Pricing AI features: seats, usage, credits and outcomes

Video lecture

Pricing AI features: seats, usage, credits and outcomes

16 chapters · about 9 min · full transcript

Coming soon

Chapter 1 of 16

Pricing AI features

Free for all → losses on heavy users
Pricing models for AI
Test against real usage

The narrated lecture is in production

Every chapter is scripted and ready. Browse the chapters and read the full transcript now — the video will appear here when it’s published.

Chapters

Transcript of the narration, chapter by chapter.

0:00 Pricing AI features

A social media platform added AI captions for free to every plan. Customers loved it. Then the finance team noticed that a handful of agency accounts were generating tens of thousands of captions a month, and for those accounts the feature was losing money on every plan. Nobody had done anything wrong. The pricing simply did not fit how AI costs behave. In this lesson you will learn the main pricing models for AI features and how to test them against real usage.

0:36 Why AI pricing is tricky

Why is AI pricing tricky? Two reasons. Every task has a real cost, and usage is wildly uneven: a few power users can consume ten to a hundred times the median. Here is an analogy. An all-you-can-eat buffet works when most guests eat a normal amount. If a few guests eat twenty plates every visit, the buffet needs a rule, a price for extra plates, or a separate menu. AI features in flat plans are buffets. Your job is to design the rules without upsetting normal guests.

1:14 Pricing models

Here are the options. Included in the existing plan: great for adoption, risky for margin. Per-seat add-on: simple and predictable, but heavy users can be unprofitable. Usage-based: aligns cost and revenue, but can cause bill anxiety. Credits or allowances: predictable with a safety valve for heavy use. Outcome-based: pay per resolved ticket or processed document, tied directly to value, but you need an agreed way to measure the outcome. And hybrid: a platform or seat fee plus included usage plus overage, which is where most business AI products end up.

1:53 Design steps

How to design it. Anchor on value first: what is a task worth to the customer, in time saved or revenue gained? Price well below that value and well above your cost. Know your cost distribution: the median and ninety-fifth percentile cost per customer each month. Protect margin with guardrails like fair-use limits or cheaper models for bulk jobs. Keep it explainable in one sentence. And test with new cohorts or in one market before rolling out.

2:26 Simple example (placeholder numbers)

A simple example with placeholder numbers. A proposal-writing feature saves a freelancer about thirty minutes per proposal. If their time is worth ten dollars an hour, a proposal saves them five dollars of value. Your cost per proposal is two cents. Anything from twenty cents to a dollar per proposal leaves the customer far ahead and you with a healthy margin. Pricing becomes a question of packaging, not of whether the numbers work.

2:58 Margin check

Now the margin check, and this is where most plans break. Model the monthly revenue and serving cost for your median user, your ninetieth percentile user and your ninety-ninth percentile user, under each plan. The lesson's script does exactly this. You will often find a flat seat add-on is healthy for the median user and negative for the top one per cent. That is precisely what allowances, credits, fair use and cheaper models for bulk work are for.

3:31 Regional considerations

Regional details matter. Local pricing in rupees, dirhams, riyals or pounds often converts better than dollar-only pricing, and you must handle VAT where it applies, including in the UAE, Saudi Arabia and the UK. Payment methods vary by market, and usage billing depends on reliable collection. And many enterprise and government buyers in the Gulf prefer predictable annual contracts, so usage components may need caps.

3:59 Business example: social platform (illustrative)

Now the realistic example, with illustrative outcomes. That social media platform moved to a hybrid. Each plan includes a monthly AI allowance sized to cover most users. Heavy users buy credit packs. Bulk jobs, like generating captions for a hundred posts at once, run on a cheaper model. Most customers never hit a limit, heavy users pay for what they use, and the feature's margin recovered. Clear in-app usage meters kept billing complaints low.

4:31 Communicating pricing

Communicate pricing well. Show usage in the product: credits remaining and usage by team. Alert customers before they reach limits. Explain what counts as a task or credit, with examples. And be transparent about which features use AI and any fair-use policy. The worst AI pricing experience is a surprise invoice, and it generates support tickets, churn and bad reviews.

