The most complete pricing disclosure recorded on this list, and the only vendor here publishing a machine readable version of it. The live page renders its figures through script, which returned nothing to a machine on this pass, and the vendor has anticipated exactly that: a link labelled for artificial intelligence agents leads to a plain text file at a stable address carrying the full pricing, dated, stating when it was last updated and pointing to the live page as authoritative.

That file gives the free plan and its limits, the self serve tier at eighty nine dollars per seat monthly or fifty nine billed annually, and both upper tiers as quoted. It prices the generative addition separately and at both cycles, at twenty nine dollars per seat monthly or twenty four annually, so the complete self serve purchase path including the artificial intelligence capability can be modelled to the dollar before any contact with the vendor.

Beneath that sits a full feature comparison across four tiers and ten capability groups, and a billing section stating that the self serve tier is charged by card while the others are invoiced with annual and multi year options. Two small imprecisions belong on the record without moving the grade.

The blanket statement that annual billing saves about thirty percent holds for the subscription, where the saving is roughly thirty four percent, but not for the addition, where it is about seventeen. And the stated philosophy of plans without excessive additions sits above a structure where the flagship generative capability, caption translation and compliance archiving are all additions. Ask the rate for the quoted tiers and the usage rate for caption translation.

Nothing published on leaving, and the deliverable stops working when a customer does. No export scope or format, post termination access window, retention period or deletion timeline was located on any surface reached. The exposure has a shape particular to hosted video.

Everything this product creates is served from the vendor's infrastructure to a viewer's browser through a link, so videos already sent to prospects, embedded in proposals, linked from electronic mail threads or placed on the customer's own website depend on the subscription continuing. Nothing published states whether those links survive a cancellation, for how long, or whether a customer can retrieve their video library in a playable form rather than simply losing access to it.

A second gate compounds it: the application interface, which is the obvious route for bulk retrieval of a large library and its analytics, is available on the top tier alone, so most customers have no programmatic way to extract what they have accumulated. Custom avatar models raise a third question, since nothing states whether a departing customer's synthetic likenesses are deleted or retained. Ask what happens to live video links after cancellation, how a library and its analytics are exported below the top tier, and whether avatar models are deleted.

Most delivery runs through the customer's own connected tools, which lightens this axis, and the medium introduces a problem nothing addresses. Videos generally reach prospects through connected sales engagement platforms, record systems and marketing automation, so sending infrastructure, authentication and bounce handling largely sit in those systems and with the customer, and this vendor is not operating a sending reputation of its own for that traffic.

The agent dispatches through the same connections. What is unaddressed is specific to video in electronic mail. A video message is delivered as a thumbnail image linked to an externally hosted page, which is a pattern filters treat with more suspicion than plain text, and the vendor publishes nothing on how thumbnails are served, what domains the links resolve to, whether a customer can use their own domain for them, or what effect any of it has on inbox placement.

For a product whose entire value proposition is that the message gets opened, the absence of any guidance on whether the format helps or hinders arrival is a real gap. Ask what domain hosted video links resolve to, whether a custom domain is available, and what guidance exists on inbox placement for video messages.

Coverage stated across four dimensions at once and evidenced by name rather than by adjective. Company size is addressed explicitly from startups through mid market to enterprise, industries are named as financial services, software and consumer sales, use cases run across prospecting, selling, deal progression, post sale and marketing, and the commercial ladder runs from a genuinely usable free tier to quoted enterprise agreements.

Scale evidence is concrete, with the vendor stating twelve million people and one hundred and sixty thousand companies, and named enterprise customers a reader will recognise across software, professional networks and financial information. Review platform recognition spans enterprise, mid market, small business and a European region, which is unusual breadth and suggests the claim to serve all three sizes holds rather than being aspirational.

Generated avatars deliver in more than twenty five languages, which is a real internationalisation capability rather than a translated interface. Two things hold it below the top band. One published use case, internal corporate communications, sits outside the go to market frame entirely, so part of what the product is sold for is not what this index measures. And no minimum volume or team size guidance is published, so a buyer cannot tell where the free tier stops being adequate. Ask what video volume the paid tiers assume, and which regions are supported for hosting.