Close CRM AI: A Published Allowance, and No Spending Cap

Close CRM is the only product in this category that tells you how much AI you get as a number. It is also the only one with no limit on how much you can spend.

Both of those are worth understanding before you switch anything on.

OUR TAKE: The AI here is real, quantified and genuinely useful for an outbound team, and the commercial design around it has two holes. The included allowance is advertised per user and stops scaling at ten users, a ceiling that appears only in the support documentation. And there is no spending cap of any kind, on a feature that places telephone calls. Use it. Put an alarm on it.

## What you get, as a number

Every tier includes a monthly allowance per user: 500 credits on Solo, 1,000 on Essentials, 1,500 on Growth and 2,000 on Scale^[1].

That is unusual. Most vendors in this category either bundle AI without saying how much, or sell it as a per-seat add-on at a flat rate. Publishing a credit figure lets a buyer reason about volume before committing, which is the whole point of a published number.

Credits pool at account level. They reset monthly. They do not carry over^[3].

## The ceiling the pricing page does not mention

The per-user scaling stops at ten users. Additional users beyond that cap add no further included credits^[3].

Tier	Advertised per user	Actual account ceiling
Solo	500	500, one user
Essentials	1,000	10,000
Growth	1,500	15,000
Scale	2,000	20,000

A thirty-seat account on Growth receives 15,000 credits a month, not the 45,000 the pricing page implies. At fifty seats it still receives 15,000.

The pricing page states the per-user figure with no qualification beside it, and the tooltip repeats it. The ceiling exists only in a support article. For a company above ten seats this is the single most expensive discrepancy we have found on any vendor's pages in this category.

No cap, on a feature that makes calls

There is no global credit limit and no per-agent limit^[3].

Voice agents place and hold calls on your behalf. The vendor states that those calls accrue both AI credits and telephony charges for their full connected duration, including periods of dead air^[3].

So the meter runs twice. Nothing stops it. A misconfigured agent, a loop, or simply more adoption than you expected produces a bill you discover at the end of the month.

That is not an argument against the feature. It is an argument for an alarm on the account before the first agent goes live, and for reading the usage dashboard weekly rather than monthly in the first quarter.

## What extra credits cost

Additional credits are sold as monthly subscriptions in seventeen bands, from twenty dollars for 1,200 credits to ten thousand dollars for a million^[3].

The price per credit falls from about 1.7 cents at the smallest band to one cent at around the tenth, and then stays flat all the way to the largest. In other words, volume buying helps up to a point and then stops helping entirely.

None of that appears on the pricing page, which says only that additional credits are available for purchase.

One asymmetry worth knowing: a plan upgrade is prorated immediately, and a credit subscription upgrade is not. Moving up a credit band charges the full price of the new band with no proration^[2].

What the AI actually does here

Three things, and they follow from what the product is.

Inside workflows, from the third tier upward, it makes automation conditional on judgement rather than only on fields^[1].

In voice agents, it places calls. This is the capability that distinguishes this product and the one that consumes most of the credits.

And across the account, it summarises and drafts, which is the ordinary assistive layer every vendor now has.

The first and third are unremarkable elsewhere. The second is not. It is why the credit question matters here more than at competitors, where the AI drafts text and stops.

## How it compares on price

Four vendors, four ways of pricing AI, and only two of them can be budgeted.

Salesforce sells its sales agent product from $125 per user per month on top of a seat, so a $195 seat becomes $320^[5]. Expensive and completely legible.

Close publishes an allowance and a price per credit, which is legible until you pass ten users, at which point the published allowance stops being the one you receive.

HubSpot meters through credits without publishing a consumption rate, so the cost arrives monthly.

Pipedrive is the least legible of the four: its meeting intelligence feature is ticked as included on all four plans while the same page's data prices it at $25 per seat per month, and no allowance is published for anything.

Ranked by how easily a buyer can budget: Salesforce, then Close, then HubSpot, then Pipedrive. Ranked by cost, that order roughly reverses.

A note on what a credit is

Neither we nor the vendor can tell you what one credit does, and that is worth stating plainly rather than glossing.

The published allowance is a number of credits. What is not published is a rate of consumption: how many credits a summarised call costs, or a drafted email, or a minute of a voice agent on the telephone.