4:57 Common mistakes

Common mistakes. Pricing on cost alone and leaving most of the value on the table. Unlimited AI in flat plans with no fair-use protection. Credit systems so complex that nobody, including your sales team, can explain them. And outcome-based pricing without an agreed, auditable definition of the outcome, which leads to disputes.

5:20 Another example: school platform in Pakistan

Another example, from B2B software in Pakistan. A school-management platform added AI report-card comments for teachers. Schools pay annually per student, and budgets are fixed. Usage-based pricing would scare principals. So the company included a generous allowance of AI comments per student per term in the premium plan, set fair-use limits that almost nobody reaches, and used a cheaper model for bulk generation. Predictable for schools, profitable for the company.

5:50 Price the value, explain the cost

One more pricing principle: price the value, explain the cost. Customers accept higher prices when they see the outcome, like hours saved or more resolved tickets. But they get frustrated when they cannot predict the bill. So lead with value in your pricing page and sales conversations, and give customers the tools to understand and control their usage. Value creates willingness to pay; transparency keeps it.

6:19 FAQ: add-on or core plan?

A question sales teams ask: should AI be a separate add-on or part of the core plan? If AI is central to the value of your product, put it in the core plans with sensible allowances, because customers will expect it. If it serves a specific group, like power users or particular departments, an add-on lets those customers pay for it without raising prices for everyone. Revisit the decision as adoption grows. Many products start with an add-on and fold AI into core plans once it becomes essential.

6:57 Try this now

Try this now. Take one AI feature and write three numbers: value per task to the customer, your cost per task, and the tasks per month for your median and heaviest users. Put them into the plan-check script for two pricing options. If either option loses money on your heaviest users, sketch the guardrail that would fix it.

7:22 Watch me do it (illustrative)

Watch me do it. Our AI proposal writer costs about two cents per proposal. I pull usage for the last month: median user writes twenty proposals, the ninetieth percentile two hundred, the top one per cent over two thousand. I model three plans in the script. Flat add-on at fifteen pounds per user: healthy for the median, fine at P ninety, negative for the top one per cent. Pure usage at ten pence per proposal: profitable everywhere, but interviews show customers dislike unpredictable bills. Hybrid: fifteen pounds including three hundred proposals, then credit packs. Margins are healthy at every percentile, and only the heaviest users ever buy credits. I check local currency and VAT handling for the UK, UAE and Saudi Arabia, draft the one-sentence explanation for sales, and propose a six-week test with new sign-ups, tracking conversion, margin and billing tickets.

8:23 Recap

Recap. AI pricing balances value capture, cost coverage and predictability. Choose from included, per-seat, usage, credits, outcome and hybrid models. Anchor on value, check margins for median and heavy users, add guardrails, respect regional realities and communicate clearly. Next module: delivering AI features with evaluation, safe launches and impact measurement.

Pricing is where AI economics meet customer value

AI features have real marginal costs and wildly uneven usage: a few power users can consume 10–100x the median. Traditional flat per-seat pricing can turn your best customers into your least profitable ones. At the same time, customers dislike unpredictable bills. Good AI pricing balances value capture, cost coverage and predictability.

The main pricing models

Model	How it works	Pros	Cons	Fits when
Included in existing plan	AI features bundled at no extra charge	Adoption, competitive defence	Margin risk from heavy use	Costs are low or usage is naturally bounded
Add-on per seat	Extra fee per user for AI features	Simple, predictable	Heavy users can be unprofitable	Usage per user is fairly even
Usage-based	Pay per task, per token-like unit, per minute	Aligns cost and revenue	Bill anxiety, harder to budget	Usage varies a lot; technical buyers
Credits / allowances	Plans include N credits; top-ups or overages	Predictable with a safety valve	Credit design complexity	Mixed usage; many feature types
Outcome-based	Pay per resolved ticket, qualified lead, processed document	Directly tied to value	Defining and measuring "outcome"; disputes	Clear, measurable outcomes (for example some support tools charge per resolved conversation)
Hybrid	Seat or platform fee + included usage + overage or outcome fee	Balances all goals	More to explain	Most B2B AI products at scale

