Demandbase

An incumbent from 2006 with a genuine model layer underneath and a recent agent layer on top, and the removal test turns on which one is meant. Stripping the agents leaves account identification, intent scoring, buying group detection, the advertising platform and the orchestration engine all working, since the vendor itself separates the two and calls the scoring layer the trustworthy data driven AI customers already rely on.

That places it with the established platforms rather than the AI natives. The scoring layer is more than a rules engine, resolving anonymous web traffic to companies and ranking them on many signals, but it long predates the current marketing and the company sold the same function before the term was applied to it. Ask which capabilities disappear if the agent layer is disabled at account level, and whether the agents carry separate cost.

An administrative control that is named rather than implied, and an architecture described in enough detail to reason about. The vendor states that access to the agent layer is governed at the tenant and account level, that administrators can enable or disable those features for the whole account, and that it is expanding administrative controls over time. Naming the off switch and who holds it is the specific thing most vendors in this cohort leave out.

The described architecture reinforces it: queries route through a gateway, and mandatory governance and guardrail checkpoints are described as structural stages in the workflow rather than as policy. What the agents do is also comparatively bounded, since they summarise accounts, draft communications, research and orchestrate plays rather than sending to buyers on their own authority, and the advertising optimisation they perform spends budget rather than contacting people directly.

Held below the top band because the control described is binary and account wide, with no published per role or per agent constraint, no stated spend limit on autonomous campaign optimisation, and no runtime stop for work already in flight. Ask what an agent can change in a live campaign without approval, whether a spend ceiling applies, and whether controls exist below the account level.

The vendor names its model infrastructure, which almost nothing else in this cohort does. The AI principles page states that the agent layer runs on a named third party model hosting environment, describes the architecture as a mixture of experts system with dedicated components for retrieval augmented generation, structured query, web search and actions, and names the prompting strategies used to structure outputs.

It also states that observability and evaluation are built into the orchestrator so that reasoning and data extraction can be monitored and validated. Naming the host, the architecture pattern and the evaluation hook is materially more disclosure than a vendor asserting that it uses third party models meeting its standards. Two gaps hold it below the top band. No specific model, family or version is named within that environment, so a buyer knows the platform but not the model.

And no accuracy, error rate or evaluation result is published for either the agents or the account identification and intent scoring that the platform is actually bought for, which is the more consequential absence of the two. Ask which models are in use within the hosting environment, and for measured identification accuracy and match rate.

Named enterprise customers in quantity, and outcome measurement that stops at the case study. A large logo wall carries household names across technology, financial services, life sciences and publishing, and a customer story library sits behind it, so the installed base is not in doubt.

What is absent is any study measuring pipeline or revenue effect across a stated population over a stated period, which for a platform whose entire proposition is that account based targeting outperforms broad demand generation is the measurement the category exists to argue about.

Independent review analysis converges on a consistent operational finding that the vendor does not surface: breadth makes onboarding and day to day management demanding, deployments underperform when treated as plug and play, and successful use assumes a dedicated operations team or an external partner. That is a real cost of ownership signal from outside the vendor. Ask for pipeline influence measured against a holdout rather than platform attributed influence, and for median time to first campaign.

The platform orchestrates rather than sends, and the compliance surface it does own is advertising rather than messaging. Sequences and email execution happen in the customer's own marketing automation or sales engagement system, so consent capture, suppression and unsubscribe handling sit with those systems and their operators.

What Demandbase controls is ad delivery to identified accounts, where the obligations are the advertising industry's rather than the messaging statutes, and a cookie policy and preference centre are published for the vendor's own site.

What is absent is any statement of what the orchestration layer enforces when it pushes an audience or a play into a downstream system: whether suppression state travels with the audience, whether a person who has objected is excluded from an ad segment, or whether that is entirely the customer's problem. Ask whether opt out and suppression state is carried into audiences pushed to downstream execution systems and to the advertising platform.

A complete instrument set with real specificity in the numbers, and the substance sits behind a request. The trust site names cross border transfer documentation, a data protection officer, a cookie disclosure, a subprocessor list, a processing addendum and the master subscription agreement, so a buyer can see the full shape of what governs the relationship. A privacy centre with a rights request tab and a do not sell route is published and operated through a consent management platform.

The security policy states retention in figures rather than in purpose language, committing to hold customer data no longer than thirteen months after termination and to delete within thirty days of written request, and the vendor repeats those same figures on its AI page, which is the kind of consistency across documents that suggests a real policy rather than page level drafting.

The gap is verification and it is specific to what this vendor does: the platform resolves anonymous individuals to companies and runs an advertising exchange, so the lawful basis for identifier collection and the onward sharing to advertising partners are the load bearing questions, and neither the transfer documentation nor the subprocessor list is readable without an approved request. Ask for the subprocessor list and the transfer documentation, and for the stated lawful basis for identifier collection used in web identification.

