Oliv AI

Nothing survives the removal test here, which is the cleanest case this axis gets. Every named product is an agent, the three layers underneath are synthesised rather than stored, and the vendor states plainly that its signal layer is more than a hundred small models each fine tuned to a single revenue question and reasoning over the customer's full context.

Strip that out and there is no pipeline tool left behind it, because the pipeline view is an output of the synthesis rather than a database with intelligence added. The vendor's own framing agrees and goes further, calling itself artificial intelligence native infrastructure for revenue teams and arguing that the barrier to adoption is not the models but the harness around them. Even the commercial framing points the same way, describing conventional applications as a commodity that should be priced cheaply so the savings can fund agents.

The vocabulary of oversight is more complete here than at almost any vendor in this index. The harness is enumerated by name across six components including guardrails, audit, monitoring and logging, and permissions and approvals are listed as things enterprise agents require.

The process layer is described as encoding the customer's rules, handoffs and approvals into how agents work, onboarding states that agents operate by the customer's rules rather than generic defaults, and the customer relationship agent is shown surfacing field suggestions that were accepted before being applied, with the manager facing material confirming that fields are filled by the model and confirmed by representatives. Expansion is customer controlled by explicit statement.

Held off the top band for two reasons. The mechanics behind those words are never specified, so approval queues, escalation thresholds and role definitions exist as vocabulary rather than as described behaviour. And the vendor's own copy raises a sharper question than any of it answers: the supervising agent learns how the team operates, turns that into the rules every other agent follows, and is shown updating three process files on its own initiative to enforce a change in selling behaviour. An agent rewriting the rulebook that constrains the other agents is the point at which oversight most needs describing, and it is not described.

The architectural disclosure is genuinely unusual and it is the reason for the band. Rather than gesturing at intelligence, the vendor publishes a five layer account with a dedicated page for each layer, and states the shape of the model estate directly: more than a hundred small models, each fine tuned to one revenue question, reasoning across the captured context.

Naming an approach and a count, and separating capture from object resolution from signal synthesis from process encoding, tells a buyer more about how the thing works than a provider name alone would. Held off the top band on the two things a procurement review would still ask. No base model, provider, family or version is named anywhere, and fine tuning implies a training corpus that goes entirely unidentified.

Worth noting for accuracy rather than as a disclosure: a named third party assistant appears prominently on the homepage as the thing this vendor is not, positioned as adequate for prototyping and inadequate in production, which identifies a competitor rather than a dependency.

Nine named and recognisable customers appear as logos across healthcare software, insurance, developer tooling and enterprise adoption, two named individuals give attributed testimonials with title and employer, a customer stories page exists, and a sandbox lets a buyer try the product without talking to anyone. That is a real base and it is what holds this row above the bottom band. What is missing is any number attached to any of it.

Not one quantified customer result is published anywhere read, and the deployment timeline promises measurable return on the first agent by week four without showing a single example of what was measured. One detail deserves flagging because of where it sits.

The most numeric looking content on the homepage, stalled deals at sixty eight percent, renewal rates falling from ninety four to seventy one, a three hundred and ten thousand dollar forecast swing, all appears inside simulated agent notification cards. Those are interface mock ups demonstrating what an agent would say, not results any customer reported.

Two agents reach people outside the customer's own company and neither arrives with a compliance surface. One builds tailored outreach and runs campaigns against target accounts. The other calls and qualifies inbound leads before their interest cools, which means an agent placing telephone calls to individuals.

Across every page read, no electronic marketing or telemarketing statute is named, no consent position is taken, and suppression, complaint handling, calling windows, registry scrubbing and messaging registration are all absent. A third agent telephones the customer's own representatives every evening to collect pipeline updates, which raises employee facing questions rather than recipient facing ones and is worth separating for that reason. The terms of service exist and were not read, so a compliance position may sit inside them, and none is visible on any marketing or product surface.

A privacy policy and a cookie policy exist at their own routes and were not read, so this row records a flag rather than a judgement on contents. What is established is the scale of what flows in, and it is among the largest of any vendor in this index.

