AI Insights
Most AI products should launch with a hybrid price: a predictable platform fee that covers your fixed costs, plus a variable component tied to the usage or outcome that actually drives your inference bill. Pure per-seat pricing is the wrong default for anything agentic, because the better your product works, the fewer humans a customer needs. Pure usage or outcome pricing is the wrong default for a first launch, because you do not yet have the data to price it safely. This guide walks through the three base models, the numbers real vendors are charging, and a decision framework we use with founders in Melbourne and Sydney who are pricing an AI product for the first time.
Per-seat pricing pays you for logins, not for the work your product does, and AI products increasingly do the work. That creates a direct conflict: a support agent that resolves tickets reduces the number of support seats a customer buys, so a seat-priced vendor is rewarded for building something that works less well. Investors called this early. Andreessen Horowitz’s enterprise team argued in its December 2024 newsletter that AI is driving a shift towards outcome-based pricing precisely because the software is now replacing labour rather than assisting it.
The market has followed. Kyle Poyar’s 2026 State of B2B Monetization survey of more than 230 software companies found that hybrid pricing is now the single most common primary structure, at 37% of respondents. Seat-only is shrinking, not disappearing: it still fits products where a human is the unit of value, such as an editor, a design tool or an internal copilot used by a fixed team.
Neomeric, a Melbourne-based AI product and consulting company — and the team behind NeoMind, Australia’s onshore AI teammates platform — has priced products across all three models. Our own NeoMind uses a hybrid: a flat monthly fee per AI teammate, with fair-use limits rather than metered tokens, because Australian small businesses told us they would rather pay a little more for a bill they can predict.
The three base models are per-seat (pay per human user), usage-based (pay per unit of consumption such as a token, action, minute or conversation) and outcome-based (pay only when a defined result is achieved). Almost every AI product in market is one of these or a blend.
Fixed price per named user per month. Easy to forecast, easy to sell to procurement, and it maps cleanly onto how customers already buy software. The weakness is the misalignment above: your costs scale with usage while your revenue scales with headcount, so a heavy user on a cheap seat can be unprofitable.
Price per unit consumed. Salesforce is the clearest public example: its Agentforce pricing documentation describes Flex Credits sold at US$500 per 100,000 credits, with a standard agent action consuming 20 credits, alongside a per-conversation option at US$2 and a per-user licence option. Usage pricing aligns revenue with your inference costs, which protects margin, but it moves budget risk onto the customer, and customers hate surprise bills.
Price per result. Intercom’s Fin is the reference case: multiple 2026 analyses of Intercom’s published pricing report US$0.99 per resolution, where a resolution is a conversation the AI answered and the customer either confirmed it helped or left without asking for more. Outcome pricing is the easiest to sell on value and the hardest to operate: you need an unambiguous definition of the outcome, telemetry to prove it, and a dispute process for when the customer disagrees that the outcome happened.
Honest cost benchmarks, the hidden costs vendors don’t quote, and a 10-line scoping worksheet.
Get the free Australian AI MVP Cost Guide 2026 — we’ll email it straight to you.
Buyers say they want consumption and outcome models, but they behave as if they want predictability. Futurum Group’s 1H 2026 buyer survey found 43% of buyers prefer consumption-based models and 27% favour outcome-based structures. Yet the same research notes that vendors are converging on hybrids, because a pure consumption model makes budgeting impossible for a CFO and a pure outcome model makes the vendor carry all the delivery risk.
In our experience the Australian SMB buyer is even more predictability-sensitive than the enterprise buyer. A trades business or a clinic will accept a slightly higher flat fee over a cheaper metered one almost every time, because a variable bill is one more thing to manage. Enterprise procurement teams, by contrast, are now used to consumption pricing from their cloud contracts and will negotiate committed-use discounts.
Answer four questions in order: who is the unit of value, can you measure the outcome unambiguously, how volatile is your cost per unit, and how mature is your usage data. The answers point to a model.
| Question | If the answer is… | Lean towards |
|---|---|---|
| Who is the unit of value? | A human using a tool | Per-seat (with usage fair-use caps) |
| An agent doing work | Usage or outcome | |
| Can you measure the outcome unambiguously? | Yes, and the customer agrees on the definition | Outcome-based |
| No, or it depends on the customer’s process | Usage-based | |
| How volatile is cost per unit? | High (long contexts, voice, multi-step agents) | Usage or hybrid with hard caps |
| Low and predictable | Flat fee is safe | |
| How much usage data do you have? | Less than three months of real customers | Hybrid with generous fair-use, revisit at 90 days |
| A year or more across segments | Whatever the data says maximises margin |
Two of those questions depend on knowing your cost per unit, which is why we build a cost model before we build a price list. If you have not done that yet, our six-step guide to building an AI product cost model is the prerequisite for everything below.
