Maxpertise does not charge per token, credit, or seat because a small business cannot budget against a bill it cannot predict, and a flat monthly retainer for a function that runs solves that problem directly. Much of the AI and SaaS market is moving the other way: billing tied to model usage, active seats, or credits consumed, which shifts the unpredictability onto the buyer. We build and run AI teams on the opposite model: one retainer per role, scoped in a written proposal, no meter running underneath it.
Key Takeaways
It means the bill moves with activity: a request processed, a login active that month, a lead engaged, each one adds to the total. Outcraft AI's new inbound sales-agent billing replaced flat monthly license and per-minute tiers with a straight per-lead charge, counted only once its AI SDR has engaged that lead, at rates that step down from about $3 per lead at low volume to under $1 per lead at scale (GlobeNewswire, September 2026). That is not a hypothetical trend description; it is a named vendor's actual pricing page as of this month.
The appeal to a vendor is obvious: billing scales with the vendor's own delivery cost, and it lets a buyer start small. The cost to the buyer is less obvious until the invoice arrives: the total is not knowable in advance, because it depends on how much work the agent ends up doing, not on what was agreed.
Because a small business has no forecasting team to model the range of outcomes, and no slack in the budget to absorb the high end of a metered bill if usage runs hot. An enterprise buyer with a dedicated FinOps function can model a usage curve, negotiate committed-use discounts, and staff someone to watch the meter. A five-to-eighty-person business does not have that person. The bet a metered bill asks them to take, that this month's usage lands somewhere reasonable, is a bet made with information they were never given the tools to evaluate.
That gap matters more right now than it would have two years ago. McKinsey's 2025 research found about 80% of companies report using generative AI already, yet an equal share report no significant bottom-line impact, and roughly 90% of the more transformative, function-specific use cases are still stuck in pilot (McKinsey, June 2025). Metering a bill against usage before the business has proof the usage is paying off is asking the buyer to underwrite the vendor's own unproven case.
Because the client is buying a function that runs, not a quantity of model activity, and pricing it as a function keeps the incentive aligned with the outcome instead of with the volume. We do not publish a rate card, and we never structure an engagement around metering a client's requests, active users, or usage volume. The proposal that comes back after a written intake is one monthly retainer per role, scoped to what that role actually needs to do, three-month minimum, first month risk-reduced, 30-day proof guarantee.
The engineer or role is not incentivized to generate more billable activity, because there is none to generate. If a role can be built to do the job in fewer steps, that is a better outcome for everyone, not a smaller invoice to chase. Whichever lane runs the work, managed or embedded, the code and the data stay yours, with a clean exit and no lock-in whenever you want one.
A function that runs: one of the five AI team roles, back office, operations analyst, lead qualifier, account manager, or SDR, scoped in a written spec before anything is built, delivered under two human gates, you approve the plan, you review the work. See how it works for the full method, or how to get an AI team for your small business for how the roles get chosen and sequenced.
This holds regardless of industry. Whether the role is chasing invoices in accounting or qualifying leads in HVAC, the retainer is scoped to the function, not metered against how many times the agent ran that week; see the industries pages for how it plays out by vertical.
No. The retainer is scoped to a defined role and set of workflows in the written spec, not to an unlimited volume of arbitrary requests. If the scope needs to grow, that is a new spec and a new conversation, not a meter ticking upward without your knowledge.
Not always. It can make sense when usage is genuinely unpredictable and low-stakes, and the buyer has the tooling to monitor it. For a small business budgeting against a fixed monthly number with no dedicated finance function watching a dashboard, a flat retainer removes a risk that usage-based pricing does not remove, it just relocates it onto the buyer.
Scope drives price: what the role needs to do, how many workflows it covers, and what data it needs to connect to. There is no published rate card because scope varies too much for a single number to be meaningful; the number arrives in a written proposal after a written intake, no call required.
Because it is a live, dated example of where the broader AI-agent market is heading, not a claim we are making up to contrast against. Seeing an actual vendor's per-lead metering, launched this month, makes the tradeoff concrete instead of abstract.
A meter tells you what happened after the invoice arrives. A retainer tells you what you are paying before the month starts. Send a written intake describing the function you need run, and the proposal that comes back is a single number, not a formula.
Internal links to add from older posts within a week: how-to-get-ai-team-for-small-business (anchor: "AI team roles"), how-much-cost-hire-ai-engineer-2026 (anchor: "what hiring an AI engineer costs"), ai-sdr-per-lead-pricing-real-example (anchor: "usage-based AI SDR pricing", once that post exists).
Maxpertise is an AI-native engineering company. We embed native AI engineers inside your team, live in about 10 days. Please enable JavaScript to view the site, or email contact@maxpertise.net.