Outcome-Based AI Pricing: Paying per Resolution, Task or Result

You pay each time the AI completes a defined outcome instead of paying per seat: Intercom charges $0.99 per Fin outcome, Help Scout $0.75 per AI Answers resolution, and HubSpot about $0.50 in credits per resolved Customer Agent conversation. The vendor's definition of "resolved" decides your bill, so read it before comparing prices.

Silence usually counts as success. Intercom's pricing page counts an outcome when a customer "doesn't ask for more help after Fin responds", treated as an assumed resolution after 24 hours of quiet, and even when Fin completes a workflow that ends in a handoff. Zendesk closes a messaging conversation after two hours without a reply by default, or 72 hours on email, then an AI check decides whether it's a billable verified resolution; assisted escalations and unverified contained resolutions are free (since 18 May 2026). A customer who gave up and phoned instead can still be billed as resolved.

Follow me on Instagram@sagnikteaches

Four meters AI vendors charge by

Outcome pricing is one of four ways vendors meter AI in customer-facing work. Each puts the risk of a bad month in a different place.

Connect on LinkedInSagnik Bhattacharya

Outcome pricing sounds like the vendor sharing your risk, and partly it is: nobody pays for a bot that answers nothing. But the vendor writes the definition, so the size of the risk you've handed over depends entirely on its wording. Per-seat versus usage-based pricing covers the broader choice between those meters.

Subscribe on YouTube@codingliquids

How each vendor defines a resolution

Three differences drive most of the gap between bills. Intercom can count a completed Procedure that ends in a handoff; Zendesk and Help Scout exclude escalations. Zendesk adds a model-based check before billing; the others rely on the customer's behaviour. And the inactivity window varies from 2 hours to 72, which matters because a longer window gives an unhappy customer more time to come back and cancel the "resolved" status. The primary sources are worth reading in full: Intercom's pricing page, Zendesk's resolution tiers and Help Scout's pricing.

That Zendesk setting carries a useful irony: lengthening the messaging window makes your AI's resolution rate look worse and your bill more honest. If you use Zendesk, try a longer window, such as a day, for a month and compare the verified count with your own sample of transcripts. The difference is roughly what the default window was counting as resolved before customers had finished.

One month of dealership chats, billed three ways

Imagine an illustrative car dealership whose website chat handles 1,000 conversations a month. The owner exported a month of transcripts and sorted them by what actually happened:

• 450 got a correct answer (opening hours, how to book a service, whether a car is still available) and the customer said thanks or simply left satisfied.
• 200 got an answer that didn't help and went quiet. Call logs showed many of them phoned the showroom within the hour.
• 250 were handed to staff: test drives, part-exchange valuations, finance questions. Of these, 100 went through a configured booking workflow that ends in a handoff.
• 100 left before any answer.

Applying each vendor's definition to the same month:

The same month produces between about 500 and 750 billable outcomes depending on the definition, while the real number of resolved customers is 450. The 200 silent failures cost $198 on Intercom's price or $150 on Help Scout's, for customers who then took up a salesperson's time on the phone anyway. Seat costs sit on top if your staff take handoffs in the vendor's own inbox; Intercom's page says Fin used with another helpdesk has no seat or platform fees, only outcomes and the minimum.

Is it still worth it? Probably. If those 450 genuine answers would each have taken a salesperson about four minutes, that's 30 hours a month, around $750 at an illustrative $25 an hour. Even the highest bill above is roughly break-even on that sum, and fixing the silent failures moves it clearly into profit. Adding a closing question ("Did that answer your question?") and routing part-exchange questions straight to a person cut the silent group sharply in the dealership's second month. The lesson is to measure the cost per genuine resolution, not the headline price per outcome.

Per task and per result pricing beyond customer service

The same questions apply wherever an AI product charges per unit of work:

• Per lead. HubSpot's Prospecting Agent uses 100 credits, about $1, per lead it recommends for outreach. You pay per recommendation, not per meeting booked or deal won.
• Per conversation. Jobber Receptionist costs $29 for 30 conversations, then $0.79 each. A property maintenance firm with 120 conversations a month pays $29 + 90 x $0.79 = $100.10. If 48 of them become booked jobs, that's about $2.09 per booking, a number worth tracking because the invoice won't show it.
• Per message. Meta's Business Agent answers in WhatsApp, Instagram and Messenger at $2 per million tokens, roughly 4-5 cents a message by Meta's estimate. A wholesaler answering 3,000 trade-customer messages a month would pay around $120 to $150, with every message counted whether or not it helped.
• Per result from people. Some consultants offer success fees tied to savings or revenue. The definitional problem is identical; success fees explained covers how to write the measure down.