Designing a pricing experiment

Anchor on value. Estimate value per task for the customer (time saved, revenue gained). Price well below value, well above cost.
Know your cost distribution. Median and 95th-percentile cost per customer per month under each pricing option.
Protect margin with guardrails. Fair-use limits, rate limits, cheaper models for bulk operations, or credits.
Keep it explainable. A sales rep in Riyadh or a founder in Lahore should explain it in one sentence.
Test. Try pricing with new cohorts or in one market; monitor conversion, expansion, churn, gross margin and support tickets about billing.

A simple pricing model sketch

value per task (customer)       = minutes saved × customer's cost per minute (or revenue uplift)
cost per task (you)             = from your unit-economics model
price per task (if usage-based) ≈ somewhere between cost × target markup and a fraction of value
plan design check               = for P50 and P95 users: revenue − serving cost ≥ target gross margin

Regional considerations

Purchasing power and currency: local pricing in PKR, AED, SAR or GBP may convert better than USD-only pricing; check tax treatment (for example VAT in the UAE, KSA and UK) and invoicing requirements.
Payment methods: cards, bank transfers and local wallets vary by market; usage-based billing requires reliable payment collection.
Procurement: enterprises and governments in the Gulf often prefer predictable annual contracts; usage components may need caps.

Communicating AI pricing

Show usage in-product (credits remaining, usage by team).
Alert before limits; never surprise customers with large overages.
Explain what counts as a task or credit, with examples.
Be transparent about which features use AI and any fair-use policy.

Hands-on: margin check across user types (illustrative)

# plan_check.py: does a plan stay profitable for median and heavy users? (all numbers are placeholders)
plans = {
    "seat_addon": {"price": 20.0, "included_tasks": None},          # $/user/month, unlimited
    "credits":    {"price": 20.0, "included_tasks": 1500, "overage_per_task": 0.02},
}
cost_per_task = 0.008
users = {"P50": 300, "P90": 1800, "P99": 9000}                     # tasks per user per month

for name, p in plans.items():
    for label, tasks in users.items():
        revenue = p["price"] + (max(0, tasks - p["included_tasks"]) * p["overage_per_task"] if p["included_tasks"] else 0)
        margin = (revenue - tasks * cost_per_task) / revenue
        print(f"{name:10s} {label}: tasks {tasks:>5}, revenue ${revenue:6.2f}, gross margin {margin:6.1%}")

Run it with your real distributions. You will often find that a flat seat add-on is healthy for the median user but negative for the top 1%, which is exactly what credits, fair use or routing are for.

Worked example: an agency SaaS in the UK and Gulf

A social media management platform added AI caption and reply generation. Flat inclusion led a handful of agency accounts to generate huge volumes, pushing the feature's gross margin negative for those accounts. They moved to a hybrid: each plan includes a monthly AI allowance sized to cover most users, with credit packs for heavy users and bulk jobs routed to a cheaper model. Most customers never hit limits; heavy users paid for what they used; the feature's margin recovered (illustrative). Clear in-app usage meters kept billing complaints low.

Pitfalls

Pricing on cost alone and leaving value on the table.
Unlimited AI in flat plans without fair-use protection.
Credits so complex nobody understands them.
Outcome pricing without an agreed, auditable definition of the outcome.

How to measure success

For each pricing option you can show value per task, cost distribution, margin for P50/P90/P99 users, and results from a controlled pricing test.

AI pricing must balance value capture, cost coverage and predictability
Options: included, per-seat add-on, usage, credits, outcome-based and hybrid
Check margin for median and heavy users, not just the average
Consider local currency, tax, payment methods and procurement preferences
Show usage in-product and never surprise customers with overages

Model two pricing options for one AI feature using your cost-per-task data and user distribution; check margins for P50, P90 and P99 users and propose a test.

Enrol for free to save your progress

Reading is always free. Enrol to keep your place, take the final assessment and earn a verifiable certificate.

Pricing AI features: seats, usage, credits and outcomes