Every software pricing model is a quiet statement about what the vendor wants you to do less of. Per-message pricing wants you to talk to customers less. Per-seat pricing wants you to involve fewer people. Before you sign anything for an AI communication tool, it's worth asking: what behavior does this meter punish? — because whatever it punishes, your team will learn to avoid.
This post explains the model we landed on for Cura — per 1,000 monthly active customers (MAC) — and, more usefully, gives you a framework for evaluating anyone's pricing, including ours.
The two incumbent models, and how they backfire
Per message (or per "conversation," or per credit). Common because it maps neatly to the vendor's compute costs. But look at the incentive it hands you: every reply now has a price tag. The clarifying question your AI should ask before booking? Costs money. The reminder, the no-show recovery nudge, the "how did it go?" — all metered. Teams on per-message plans end up rationing exactly the behavior they bought the tool for: thorough, attentive conversation. Worse, the bill is unpredictable — a busy month, a promotion that lands, a holiday rush, and your software bill spikes because business was good.
Per seat. Familiar from every SaaS contract, and reasonable for tools where value scales with users. But an AI agent's whole premise is that the system does work, not the seats. Charging per seat means you hesitate to give the receptionist access, the part-timer stays out of the inbox, and visibility narrows to whoever's login is "worth it." For a tool that's supposed to be your shared memory of every customer, pricing that discourages access works directly against the product.
There's a subtler problem with both: neither has anything to do with the value you receive. Value, for a service business, tracks one thing — how many customer relationships you're actively maintaining. Messages and seats are costs of maintaining them, not measures of it.
What "monthly active customers" means
A monthly active customer is a customer your business actually interacted with this month — someone who messaged you, got a reply, received a reminder, was followed up with. Not your whole contact database; the database doesn't cost you anything to keep. Just the relationships that were live this month.
Cura's plans are priced per 1,000 MAC:
- Starter — $99/month per 1,000 MAC. Up to 5 users, 3 workspaces, 5 channels, the AI agent with conversation summaries, CRM, knowledge base, analytics.
- Standard — $179/month per 1,000 MAC. Up to 12 users, 10 workspaces, 12 channels, AI workflows (the follow-up playbooks), a bigger knowledge base, SLA and dedicated onboarding. Most businesses land here.
- Advanced — $399/month per 1,000 MAC. Up to 25 users, unlimited workspaces and channels, custom AI agent and custom workflows, the longest history and analytics retention.
Notice what's not metered inside that: messages. Within fair-use bounds (each tier includes AI usage at a multiple of your MAC — 10× on Starter, up to 40× on Advanced — generous enough that ordinary businesses never think about it), a conversation that takes twelve messages to get right costs the same as one that takes two. The fair-use line exists to keep the economics honest at the extremes, not to make you count replies.
Why this alignment matters in practice
The meter only moves when your business does. Your bill scales with active relationships — which scale with revenue. A slow month means fewer active customers, not a fixed bill for idle capacity. A great month means more active customers, and the software bill grew because the business did. That's the only kind of bill growth nobody resents.
Thoroughness is free. This is the big one for an AI product. We want Cura to ask the clarifying question, send the reminder, follow up after the visit, recover the no-show. Those extra touches are where the product earns its keep. A pricing model that taxed each touch would push you — and frankly, would push us when designing features — toward stingier automation. Per-MAC pricing means the incentive points the right way: make every active relationship as well-tended as possible.
Everyone can be in the system. Seats are capped per tier but not individually metered, so the whole front desk gets access without a procurement conversation per person. Shared context is the point.
How to budget for it
Two honest notes for your spreadsheet:
- Your MAC will likely grow once follow-up gets systematic. Recovered no-shows and re-engaged regulars are new active customers — that's the tool working, and it means the bill nudges up as retention improves. Per-MAC cost of a retained customer is a rounding error against their revenue, but model it.
- Add-ons exist at the edges. An extra user is $29/month, an extra workspace $49/month, and one-time custom AI setup runs $300–2,000 depending on complexity. Predictable, listed, no surprises — but put them in the plan.
There's no free tier, which is itself a pricing statement: the product's value shows up in your real inbox with your real customers, so the front door is a demo on your busiest channel rather than an empty sandbox.
Five questions for any AI vendor's pricing page
Whatever you end up buying, run the pricing through these:
- What's the unit, and does it track my value or the vendor's costs?
Messages and tokens track their costs. Active customers, bookings, locations track your value.
- What happens in my best month?
If the answer is "your bill spikes unpredictably," the model taxes success.
- What behavior does the meter discourage?
Have the team imagine working under it. If anyone would think "don't send that, it costs," walk away.
- Where are the cliffs?
Caps, overage rates, fair-use multiples — ask for the numbers, not the adjectives.
- Can you predict next quarter's bill within 10%?
If the vendor can't show you how, the pricing is a variable you don't control.
Pricing is product. The meter a vendor chooses tells you what they optimized for — make sure it's the same thing you are.