The startup marketing stack, reviewed

The buying decision moved from seat count to consumption. HubSpot shifted its Breeze Customer Agent and Prospecting Agent to outcome-based credit pricing on 14 April 2026, at 50 credits per resolved conversation and 100 credits per recommended lead, with credits at US$10 per 1,000 monthly or US$9 annually. That prices a resolved support conversation at roughly US$0.50, down from US$1.00, and a recommended lead at roughly US$1.00.

Salesforce sells Agentforce three ways at once: US$2 per conversation, Flex Credits at US$500 per 100,000 with roughly 20 credits consumed per action, and per-user licences from US$125 a month. Salesforce has reported 18,500 Agentforce deals with about 9,500 paid and US$540 million of Agentforce ARR by Q3 FY2026. The original per-conversation rate priced out SMBs and non-profits, which is why the other two models exist.

Gartner's 2026 CMO Spend Survey records the same shift on the buyer side. 56 per cent of respondents increased the share of their martech budget allocated to consumption-based pricing, while martech's overall share of the marketing budget fell to 19.4 per cent, a five-year low.

The 2026 comparative reviews converge on a stage split. Salesforce with Agentforce suits enterprise deployments with dedicated administration and complex permissioning. HubSpot wins on speed to value, unified data by default and lower operating overhead, which is the profile of a company that does not have a systems administrator and is not going to hire one this year.

For an early-stage company the practical test is not the feature matrix. It is whether the platform can be configured against the go-to-market blueprint by the people already on the team, and whether the agent spend can be predicted well enough to sit in a CAC model at roughly US$0.50 per resolved conversation. Outcome-based pricing makes that easier to forecast than seat licences did, because the cost moves with the volume it produces.

The 2026 Marketing Technology Landscape counts 15,505 products against roughly 150 in 2011, and growth has finally stopped at 0.79 per cent. Underneath that, 1,488 products were added and 1,367 removed, so category leadership is less stable than the headline suggests. A stack chosen on a feature comparison will be re-litigated within two years.