AI Customer Support Pricing in 2026: Per Resolution vs Per Seat vs Flat Rate
Every AI customer support vendor bills something different. One charges per ticket the bot closed. One bundles it into your seat price. One charges only when the AI produced a verified outcome, and a fourth sells a flat monthly block with no usage meter. Two can look like the same $99 on a pricing page and land 400% apart on the invoice.
That is not a vendor trick. It is a pricing model difference, and if you do not name the unit before you sign, you are guessing which number the renewal will use.
I read the pricing pages of six major vendors in November 2026 and modelled what each costs a small SaaS team at three ticket volumes. This is a pricing comparison, not a feature comparison. The question that decides your bill: what are you paying for.
Why Pricing Model Matters More Than Price Per Ticket
Here is the trap. A vendor quotes “$0.99 per resolution” and another quotes “$29 per agent per month.” The first looks cheap, the second looks expensive, and most founders compare them directly. That comparison is meaningless because they are not selling the same thing.
The $0.99 charge is per resolved conversation, and it keeps accruing as volume grows. The $29 is per seat. It does not matter how many tickets arrive, because the AI is doing the handling.
Those numbers cross over. At low ticket volume the per-resolution model wins, because you only pay for the automated slice. At high volume the seat model wins, because the flat fee stays flat while the usage bill climbs. For most vendors the crossover lands around 1,000 to 3,000 AI-handled conversations a month.
The practical version: under roughly 500 tickets a month, per-seat is almost always cheaper. Above 3,000, model the actual resolution rate, because a vendor quoting a low headline rate will bill you more as your AI gets good. A better AI closes more conversations, and every closed conversation is a billable unit.
The Six Pricing Models You Will Run Into
Every AI support product on the market today is a variation on one of these six patterns.
Model 1: Per Seat, AI Included
You pay per human agent and the AI handles unlimited conversations on that plan. Most predictable model, because cost tracks headcount, not ticket count.
Model 2: Per Seat, AI Metered Separately
Same seat fee with AI usage as a line item on top. Looks like per-seat pricing until the invoice arrives.
Model 3: Per Resolution
You pay each time the AI closes a conversation without a human. The model most AI-first startups use, because you only pay when the AI does work.
Model 4: Per Outcome
Similar, but success is defined more narrowly: the customer got an answer and marked it solved. Friendlier in theory, harder to forecast.
Model 5: Flat Monthly with Allowance
You buy a block. Zendesk works this way. The included allowance scales with your agent count, so adding a seat also adds AI capacity.
Model 6: Credit-Based
You pre-buy credits or actions and spend down. Best unit price, worst cash flow, no refund on unused capacity.
Intercom: Per Outcome, From $0.99
Intercom now sells outcomes rather than raw resolutions, at $0.99 per Fin outcome on every plan. The platform itself is seat-based: Essential starts at $29 per seat per month on annual billing, Advanced is $85, Expert is $132.
Advanced and Expert bundle free Lite seats, 20 and 50 respectively, so a small team gets coverage without paying full seat rates.
The add-ons are where the real cost hides. Copilot runs $29 per agent per month, with 10 free Copilot conversations per agent per month included. Proactive Support Plus is $99 a month. Pro, the advanced analytics tier, starts at $100 a month.
For a founder: one person on a 500-ticket-a-month queue, Essential at $29 plus $0.99 per outcome on 300 AI-closed conversations, is roughly $326 a month. Add Proactive Support Plus and you are at $425, about what one part-time support hire costs in a low-cost market.
At 3,000 tickets a month with the same 60% resolution rate, you hit about 1,800 billable outcomes, or $1,782 plus your seat fee. The seat model wins clearly at that volume, so do not default to the outcome price when modelling forward.
Zendesk: Per Seat with Included Automated Resolutions
Zendesk is the clearest example of flat-with-allowance. In 2026 its AI agents sit inside every Support and Suite plan rather than as a separate product.
The unit is the automated resolution, abbreviated AR. Support Team and Suite Team include 5 ARs per agent per month, Suite Professional includes 10. Committed volume above the allowance runs $1.50 per resolution, pay-as-you-go runs $2.00.
The allowance scaling with seat count is worth internalising. Ten agents on Support Team gives you 50 included ARs a month automatically, so growing the team grows your free AI capacity at the same time. That is a meaningfully different model from a vendor where every closed conversation is a line item.
The catch is the per-agent base price. Support Team and Suite Team sit around $19 per agent per month for the entry tier, and Suite Professional climbs once messaging, analytics, and custom workflows are included. A five-person team is not a $95-a-month operation.