Usage meters are also creeping into seat-based products. Microsoft has told its resellers that new Microsoft 365 Copilot Business licences bought through its Cloud Solution Provider programme from 2 November 2026 come with pay-as-you-go usage billing switched on by default for extras such as Copilot Cowork, with a preset monthly limit. A seat price no longer guarantees a fixed bill.

Estimating the bill before you have any data

Before launch you don't know your resolution rate, but you can bracket the bill. Four numbers do it: monthly conversations, the share the AI will answer, the share of those the vendor's definition will bill (including silent customers), and the price per outcome.

Monthly outcome bill = conversations x share answered x share billed
                       x price per outcome   (+ seats, minimums, overage)

For an illustrative estate agency expecting 400 website and portal enquiries a month, with the AI answering between 40% and 60% and the definition billing 85% to 95% of those, the range on Help Scout's $0.75 is 400 x 0.40 x 0.85 x $0.75 = $102 at the low end and 400 x 0.60 x 0.95 x $0.75 = $171 at the high end. On Intercom's $0.99 the same range is about $135 to $226. Budget for the top of the range in the first two months, then replace the guesses with real figures from the sampling routine below.

Volume changes which meter suits you. Outcome prices scale in a straight line, so it's worth seeing the numbers at a few sizes:

At 100 a month, the fee is small but so is the time saved, and a well-written help page might do nearly as well. At 2,000, the fee is significant enough to justify negotiating, comparing flat-priced options and, above all, measuring how many of those outcomes are real. The staff-time column assumes every billed outcome was genuine, which the dealership's figures show it won't be.

Caps, pools and alerts that stop the bill running away

• Hard caps. Help Scout lets you set a monthly cap on AI Answers resolutions; when it's reached, AI Answers switches off for the rest of the cycle. Decide what customers see then, usually the normal contact form.
• Credit pools. HubSpot's credits come in monthly pools that don't roll over. Size the pool to typical months and buy extra only when needed, rather than paying for headroom you lose.
• Allowances and minimums. Zendesk draws resolutions from an allowance; Intercom's standalone Fin has a monthly minimum. Know both numbers before your quiet season.
• Preset usage limits. For Microsoft's usage billing, find out what the preset monthly limit is and who can change it.
• Your own alert. A weekly glance at billed outcomes against conversations catches a spike before the invoice does.

Spikes are the scenario caps exist for. Picture an illustrative courier firm whose parcel-tracking page failed for most of a Monday. Around 1,200 customers opened the website chat asking where their parcel was. The bot, following its instructions, pointed each one to the tracking page, which was down. Most customers gave up and went quiet. Under a definition where silence counts, a thousand quiet conversations at $0.99 each is $990 for a day on which the bot helped almost nobody. A monthly cap would have limited the damage, and a daily alert on billed outcomes would have flagged it by lunchtime. The firm now pauses the bot whenever the tracking system is down, and asked its vendor, in writing, how it treats outcomes during an incident on the customer's side.

Checking whether "resolved" meant resolved

Once a month, export 40 conversations the vendor billed as resolved and label them. An assistant speeds this up if you paste in the transcripts with the customers' names removed:

Below are 40 chat transcripts that our vendor billed as "resolved".
Label each one:
A = question clearly answered and the customer acknowledged it
B = answered, customer went quiet, answer looks complete
C = answered, customer went quiet, answer looks wrong or incomplete
D = customer asked for a person or said they would call instead
Give the transcript ID, the label and one line of reasoning.
Do not guess what happened after the chat ended.

An illustrative extract from the dealership's second month:

1183  A  Customer: "perfect, thanks"
1190  C  Asked about part-exchange value; bot replied with opening hours
1204  D  Customer: "I'll just ring the showroom"
1231  B  Service booking link given; no reply
Summary: A 17, B 11, C 8, D 4

One label needed correcting: transcript 1231 was marked B, but its timestamps showed the customer's final message, a follow-up question, arrived after the bot had stopped listening. Check timestamps on every B. The assistant also can't know whether a quiet customer phoned instead; the call log can. After corrections, 12 of 40 billed outcomes, 30%, were not real resolutions. Applied to 750 billed outcomes, that's about 225 a month, roughly $223 on Intercom's price, and a clear list of what to fix first. A weekly sampling routine for AI support replies turns this into a habit, and when a chatbot should hand over to a human helps with the fixes.

Questions to settle before you sign an outcome-priced deal

1. What exactly makes an outcome billable? Get the definition into the contract, not just the pricing page.
2. What is the inactivity window, and can you change it?
3. Are handoffs, or workflows that end in a handoff, ever billable?
4. Can you download billed outcomes with conversation IDs, and is there a dispute process?
5. What are the minimums, allowances and overage rates?
6. What caps and alerts are available, and what do customers see when a cap is hit?
7. How much notice do you get if the definition changes? Zendesk's tiers, for example, changed how resolutions were counted from 18 May 2026.
8. Which seat or platform fees apply on top?