Three acquired data assets and a live advertising exchange, and the origin of the underlying records is not described. The platform now carries technographic data from one acquisition and contact and company intelligence from another, both folded into the product, and neither the collection method nor the licensing basis for those inherited corpora is stated anywhere on the surface.

Intent is described as a combination of first party website signals and third party sources without naming a cooperative, publisher or provider. The advertising side adds an unusual disclosure that cuts the other way and is worth recording: the security policy explicitly identifies digital advertising metadata as a data class it encrypts, describing cookies and the first three octets of an internet protocol address as identifying fields shared with exchanges and participating websites.

That is a candid technical description of what actually moves, and truncating the final octet is a recognised minimisation choice. It describes handling rather than provenance. Ask which providers and cooperatives supply third party intent, what the licensing basis is for the acquired contact and technographic corpora, and whether provenance is indemnified.

Integration and activation both run through sanctioned routes, and the vendor carries less platform dependency than most of this category because it owns its own advertising infrastructure. Connectors into customer record and marketing automation systems are native and bidirectional, published through official channels with a partner integration directory, and the advertising layer runs on the vendor's own demand side platform rather than reselling access through someone else's, which means the delivery path is owned rather than borrowed.

A named social platform integration operates as a partnership. Nothing resembling credential storage, scraping or unsanctioned automation appears anywhere. Held below the top band because the exposure that does exist is unaddressed in public: web identification depends on third party identifiers whose availability is set by browser vendors rather than by Demandbase, and no published position describes what happens to identification coverage as those identifiers are restricted further.

Owning the exchange does not remove that dependency, it relocates it. Ask what proportion of identification depends on third party cookies today, and what the stated plan is as browser restrictions tighten.

A training boundary stated in unusually plain terms, with the retention numbers to go with it. The vendor states that customer data is used only to train and fine tune models for that customer's own account, that it does not use customer data to train large language or generative models at all, and that proprietary customer data is never used to train third party models accessed through its named hosting environment.

Those are three distinct commitments rather than one claim restated, and the third is the one that matters most because it addresses the vendor's actual supply chain rather than its own systems. Retention is quantified: customer data held no more than thirteen months after termination, deleted within thirty days on written request, and agent chat history purged after twenty eight days. A separate governance policy for the AI management system is referenced by name.

What holds it below the top band is instrument tier rather than content. All of this sits on a marketing page, the governance policy itself is not published, no independently audited management standard for artificial intelligence is claimed, and the two vendors at the top of this axis both hold one. Ask whether the no training commitment appears in the subscription agreement, and request the governance policy.

The buyer facing output is advertising rather than correspondence, which changes the question rather than removing it. A person at a targeted account experiences this platform as display, video or connected television advertising served because their employer was identified and scored, and advertising carries its own disclosure conventions that are broadly understood, so nobody is being deceived about who is speaking.

The agent layer does draft personalised communications, but those are sent by a human through a downstream system under that person's own name. What is unaddressed is the identification itself, which is the part a recipient would find surprising: a person browsing is resolved to their employer, added to a buying group, scored, and targeted, and nothing published describes what notice that person receives or how they would learn any of it happened.

A do not sell route exists on the vendor site and would not be found by someone who has never heard of Demandbase. Ask what notice a person at a targeted account receives about identification, and whether AI drafted content is marked as such for the seller who sends it.

Bidirectional depth into the systems that matter, plus an owned activation channel most competitors rent. Native two way synchronisation runs with three major customer record platforms rather than one, which is broader than most of this category manages, alongside marketing automation connectors and a published partner integration directory on its own subdomain.

The advertising exchange is itself an integration surface, connecting to programmatic supply, a named social platform and connected television inventory from inside the same system, so a customer activates without assembling separate ad technology. A public status page and a customer community are maintained. Two limits hold it below the top band. No developer documentation or interface reference was reached on the routes taken this pass, so the programmable surface is unverified.

And the independent review evidence consistently reports that realising this breadth requires dedicated operations capability, which is a real constraint on the depth being usable rather than merely present. Ask for the interface documentation and the current verified integration inventory.

The infrastructure is described in more technical detail than most vendors offer, and the residency question is still unanswered. The security policy names two public cloud providers as primary platforms, states that the estate spans multiple regions and multiple availability zones within each region for redundancy and recovery, describes the shared responsibility split, and specifies encryption at rest through both providers key management services with keys held in those services or a named software vault.

Environments are stated to be physically and logically separated by function. That is a real architectural description rather than a badge. What is missing is the customer facing part: no region is named, no residency election is described, no tenancy model is stated, and no recovery time or recovery point figure is published even though continuity and recovery plans are said to be tested annually.