The capture layer takes every meeting, every email, every call and every document automatically, across two mail providers, two conferencing platforms, telephony including a dialer already graded here, two systems of record and a stated seventy or more further applications, and writes all of it to the right account, contact and deal. Recorded customer conversations, internal representative conversations and document contents therefore accumulate continuously.

Against that: no processing addendum was located in any footer or navigation, no sub processor list, no retention position, no lawful basis and no data subject route. Reading the policy is the highest value check available on this vendor.

Most of what this platform holds originates with the customer, which is the clean part: their calls, their emails, their documents, their system of record. Two layers sit outside that and neither is sourced. A research agent produces what the vendor calls deep account dossiers covering buyer pain points and decision maps, which means information about individuals who are not customers and have not been asked, and independent reporting from the earlier product generation described that research drawing on the open web, a professional network, a company database and regulatory filings.

And the integration list names a professional network directly among the platforms it syncs from. No supplier, licence, collection method, refresh cadence or lawful basis is stated for any of it on the pages read, and the distinction between context the customer already owned and intelligence assembled about third parties is never drawn.

The exposure surface is wide and no position is taken anywhere on it. Capture runs across two conferencing platforms, two mail providers, several dialers, competing conversation intelligence tools and two systems of record, and the vendor's own integration answer names a professional network among the platforms it syncs from. A browser extension is published in an extension store. Meeting capture implies an attendee or a recording connection on calls hosted by third parties.

Every one of those is a place where another platform's terms govern what an automated participant may do, and no conformance statement, method description or account model appears on any page read. Held at this band rather than lower because everything here is presented as an ordinary integration rather than as circumvention, nothing rents or rotates an identity, and the vendor's posture toward the tools it connects to is cooperative to the point of letting a customer keep a competitor's recorder in place.

This vendor states the fact that makes the stewardship question urgent and then leaves the question open. Its signal layer is described as more than a hundred small models each fine tuned to one revenue question. Fine tuning is training, the material available to train on is the recorded conversations, emails and documents of every customer on the platform, and nothing published says whether any customer's content contributed to any model that serves anyone else.

That is the cross client boundary question in its sharpest form in this session, sharper than at the vendors who merely say their systems learn, because here the vendor has named the mechanism. Around it, audit and security appear as named harness components and security and compliance as another, which is vocabulary rather than commitment. No statement on training, retention, isolation between tenants or deletion was located on any page read, and the privacy policy that might carry one is unread and flagged.

One agent here telephones people, which makes this the live case rather than the theoretical one. An inbound response agent calls and qualifies leads before their interest cools, and nothing on any page read states whether the person answering is told they are speaking to a machine, on whose behalf it is calling, or how they opt out of being called that way.

That is the obligation now in force in Europe for systems interacting with people, and the marking duty rests principally on the provider of the system. Two further surfaces sit alongside it: an assistant that attends and captures meetings, with no announcement or named participant mechanic described, and an outreach agent whose campaigns go out under a representative's name with no authorship position. Held at this band because no identity is manufactured, no sender is substituted and nothing celebrates concealment.

Breadth is real and specifically named rather than claimed as a number. The integration answer lists two systems of record, two mail providers, two conferencing platforms, three dialers, three competing conversation intelligence tools and a professional network, and the capture layer adds telephony and a stated seventy or more further applications. A browser extension is published. A sandbox lets a buyer connect and try without a sales conversation.

The posture is unusually accommodating: a customer can keep their existing meeting recorder from any of five named competitors rather than being required to move, and migration from a competing tool is offered at no cost with recordings, transcripts, notes and metadata included. Held off the top band on the programmatic half, which is absent.

No published interface, webhooks, protocol server, developer documentation or marketplace listing was located, and the agent marketplace is a request form rather than a catalogue anything can be built against. The two integration counts given on the same page also disagree with each other.

Hosting provider, country, region and residency options are absent from every page read, and there is no security page or trust centre where they would ordinarily sit. That matters more here than at a lighter product, because this platform continuously ingests recorded customer conversations, internal representative calls, mailbox contents and documents from across a company, and because two of its named customers sell software into healthcare, where processing location is routinely a procurement gate.

