Every ranked list of agentic marketing platforms is written by someone selling into the category, and at least three of them rank their own product first. Here is the buyer's side: what counts as agentic, what the enterprise tier actually costs, what the evidence does and does not show, and the point at which a platform starts to earn its money.
To evaluate an agentic marketing platform, score it on ten criteria and treat four of them as gates: bounded authority, auditability, data readiness and exit. A high score everywhere else cannot compensate for a zero on those four.
Before any of that, apply one test. The software must be able to take an objective, act on your connected systems, read what happened, and change its next action as a result.
That loop is the only thing every published definition of the category agrees on. It is also the only thing you can actually watch in a demo.
As of September 2026 the first page of results for this category carries at least six ranked lists. Netcore, Profound, Superscale, AdMesh, Hellyeah and DigitSelect all publish one, and at least three of them place their own product at the top.
They do not agree on what the category contains. Superscale disqualifies HubSpot Breeze; Netcore recommends starting with it. Two lists cannot both be right about the same product, and neither is a neutral referee.
There is no neutral referee to appeal to. Gartner has published no Magic Quadrant for agentic marketing platforms and Forrester has published no Wave.
Gartner's public position on the wider field is a warning rather than a definition. It named the practice of relabelling existing software as agentic, and estimated in June 2025 that only around 130 of the thousands of vendors using agentic language were genuinely in the category. That is an estimate, not a count, and it should be read as one.
If the firm whose business is defining software markets has declined to define this one, a buyer's rubric is not presumptuous. It is the only instrument available.
Four definitions are in circulation and every one of them traces to a party selling into the category. The incompatibility between them is itself the finding.
Salesforce describes agents that execute, adapt and optimise in real time on unified data. Hightouch defines the category as delegating entire marketing workflows, from research and planning through to execution and optimisation, with human review and approval built in.
Read that second one twice. Entire workflows delegated, and human review built in, inside the same sentence. The tension is not sloppy drafting; it is the category admitting what it can currently do.
The common thread is the closed loop. Strip the marketing language from all four definitions and what remains is act, observe against a goal, change the next action.
So make the demo prove the loop. Ask to see the system change its own next action in response to a result it did not predict, and ask what it would have done differently if the result had gone the other way.
If every path ends in a recommendation for a person to approve, you are looking at a copilot. That may still be the right purchase. It is not the purchase you were quoted for.
Most of the category does not publish a price, and the pattern of who publishes what is informative on its own.
Salesforce publishes. Marketing Cloud Next Growth Edition lists at £1,200 per organisation per month billed annually, which is £14,400 a year before channels, contacts, add-ons, implementation and support. The equivalent US list price is $1,500 a month, or $18,000 a year, with the Advanced edition at $3,250 a month.
HubSpot publishes, and prices some agents by outcome rather than by licence. Its Customer Agent is billed at roughly $0.50 per resolved conversation and its Prospecting Agent at roughly $1 per lead recommended for outreach.
Superscale publishes a self-serve ladder running from free to $799 a month. Adobe, Netcore and Iterable publish nothing for their agent layers.
Hightouch is the cleanest illustration of the category's opacity. Its composable CDP line carries real numbers on the pricing page. Its Agentic Marketing Platform line, the thing the company's home page leads with, carries none.
No vendor in the set publishes an implementation or onboarding fee. That absence is the finding, and it is worth saying to a sales team out loud rather than accepting an estimate.
The prerequisite is where the real cost sits, and Adobe states it contractually rather than as guidance. Its product description, effective 30 July 2026, requires the customer to maintain a licence to the underlying products and services in order to retain access to the agents, and specifies that AI credits are licensed annually and do not roll over.
Salesforce routes agent capability through its data layer on the same logic. A company without a resolved customer data layer is not buying a platform. It is buying a second project.
Merkle, an Adobe implementation partner rather than a critic, put the consequence better than any sceptic has. An agent orchestrating journeys on incomplete or poorly resolved customer profiles, it wrote after Adobe Summit 2026, accelerates bad decisions at scale.
Score each criterion 0, 1 or 2. The four marked as gates are pass or fail regardless of the total.
The gates exist because the failure modes in this category are not feature gaps. They are an agent with authority nobody bounded, acting on data nobody resolved, leaving no trail, inside a contract with no exit.
There is no independent controlled comparison of agentic and human-run marketing. The strongest controlled evidence in the category is authored by a vendor about its own infrastructure, and that is worth knowing before anyone quotes it at you.
The study is an eleven-month trial published at the ACM UMAP conference in June 2026, covering roughly 8.8 million users of a consumer delivery application with a persistent 90/10 holdout. Four months of human-curated operation were followed by seven months in which the agents ran autonomously from a fixed library.
The autonomous phase sustained a 57% lift in notification clicks and a 2.4% lift in days with engagement. The authors describe human management as a performance premium worth roughly 12 to 26%.
Now the part that does not make the vendor decks. In the autonomous phase, the lift on days with intent and conversion events was 0.2% with a confidence interval of plus or minus 0.27, which is not distinguishable from zero. The trend was declining by the end of the eleven months.
Read honestly, the best controlled evidence in this category shows a very large lift on engagement proxies and a lift at or near the noise floor on the outcomes a business actually books. All three authors work for the vendor whose system was tested, on a consumer app that looks nothing like a B2B SaaS company.
The other useful dataset points the same way. Haus ran 640 geo-lift incrementality experiments across brands spending an average of around $14m a year on Meta, and found the automated buying product beat manual in only 42% of tests, at 12% lower incremental return, as reported by AdExchanger in October 2025.
The buried headline there is the spend level. The evidence base for autonomous decisioning was generated by companies nothing like the reader of this page.