Copilot is where it gets expensive. It is included by default in a 14-day trial, then added on Professional and above. The plan includes 5 Copilot uses per agent per month, pooled across the team, capped at 500 per instance. Twenty agents means 100 pooled uses before you pay for more, and Copilot conversations power suggested replies, ticket summaries, and merge suggestions.
Also note action credits: $200 per 100,000 actions per subscription term, on top of what your plan includes. Build automations that write to external systems and you have a second meter running alongside the resolution meter.
Featurebase: Per Seat with Per-Resolution AI on Top
Featurebase layers a usage meter on top of a seat fee, which is the most common source of surprise in this category.
Seat pricing is straightforward. Free is $0 forever with 1 seat. Growth is $29 per seat per month billed yearly. Professional is $59. Enterprise is $99.
Then there is Fibi, the AI agent, at $0.49 per successful resolution, charged once per conversation and only when Fibi fully resolves it without a human handover. Human handovers are free, which is a genuinely fair structure.
There is also a separate Copilot meter: 10 free Copilot conversations per seat per month, then $19 per seat per month for unlimited usage. A five-agent team on Professional at $59 per seat is $295 a month, and unlimited Copilot across those seats adds another $95.
Model a founder handling 1,000 conversations a month at a 50% Fibi resolution rate: $295 for seats, plus 500 resolutions at $0.49 which is $245. That is $540 a month before Copilot. Intercom Essential at $29 with 500 outcomes at $0.99 comes to $524. Similar money, different products: Featurebase bundles a full helpdesk with live chat, inbox, and ticketing, while Intercom bundles a Messenger with better native AI maturity.
The $0.49 rate is the cheapest per-resolution price of any major vendor I checked. High ticket volume and a low seat count is where Featurebase has the best unit economics.
Tidio: The Flat Monthly Entry Point
Tidio sits at the accessible end of the market and it is worth including because for a lot of solo founders it is the only realistic starting point.
Tidio Lyro, their AI agent, started around $20 a month for a small volume of conversations on the entry tier, with higher tiers and more AI usage on the paid plans. The platform runs on a per-agent monthly seat fee with a free tier for small volumes.
The important thing about Tidio is that its per-conversation figure covers conversations handled, not outcomes. That is a looser unit than Fin’s outcome or Featurebase’s successful resolution, so your bill scales with AI activity rather than AI success. If your AI is performing badly, you pay the same. That is the strongest argument for per-outcome pricing, and why to model your resolution rate before signing.
Cost Modelling: The Same Company Under Five Setups
Here is where the models diverge. Model a founder with 2,000 inbound conversations a month, a 50% AI resolution rate, and three people on escalations. List prices from the vendors’ own pages, November 2026, before any startup discount.
| Setup | Platform cost | AI cost | Monthly total |
|---|---|---|---|
| Intercom Essential, 3 seats, 1,000 outcomes | $87 | $990 | $1,077 |
| Zendesk Support Team, 3 seats, 50 included ARs, 950 paid at $1.50 | $57 | $1,425 | $1,482 |
| Zendesk Suite Team + Committed ARs at $1.50, same volume | $87 | $1,425 | $1,512 |
| Featurebase Professional, 3 seats, 1,000 resolutions at $0.49 | $177 | $490 | $667 |
| Tidio + Lyro, 3 agents, 1,000 AI conversations | $99 | $50 | $149 |
Look at the spread. The same 2,000 conversations, same customers, same service level, range from $149 a month to $1,512. A ten-fold gap created entirely by how each vendor defined a billable unit.
Now the useful test. Push AI resolution to 70%, realistic once you fix your knowledge base. The AI is better, customers are happier, and the bill goes up on every per-unit vendor, because more success means more billable events.
| Setup at 1,400 outcomes | Platform cost | AI cost | Monthly total |
|---|---|---|---|
| Intercom Essential, 3 seats | $87 | $1,386 | $1,473 |
| Zendesk Support Team, 3 seats, 1,400 paid ARs at $1.50 | $57 | $2,100 | $2,157 |
| Featurebase Professional, 3 seats | $177 | $686 | $863 |
| Tidio + Lyro | $99 | $70 | $169 |
The order held but the gaps widened. If improving your AI raises your costs 40%, you picked the wrong pricing model for your trajectory.
What Each Model Is Really Optimised For
Per-outcome pricing is optimised for the early-stage founder with low volume. You pay almost nothing until the AI starts working. It is also a model that gets more expensive the more successful you become, which is a strange incentive if you are building a long-term business.