It also helps to start small. Ask whether you can switch the AI on for a subset of traffic first, such as one page of your website or one enquiry type, so the sampling routine runs on real conversations before the whole site is billed.

With those answers you can run your own month of conversations through the definition, as the dealership did, before a single outcome is billed. How to measure whether your chatbot is working covers the other numbers worth tracking once it's live.

Paying per outcome: common questions

Is outcome-based pricing cheaper than paying per seat?

It depends on volume and on how many billed outcomes are genuine. Divide the monthly outcome bill by the conversations the AI truly resolved, then compare that cost per real resolution with the staff time those conversations would otherwise take. At low volumes a per-outcome price is usually cheaper than extra seats; at high volumes, check whether a flat plan works out lower.

Can I dispute a charge for a chat that wasn't really resolved?

Ask the vendor before you sign whether billed outcomes can be listed with conversation IDs and whether there's a dispute process. If they can be listed, sample them monthly and raise clear failures with evidence, such as a transcript where the customer said they would phone instead. Whether you get credit depends on the contract, so get the definition written into it.

What counts as a resolution with Help Scout's AI Answers?

Help Scout counts a resolution when a customer receives an AI response and doesn't escalate, search the knowledge base, ask more questions or say they need more help. It charges once per conversation, even if AI Answers replies several times, and doesn't charge for greetings or incomplete answers. You can set a monthly cap, after which AI Answers switches off until the next cycle.

Is outcome pricing spreading beyond customer service?

Partly. HubSpot's Prospecting Agent uses credits for each lead it recommends for outreach, and some consultants offer success fees tied to results. Other products meter per conversation or per token instead. In every case the same question applies: what exactly triggers the charge, and does that event mean the result you actually want, such as a booking or a sale?

Further reads

• How to Read AI Software Pricing: Seats, Credits, and Usage Fees — Read seats, credits and usage fees on any price page.
• How Much Does a Website AI Chatbot Cost, and Will It Pay Off? — Whether a website chatbot pays for itself at all.
• AI Customer Service for Small Businesses: What to Automate First — Which customer service jobs to automate first.
• HubSpot AI Agents and Breeze: What Small Teams Should Switch On — Which HubSpot agents are worth their credits.
• How Much Does a WhatsApp Chatbot Cost? Fees, Tools and Setup — The full cost of a WhatsApp chatbot, fees included.
• What to Ask an AI Chatbot Vendor Before You Sign Up — More questions for a chatbot vendor before signing.
• AI Ordering vs Delivery Apps: What a Takeaway Actually Pays — Delivery app commission against AI phone and web ordering, line by line, with a break-even worksheet a takeaway can fill in with its own numbers.
• Can an AI Chatbot Handle Order Tracking and Returns? — What an order-tracking chatbot needs to see, which return decisions it can safely make, and where a person still has to step in.
• Per-Seat or Pay-As-You-Go? Legal AI Pricing for Small Firms — When a legal AI seat beats paying by use, how to work out who in the firm needs one, three monthly budgets and the costs quotes leave out.
• How Much Does an AI Agent Cost a Small Business? — The five ways AI agents are priced, the cost-per-task sum that makes them comparable, and the supervision time that usually costs more than the software.
• ClickUp vs Asana vs Monday: Which AI Features Save Time? — All three now charge for AI differently. A timed photography-studio example shows which features save hours and what each costs a five-person team.
• Hidden Costs of AI Implementation Most Small Businesses Miss — The AI costs that never appear on a pricing page, a subscription box company's real first-year bill against the headline, and a cost sheet to copy.
• What Is the Cheapest Way to Handle Enquiries Out of Hours? — Five ways to cover evening enquiries, priced at three volumes, with a charity shop's donation questions and the billing small print on 'resolved'.
• How to Run an AI Software Demo So You See the Real Product — Send your own scenarios first, bring questions the vendor hasn't seen, and inspect the admin screens and pricing terms so an AI demo shows the real product.
• How Much Does Customer Service Software With AI Cost per Agent? — Seat prices for Help Scout, Freshdesk, Zendesk, Intercom and HubSpot, what each charges for AI, and a four-agent desk costed per agent and per conversation.
• Is HubSpot Worth Paying For Just for Its AI Features? — The seat jump, the onboarding fee and the credit meter, worked through for a small florist, so you can see when HubSpot's AI agents pay back.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Intercom pricing page (Fin outcomes and seats); Zendesk help on automated resolution tiers; Help Scout pricing page and AI Answers documentation; HubSpot pricing and knowledge base on credits and Customer Agent; Jobber Receptionist pricing; Meta Business Agent pricing; Microsoft partner announcement on Copilot Business usage billing. Checked September 2026.