No published evidence supports a threshold based on ARR, contact volume or headcount. Anyone quoting you one has made it up, including anyone quoting you ours.
Two variables do carry weight. The first is whether you already have a resolved customer data layer, which Adobe enforces contractually and Salesforce and Hightouch assume architecturally.
The second is under-argued everywhere else: whether you can run a holdout. If you cannot measure incrementality, you cannot tell whether the platform worked, and both datasets above show platform-reported metrics diverging from incremental reality.
The cost side can at least be anchored in arithmetic. Gartner's 2026 CMO Spend Survey of 401 marketing leaders puts martech at 19.4% of marketing budget, inside a total marketing budget averaging 7.8% of revenue.
Apply those two published averages to a £5m ARR company and you get roughly £390,000 of marketing budget and roughly £76,000 of martech budget for the entire stack. An $18,000 licence is about a quarter of that, before implementation. At £1m ARR it is roughly the whole thing.
That is arithmetic on population averages rather than a measured finding, and it should be treated as a sanity check rather than a rule. Alongside it, Gartner reports martech utilisation at 49%, which means the base rate for adding an expensive tool to an under-used stack is poor.
So the rule we would apply, stated as ours rather than as an industry finding: an enterprise agentic platform starts to earn its cost when there is a repeated, measurable decision running at enough volume for optimisation to compound, connected and governed data to ground it, a named person who owns its operation, a bounded pilot with a real baseline and a clean exit, and expected value that comfortably beats the full annual cost rather than the licence cost.
One narrowing matters, and most versions of this argument get it wrong. That is a threshold for enterprise platforms, not for agentic capability.
Agentic capability is available well below the price of an enterprise platform, and the pricing models are the reason.
Outcome and metered pricing removes the commitment problem. HubSpot's agents bill per resolved conversation and per recommended lead, so the cost scales with work completed rather than with a seat count negotiated in advance.
Self-serve ladders remove the sales-process problem. Superscale publishes every tier from free upward, which means a team can establish whether the loop works before anyone signs anything.
Platform-native agents cost nothing beyond the media you were already buying. They are also the least bounded thing in your stack, which is exactly why the gates in the rubric apply to them too.
There is a real argument against this approach and it deserves stating. MIT's NANDA research found that buying AI tools from specialised vendors succeeded roughly 67% of the time, while internal builds succeeded about a third as often, and an assembled stack is structurally closer to a build than to a purchase.
The answer is to keep every piece a purchase. Buy narrow products with published prices and clean exits, and resist the temptation to wire them together with custom code you will then have to maintain.
Ask these before the demo, and ask for the answers in writing rather than in a call.
A vendor that answers all eight in writing is worth a pilot. A vendor that will only answer them verbally has told you something about the contract you are about to sign.
The organisation that deployed it. Autonomy does not transfer liability to the software vendor, and no regulator has suggested otherwise.
The EU AI Act's transparency obligations under Article 50 have applied since 2 August 2026 and fall on deployers as well as providers. The high-risk obligations under Annex III do not apply until 2 December 2027, and ordinary marketing personalisation is not automatically caught by them, whatever the headlines suggest.
In the UK, the Data (Use and Access) Act changes took effect on 5 February 2026, replacing the old Article 22 regime on automated decision-making with Articles 22A to 22D. The practical implication for a buyer is that the audit gate in the rubric is not a nice-to-have.
If you cannot show what the agent decided and why, you cannot answer a complaint about it.
What is an agentic marketing platform?
Software in which AI agents take an objective, act on connected marketing systems, read the results and change their next action within limits a person set. The closed feedback loop is the only element every published definition of the category shares. There is no independent or analyst definition, and no Magic Quadrant or Forrester Wave covers the category as of September 2026.
How much does an agentic marketing platform cost?
Most vendors do not publish a price. Salesforce Marketing Cloud Next Growth Edition lists at £1,200 per organisation per month billed annually, which is £14,400 a year before channels, add-ons, implementation and support. HubSpot prices some agents by outcome at roughly $0.50 per resolved conversation. Adobe, Netcore, Iterable and Hightouch's agentic line publish nothing, and no vendor in the set publishes an implementation fee.
Is there evidence that agentic marketing platforms work?
No independent controlled comparison exists. The strongest controlled evidence is an eleven-month trial of 8.8 million consumer-app users published at ACM UMAP in June 2026, authored by the vendor whose infrastructure was tested. It found large sustained lifts on engagement proxies but lifts at or near the noise floor on intent and conversion, with the trend declining by the end of the study.
How do you tell a real agentic platform from relabelled automation?
Ask it to change its own next action in response to a result it did not predict, and ask what it would have done if the result had gone the other way. Automation follows a route someone wrote in advance. If every path in the demo ends in a recommendation for a human to approve, it is a copilot rather than an agent.
When should a small B2B SaaS team buy an agentic marketing platform?
When there is a repeated, measurable decision at enough volume for optimisation to compound, connected and governed data to ground it, a named owner, a bounded pilot with a real baseline and a clean exit, and expected value that beats the full annual cost rather than the licence cost. Below that, buy agentic capability in metered or outcome-priced pieces rather than committing to an enterprise platform.
Who is liable if an agentic marketing platform breaks advertising or data rules?
The organisation that deployed it. EU AI Act transparency obligations under Article 50 have applied to deployers since 2 August 2026, and the UK's Data (Use and Access) Act changes took effect on 5 February 2026. Exportable decision logs are what let you answer a complaint, which is why auditability is a gate rather than a scoring criterion.
SaaStrix is where B2B marketers become agentic marketing leaders, and you can try it free for five days.
.svg){kind=icon}