The pricing page carries a badge suggesting alignment with a United States health statute, which implies the question has been asked by buyers, and no residency answer accompanies it. The corporate identity available is also thin: an entity name in the copyright line, a professional network page and a careers board, with no registered address or jurisdiction stated anywhere.

Five certification marks appear on the pricing page under a heading promising the buyer's data is safe, and that is the entire security surface of this website. They are unlinked images. No type, scope, period, auditor, report or issuing body accompanies any of them, no security page exists in the navigation or the footer, no trust centre was located, and no status page or vulnerability reporting route appears anywhere.

The index has recorded this pattern before and the wording holds: asserting certifications without enumerating them is weaker than it looks. Two of the five deserve individual scrutiny. One names a United States health statute, which is a regulatory regime a company complies with rather than a certification it holds, and this index has flagged the same category error at another vendor this month. Another names an international standards body without identifying which standard.

Set against a platform that holds the recorded conversations, mailboxes and documents of an entire revenue organisation, and that markets audit, security and compliance as components of the very harness it says competitors lack, the absence of any supporting surface is conspicuous. The privacy policy and terms were not read and a documented control set inside either would move this row.

The published path is real and a buyer can work it out unaided. Pricing is modular in three steps: a meeting recorder at nineteen dollars per user per month, or the option to keep a competitor's recorder instead, with five named alternatives accepted, then insight packs at twenty nine dollars per user per month each for meetings and for deals, bought only if wanted. A calculator converts a team size into an annual figure and its arithmetic checks out.

The no fees claim is enumerated rather than asserted, naming training, support, implementation and integrations as included, migration from a competing tool as free, and viewers and collaborators as never charged. Held off the top band by two things, one ordinary and one substantial. The ordinary one is that every price requires annual commitment and annual payment with no monthly option, and volume discounts are described as flexible rather than published.

The substantial one is a gap between the product being sold and the product being priced: the entire site sells agents that own responsibilities, and the pricing page prices a notetaker and two insight packs, then displays six agents with no figure against any of them and no statement of whether they are included in a pack or bought separately.

The clearest instance this session of a pattern worth naming, because the vendor documents arrival in detail and departure not at all. Inbound migration is a headline offer: free of cost, from any tool, with four competing platforms named, and the vendor specifies exactly what comes across, recordings, transcripts, notes and metadata, and states that its own team manages the whole process with no data loss.

That is a better described data movement than most vendors publish in either direction. Outbound, nothing. No export function, file format, download path, post termination data right, deletion timeline or retention period was located on any page read, and the terms of service were not opened.

The asymmetry is sharpened by the commercial terms, since every published price requires an annual commitment paid annually, so the customer who wants to leave is doing so at a contractual boundary with no documented way to take the accumulated conversation record with them.

Three agents generate outbound activity and none of them arrives with an operational discipline surface. An outreach agent runs campaigns against target accounts. An inbound response agent places telephone calls to leads. A meeting assistant sends follow up emails automatically.

Across all three, the apparatus this axis looks for is absent: no warmup guidance for a new sending domain, no volume pacing, no complaint rate threshold, no bounce handling, no authentication guidance, and on the voice side no position on caller reputation, number labelling, branded calling or attestation. The structural point worth recording is the one this index keeps returning to with agent platforms.

Removing the human effort that used to limit how much outreach a team could produce does not remove the constraints the receiving infrastructure applies, and a platform that automates the production is the natural place to state what replaces the old ceiling.

Some genuine shape and no measurements. Seven functions each get their own page, from sales development through onboarding, customer success, expansion, revenue operations and enablement, which describes the buyer precisely by role and shows the product is sold across a whole revenue organisation rather than to one team.

A dedicated page addresses startups, nine named customers give an industry impression, and the pricing calculator runs from a small team up to a stated ceiling of two hundred and fifty or more representatives, which is the only published size signal. Beyond that the coverage half is empty.

No customer count, no region or territory, no language coverage for a product transcribing and analysing conversations, no industry statement despite two named customers sitting in healthcare, and no description of who this product would be wrong for. A four week path to measurable results is published as a timeline, and nothing states the team size or data maturity it assumes.