A safe hybrid has three parts: a platform fee that covers your fixed costs at zero usage, an included allowance sized so that a typical customer never touches the meter, and an overage rate priced above your marginal cost with a hard cap the customer sets. The platform fee is what keeps you alive; the allowance is what keeps the customer calm; the overage is what protects you from the heavy tail.
Add up hosting, observability, support, compliance overhead and your own time, divide by the number of customers you realistically expect at month twelve, and that is the floor for the platform fee. Anchoring to a competitor’s price before you know your own floor is how founders end up subsidising every customer.
AI usage is heavy-tailed. A handful of customers will consume ten times the median. Size the included allowance around the 70th to 80th percentile of expected usage so most customers never see an overage line, and let the top decile pay for what they use.
Your marginal cost per request is knowable: model tokens, retrieval, voice minutes, tool calls. Price overage at a multiple of that, and give customers a spend cap they control. A cap you set for them is a limit; a cap they set is a feature. Techniques such as prompt caching, model routing and output limits can drop your marginal cost substantially, which is covered in our guide to AI API cost optimisation.
Never meter in tokens for a non-technical buyer. Meter in conversations, calls, documents, bookings or minutes: units the customer already counts in their own business. Salesforce moved from a single per-conversation price to credits partly because a single support interaction can trigger many backend actions; the lesson for a smaller vendor is to pick one meter, define it in plain English, and show it on the invoice.
Outcome pricing works when the outcome is binary, verifiable inside your own product, and valued far above your cost to deliver it. A resolved support ticket, a booked appointment, a qualified lead or a completed document review can all qualify. A “better decision” or “time saved” cannot: you cannot prove it, and the customer will not pay for what they cannot see.
The operational cost is real. You need event-level telemetry to prove each outcome, a clear rule for partial outcomes and hand-offs, a dispute process, and a floor fee so that a customer who integrates badly does not cost you money for months. Intercom’s model illustrates the trade-off: an “assumed” resolution bills when the customer leaves without asking for more, which is efficient for the vendor but is also the definition customers push back on. If you go outcome-based, publish your definition and make it generous in the customer’s favour, because a disputed outcome costs more in trust than it earns in revenue.
Price in AUD, include GST clarity on every plan page, and treat onshore hosting and Privacy Act compliance as part of the value you charge for rather than an add-on. Australian SMBs buying AI products are increasingly asking where data is processed; if you host in an Australian region and can say so, that is a pricing lever, not just a compliance checkbox. Enterprise buyers regulated by APRA have been operating under CPS 230 since 1 July 2025 and will expect you to carry the cost of vendor-risk paperwork, so price enterprise tiers with that overhead in mind.
One more local reality: the Privacy Act’s automated decision-making transparency obligation commences on 10 December 2026, per the OAIC’s consultation page. If your product makes or substantially assists decisions about individuals, your customers will need to describe that in their privacy policies, and the vendors who make that easy will win the deals.
If you are deciding between building this yourself and buying a platform, our build vs buy AI guide covers the trade-off, and if you are pricing a first version our AI MVP development cost guide gives you the AUD benchmarks to plan against.
Only if a human user is the unit of value and usage per user is fairly predictable. For agentic products that do work on the customer’s behalf, per-seat pricing punishes you for building something effective. A hybrid with a platform fee and a usage or outcome component is the safer default.
A fixed platform or subscription fee plus a variable component tied to usage or outcomes. It is now the most common primary pricing structure among software companies surveyed in 2026 because it balances vendor margin protection with customer budget predictability.
Size it around the 70th to 80th percentile of expected customer usage so most customers never hit the meter, then price overage above your marginal cost and let customers set their own spend cap.
Usually not. It requires an unambiguous outcome definition, proof telemetry and a dispute process, plus enough usage data to price it safely. Launch hybrid, gather ninety days of data, then consider adding an outcome tier where the outcome is clearly measurable.
Not for non-technical buyers. Meter in units the customer already counts, such as conversations, calls, documents or bookings, and define the unit in plain English on the invoice.
Neomeric is a Melbourne AI product studio — 7+ products shipped, including our own. Start with a free 15-minute scoping call, or a 2-week Build Sprint at A$6,900 fixed, fully credited toward your pilot.
What an AI MVP really costs in Australia in 2026 — line-item budgets, the traps that blow them out, and how to scope a build that pays for itself.
AI Insights
AI for dental practices in Australia: what AI teammates handle, what stays clinical, Ahpra and privacy rules, onshore hosting, setup time. Read the guide.
AI Insights
How to build an AI product cost model: trace one request, price tokens with caching, model real usage, add non-token costs, stress-test margin. Start now.
AI Insights
AI deployment options for Australian builders: direct API, Bedrock, Vertex, Azure, self-hosted or hybrid. What stays onshore, what it costs. Read the guide.