The number on a pricing page is the least useful thing about it. What matters is how the bill behaves when your business has a busy month, because that is exactly when you cannot afford a surprise.
A roofer after a storm, a tax adviser in January, a garden centre on the first warm weekend of spring: they all get far more enquiries than usual in a short burst. Under some pricing models, that surge makes the bill jump. Under others it does not move at all.
So instead of comparing headline rates, this piece compares the main ai employee pricing models by what they do to your bill across a quiet month and a busy one.
The four models you will come across
- Flat monthly fee. One price per month, usually with an allowance of messages or conversations and a set number of assistants. The bill is the same whether you had a slow month or a record one, up to the allowance.
- Per conversation. You pay for each conversation started. Whether a conversation that is just "hi" and nothing else counts depends on the vendor.
- Per resolution. You pay for each conversation the vendor judges was resolved without a human. It sounds like paying only for results, which is the appeal.
- Per seat, with usage on top. A charge per team member who uses the inbox, sometimes combined with one of the above. More common in larger support tools.
Hybrids exist too, such as a base fee that includes a number of resolutions and then charges for each one beyond it. They behave like usage pricing once you are past the included amount.
A worked example across three months
The rates below are invented for illustration, not taken from any vendor. Replace them with the real quotes you are comparing.
Say a small heating firm has a website assistant that handles:
- a quiet month (July): 250 conversations, of which 150 are counted as resolved
- a normal month (October): 500 conversations, 300 resolved
- a busy month (the first cold snap in December): 1,200 conversations, 750 resolved
And suppose the options on the table are a flat plan at $59 a month, a per-conversation rate of $0.30, and a per-resolution rate of $0.90.
| Month | Flat plan | Per conversation at $0.30 | Per resolution at $0.90 |
|---|---|---|---|
| Quiet (July) | $59 | $75 | $135 |
| Normal (October) | $59 | $150 | $270 |
| Busy (December) | $59 | $360 | $675 |
| Busiest vs quietest | Same | 4.8 times higher | 5 times higher |
With these example rates the flat plan wins every month, but that is not the lesson, because real rates vary and a very quiet business might see a different result. The lesson is the bottom row. Usage-based bills scale with your busiest period, which is the month you are also paying for extra materials, overtime and fuel.
It is also worth finding the crossover point, because it tells you when usage pricing genuinely is the cheaper choice. Divide the flat fee by the usage rate. At these example figures, $59 divided by $0.90 is roughly 65 resolutions a month. A business that reliably stays below that, every month including the busy ones, would pay less on the per-resolution plan. A business that goes above it even for a few weeks a year probably would not, once you add those peak months together.
Most small firms that bother to install a website assistant do so because enquiries are already arriving in reasonable numbers, which usually puts them well past any crossover point in at least part of the year. A sole trader with a handful of website visitors a week is the exception, and for them a free plan is often the sensible place to start.
Rerun the table with your own monthly enquiry numbers from last year. The chatbot ROI calculator helps if you want to set cost against the value of enquiries captured.
What resolved really means
Per-resolution pricing deserves extra scrutiny, because the unit being charged for is defined by the vendor.
Questions worth asking in writing:
- If a visitor asks one question, gets an answer and leaves, is that a resolution? Many count it as one, even if the visitor left unhappy.
- If the assistant captures a lead and your team then handles the job, is that a resolution or a handover?
- If the visitor rates the conversation badly, are you still charged?
- Can you see, conversation by conversation, which were billed?
None of these makes per-resolution pricing wrong. But a business cannot budget against a unit it cannot audit. There is also a subtle incentive issue: a model that charges for resolutions gains nothing from your assistant handing over quickly to a person, even when that is the right thing for the customer.
The honest limits of flat pricing
Flat pricing has its own questions, and it would be odd to skip them.
The allowance matters. A plan with a message allowance is only predictable if your busy month fits inside it, so check your likely peak against the limit and ask any vendor exactly what happens when you reach it. Message counts and conversation counts are also different units; a long conversation uses several messages.
You may also pay for headroom in quiet months that you do not use. For a business with steady, predictable volume that is a small price for never having to think about the bill.
For reference, SpideyChat uses flat monthly plans in USD rather than charging per resolution or per conversation: Free at $0 with 100 messages a month, Starter at $20 a month with 2,000 messages, and Pro at $59 a month with 9,000 messages, with lower prices when billed annually. Full details are on the pricing page.
Before you compare any quotes
Pull last year's enquiry numbers month by month from your contact form, inbox or analytics, and circle the busiest month. Then, for each vendor you are considering, write down what the bill would have been in that month, using their definitions rather than their marketing. Add any setup fees, seat charges or add-ons you would actually need.
If a vendor cannot tell you how to work out that number, treat that as your answer. For a longer list of charges that tend to appear after signing, read the hidden costs of AI chatbot pricing, and for how usage pricing specifically plays out in seasonal trades, per-conversation pricing in busy months.