So the figure is legible as a quantity and opaque as a budget. Two accounts on the same tier with the same seat count can consume wildly different amounts, and nothing on the pricing page helps you predict which one you will be.

This is why the trial measurement below matters more here than at vendors where AI is a flat per-seat charge. You are not testing whether the feature works. You are establishing your own consumption rate, because the vendor has not published one.

## What this means for a team above ten people

The ceiling turns the AI from a per-seat feature into a shared pool, and nobody is told.

On Growth, a team of ten has 1,500 credits each. A team of twenty has 750 each. A team of forty has 375 each, for the same per-seat price^[3].

The feature does not stop working. It gets quietly thinner as you hire, until someone notices the account is buying top-up bands every month.

For a growing company, that is the line to put in the model: included credits are a fixed number per account, not a number per person, and your effective allowance halves every time you double the team.

How to evaluate it in the trial

Fourteen days, no card^[1]. Enough to answer the questions that matter.

Run voice agents on a real list rather than a test list, for two days, and read the credit consumption afterwards. Multiply by ten working days and you have a monthly figure the pricing page will not give you.

Separately, watch the telephony meter during the same test. The two bills move together and only one of them is in credits.

Then check your seat count against ten. If you are above it, you already know your included allowance is lower than the page implies, and the question becomes which credit band to buy rather than whether to buy one.

Where the voice agent fits, and where it does not

The feature that makes this AI different from everyone else's deserves its own assessment, because it is also the one that spends the money.

It works for the top of a funnel. Qualifying inbound, confirming appointments, calling a list to establish whether a number is live and a person is interested. These are jobs with a short script, a clear success condition and a high tolerance for failure.

It does not work for a conversation that carries the sale. Anything requiring judgement about price, scope or objection handling is a human call, and attempting it with an agent produces an expensive recording of a person hanging up.

The commercial consequence follows from that. Put the agent on volume work where the cost per successful outcome is calculable, and keep it off the work where a bad call costs you the deal.

And measure it against the alternative honestly. An agent that qualifies at half the rate of a junior person but at a tenth of the cost is a good trade; one that qualifies at a tenth of the rate is a bill, and the only way to tell which you have is to run it on a real list for two days.

The questions to put in writing

Four, and they are short, because the pricing page answers none of them.

What is the included credit ceiling on our plan at our seat count? The answer should be a single number for the account, not a number per user.

What is the consumption rate for the operations we will use most, in credits? Ask specifically about a summarised call and a minute of a voice agent.

Can a spending limit be set on the account, and if not, what alerting exists? The documentation says there is no cap; ask what warning you get.

And is there any proration if we move up a credit band mid-month? The documentation says there is not^[2], so get the answer in writing rather than the assumption.

A vendor that answers all four has told you more than the pricing page does. One that answers none has also told you something.

Two sentences on the rest of the AI market

Every CRM vendor now ships an assistant, and the differences between them at the drafting layer are small.

What is not small is how they charge: a flat per-seat addition at one, a credit pool at another, an unpriced tick in a comparison table at a third. The capability converges; the commercial design does not, and that is where a buyer's attention belongs.

Close sits at the useful end of that spectrum in one respect and the uncomfortable end in another. It publishes a quantity, which almost nobody does. It publishes no consumption rate and no cap, which almost everybody does.

What AI does not fix

It does not make a bad list better. A voice agent calling an unqualified list produces the same outcome faster and bills you for the privilege.

It does not change the plan limits. Custom fields stay at 250 on every tier and connected email accounts at ten^[1]. The AI drafts; the ceilings still bind.

And it does not address the thing this product's unhappiest reviewers describe, which is a redesigned interface they found slower and support replies they found templated. No feature on this page changes either.

## For a budget

Model the included allowance at the account ceiling rather than the per-user figure, which means 10,000 to 20,000 credits depending on tier and not one credit more for seats above ten.

Then add a credit subscription band for the difference between that and your measured consumption, at roughly one cent a credit.

Then add telephony separately, because voice agents bill both.

Do that and the AI here is a planned line rather than a surprise. Skip it and you have bought the one feature in this category with two meters and no cap.