Per-seat pricing is optimised for the scaling team. Once the AI handles most of the volume, adding a human becomes a deliberate decision rather than a cost of doing business.
Flat-with-allowance is optimised for the mid-size team that knows how many ARs it needs. The risk is that the included allowance is small relative to your ticket volume, so you pay the per-unit rate for most of your traffic anyway.
Credit-based is optimised for the enterprise with predictable annual AI usage. For a founder, usually the wrong shape.
How to Model Your Own Bill Before You Sign
Do this before you talk to sales, because a vendor’s calculator is tuned in their favour.
Step 1: Get your real ticket volume. Not signups, not customers. Pull actual inbound conversations from the last 90 days across every channel. Multiply by 1.5 if you are launching something new, because launch weeks run 50% worse than baseline and you should budget for the bad month.
Step 2: Estimate your AI resolution rate honestly. If you have never run an AI support agent, assume 30%, not 60%. Vendor claims cluster around 50% to 70% deflection, and real numbers on a young knowledge base sit below that.
Step 3: Find the exact billable unit in writing. Ask sales three questions: what exact event triggers a charge, what happens when a human takes over mid-conversation, and what happens when the same customer asks the same question twice. A vendor who answers crisply has billing you can model. One who goes vague will surprise you at renewal.
Step 4: Check for a spend cap. Intercom lets you set a monthly limit and pauses upgrades when you hit it. Ask every vendor whether you can cap usage without losing data you already collected.
Step 5: Model three scenarios, not one. Low means half your expected volume at a 20% resolution rate. High means double volume at 70%. If a vendor only makes sense in your optimistic scenario, it is the wrong vendor.
Five Mistakes That Show Up on Every AI Support Bill
Confusing conversations with outcomes. A vendor charging per conversation bills you for your AI being bad. A vendor charging per outcome bills you for your AI being good. Everyone would prefer the second, but per-outcome vendors set a higher headline price, which is why per-conversation models still sell well.
Forgetting the second meter. The resolution rate is only half the bill. Featurebase charges $19 per seat per month for unlimited Copilot on top of the seat fee. Zendesk charges for action credits above the allowance. Intercom charges $29 per agent for Copilot and $99 for Proactive Support Plus. Build a row for every add-on before signing.
Ignoring the seat allowance. Zendesk bundles 5 automated resolutions per agent per month, so 12 agents means 60 included resolutions for free.
Taking the startup discount and forgetting it expires. Intercom and Zendesk both offer large discounts to qualified startups. Check when the discount ends, because your bill will not stay at $29 just because it was $29 in year one.
Buying per-seat for a chatty, low-complexity product. If 90% of your tickets are password resets and billing questions, a per-seat model charges you for a human in an inbox that barely needs one.
What I Would Actually Pick
If you are a solo founder under 500 tickets a month, start with the cheapest flat option that includes AI and accept that you are paying for capability you will not use. Do not spend an afternoon modelling a $70 decision.
If you are handling 1,000 to 5,000 tickets with a real knowledge base, per-outcome or per-resolution pricing is the right shape, and Featurebase’s $0.49 per successful resolution is the best unit price I found. Check that your resolution rate is genuinely above 40% before you commit, because below that threshold you are paying per conversation for a mediocre bot.
If you already run Zendesk, stay on Zendesk and use the included allowance. Moving to a cheaper AI vendor to save $200 a month is a bad trade when you also have to rebuild your support workflow. The included ARs that scale with your seat count are underrated.
If you are on Intercom today, do not rip it out. The $0.99 outcome price looks bad in a spreadsheet, but Intercom’s native AI is more capable than anything else I priced, and a tool that resolves hard tickets beats a cheap one that escalates everything above simple questions.
Whatever you pick, set the spend cap on day one. The failure mode is not choosing a bad vendor, it is choosing a good vendor, succeeding, and discovering that success has a per-ticket price attached.
The Bottom Line on AI Support Pricing
AI support software in 2026 is good enough that the interesting question moved from “does it work” to “how am I billed for it working.” That is good news, because pricing model is something you can negotiate, forecast, and optimise.
The unit of sale is the decision. Outcomes, resolutions, conversations, seats, actions, credits. Six vendors, six units, a ten-fold spread on identical service volume. Model your bill at expected volume and at double. If those numbers are far apart, you have found the term that decides whether your support stack is a fixed cost or a variable one, and that is the term worth fighting for at renewal.