Multi region is an availability claim, not a residency answer, and a buyer cannot tell from it where European data would sit. Ask which regions data can be pinned to, whether a European residency option exists, and what the recovery time and recovery point objectives are.

Real certifications, a named auditor, and every artifact behind an agreement. The vendor holds certification to the international information security standard in its 2022 revision and a service organisation control type two report, both recertified, with the audit firm named in its own announcement and the certification scope stated as covering the platform together with its supporting systems, development and operations.

Naming the auditor and the scope without being asked is better practice than most of this cohort. A live trust site lists the report, a penetration test report, a vulnerability assessment, two industry self assessments, a subprocessor list, the processing addendum and the master subscription agreement, and a separately published security policy describes the control environment in genuine technical detail including access control, logging, development lifecycle and encryption.

Everything substantive requires a confidentiality agreement, including the report itself and the penetration test summary. One caution recorded rather than graded: a third party vendor risk aggregator lists this vendor as holding several further certifications including a United States federal authorisation and a health information framework. Nothing on the vendor's own surface claims either, and the vendor's own instrument names only the two. The aggregator appears unreliable and its claims were not used. Ask for the report with its audit period, and the certificate with its scope and expiry.

The pricing model is explained clearly and no number appears anywhere in the explanation. The page is better than most at this grade and the distinction is worth stating precisely: rather than listing packages, the vendor answers how pricing works in prose.

It states that the flagship platform combines a platform fee covering the software and services with a flat fee per user, that advertising can be bought as a starting point, that the data products can be licensed standalone, and that the platform is deliberately not split into separate marketing and sales products. A buyer therefore understands the shape of the bill, which is real disclosure, and cannot estimate its size, which is what this axis measures.

No platform fee, no per user rate, no minimum and no advertising spend commitment is published, and the only path is a form. Third party procurement figures spread widely, from roughly eighteen thousand dollars at the low end through a median near sixty five thousand to well past two hundred thousand for advertising heavy deployments, with independent analysis reporting that media costs on the vendor's own exchange run above standard programmatic rates. Fourth consecutive vendor at this grade. Ask for the platform fee, the per user rate, the minimum commitment, and the media margin on advertising spend.

Two exit commitments published as specific numbers, which is rare enough on this axis to carry the grade. The security policy states that customer data is retained for up to thirteen months following expiry or termination and that on written request the vendor will delete it within thirty days, and the AI page repeats both figures and adds that agent chat history is purged after twenty eight days.

A stated deletion window a customer can hold a vendor to is more than almost anything else in this index offers here, and the consistency of the figures across two independently maintained pages is itself a small signal of reliability. The master subscription agreement is named on the trust site and is therefore obtainable rather than hypothetical. What is missing is scope.

Nothing states what a departing customer can export, in what format, or whether account scores, engagement history, audience definitions and advertising performance data leave with them, and that last category is the exposure specific to this platform, since attribution history accumulated on the vendor's own exchange has no obvious home elsewhere. Ask what exports and in what format, specifically whether engagement scoring history and advertising attribution data are included.

The platform runs no sending infrastructure, so this axis applies to delivery of a different kind and the vendor's own instrument acknowledges it. Email execution happens in the customer's downstream marketing automation or sales engagement system under that customer's domains and reputation, so none of the warmup, rotation, bounce or complaint machinery this axis normally examines belongs to Demandbase.

The analogous discipline is advertising delivery, where the vendor operates its own exchange and where the equivalent questions are inventory quality, viewability, invalid traffic filtering and brand safety controls. Its security policy does name bot detection as a maintained control, which touches the invalid traffic question from the security side.

Nothing published addresses viewability standards, fraud filtering rates, or what happens to spend delivered against invalid traffic, which is the money equivalent of a bounce rate on this platform. Ask what invalid traffic filtering and viewability standards apply to delivered media, and how spend against invalid traffic is credited.

The market is stated explicitly by role, industry and segment, and the independent evidence sharpens rather than contradicts it. The vendor publishes solution pages for four buyer roles covering marketing, sales, customer success and revenue operations, four named industries covering business services, financial services, manufacturing and technology, a distinct enterprise segment page, and six use case pages spanning foundation building through expansion and retention.

It also publishes direct comparison pages against three named competitors, which is an unusually direct statement of the competitive frame it accepts. The customer evidence matches the enterprise claim with named references across technology, financial services, life sciences and publishing.

What holds it below the top band is a gap the vendor does not close itself: independent analysis places the platform as purpose built for organisations above roughly one hundred and fifty million dollars in recurring revenue, and reports that pricing and complexity put it out of reach for most smaller and mid market teams, while the vendor's own pages describe flexible pricing that grows with the team. Those two statements point different directions and only one of them comes from someone with no stake. Ask what the smallest viable deployment is in users and total commitment.