Zoho CRM AI playbook: agents at $23, and no published ceiling anywhere

Zoho puts AI agents and predictive intelligence on its Professional edition, at $23 per user per month on annual billing^[1]. In this category that is remarkably early, and it is the most interesting thing about the product's AI story.

The second most interesting thing is what the vendor does not publish, and it is a bigger gap than the price advantage.

OUR TAKE: The AI arrives at a tier most teams were going to buy anyway, which makes it close to free in practice. But there is no published allowance, no consumption rate and no stated cap, so you cannot size it before buying and you cannot predict what happens when you use it heavily. Ask for the ceiling in writing, then start with one task and measure.

## What arrives where

Edition	Annual rate	AI named on the card
Free	$0 for 3 users	none
Standard	$14	none named
Professional	$23	AI agents that execute tasks; predictive intelligence for conversion, closing and churn likelihood
Enterprise	$40	extended Zia capabilities: insights, predictions and recommendations
Ultimate	$52	the vendor's FAQ describes the highest feature limits

Two observations from that table rather than from marketing copy.

The agents arrive with process management, inventory management, CPQ and portals, in the same step. So a team buying Professional for its automation gets the AI as part of the same purchase, rather than as a line item bolted on top.

And the step that unlocks it is $9 per user per month above Standard. That is the cheapest path to agentic features we have measured in this category.

## The number that is missing

No published allowance. No consumption rate. No stated cap.

That is unusual now, and it is worth being precise about why it matters. Metered AI has a ceiling whether or not the vendor names one. If the ceiling is generous, publishing it is a selling point. If it is tight, publishing it is a disclosure. Silence tells you nothing either way, which is the problem.

Compare how the other two CRMs we have read in full handle the same question. HubSpot publishes credit allowances by tier, 500 at Starter and 3,000 at Professional, and names which agents exist at which tier^[4]. Salesforce publishes Flex Credit allowances of 500 thousand, 1 million and 2.75 million per org per year^[3].

Both of those let a buyer size the purchase. Zoho's pricing is better and its disclosure is worse, and for a team planning to lean on the AI, the second thing may cost more than the first saves.

The four questions to put in writing

Ask these before you commit a team, and ask for answers by email rather than on a call.

Is there a cap on agent runs or predictions at our edition, expressed as a number. What happens when we reach it: throttling, an upcharge, or a hard stop. Which models run the agents, and can that change without notice. And is our CRM data used to train anything, with what controls and what retention.

None of those four appeared on any page we read. All four are ordinary procurement questions, and a vendor that answers them in writing has given you something worth more than a feature list.

## What the AI is actually for here

The capability names on the cards are specific enough to plan against, which is more than most AI marketing offers.

Predictive intelligence scores the likelihood of a lead converting, a deal closing or a customer churning. That is three separate jobs and they suit different teams. Conversion scoring is for a team with more leads than capacity. Close prediction is for a manager who needs a forecast they can defend. Churn prediction is for a business with renewals, and it is the one most teams underuse.

AI agents are described as executing tasks on your behalf for a business use case, which is broad. Treat the breadth as an invitation to be specific: name the task, then ask whether an agent can do it, rather than buying the category and looking for a use afterwards.

Extended Zia at Enterprise adds insights, predictions and recommendations, which is the layer that reads your data and tells you something rather than acting on it.

What the free and Standard editions give you instead

Neither names an AI feature on its card, which is worth knowing before a sales conversation implies otherwise.

What they do give you is the data the AI would run on. The free edition includes workflow automation that routes leads and updates fields when a deal closes, plus standard reports, for three users. The Standard edition adds email and mass email, built-in calling, multiple pipelines and sales forecasting at $14 per user per month.

That matters for sequencing. Predictions are only as good as the history behind them, so a team that spends three months getting its pipeline data clean on Standard and then moves to Professional will get more from the AI than a team that buys Professional on day one with an empty system.

Put differently: the $9 step is available whenever you want it, and there is a decent argument for taking it in month four rather than month one.

## A ninety-day plan

Start with one job, not one agent.

Pick the task your team does most often that is mostly reading and summarising, because that is where metered AI pays back fastest and where a poor output costs the least. For most sales teams that is turning call and email history into something a manager can read.

Measure before you expand. Even without a published allowance you can measure your own consumption indirectly: how many agent runs per week your team actually triggers, and whether anything changes in the product's behaviour as that number grows. If the vendor will not state a ceiling, your usage log is the only early warning you will get.

Then expand one job at a time, and keep a written note of which agent does what. In a product this configurable, the failure mode is not a bad agent. It is six agents nobody can account for.

Where the AI should not be your reason to buy

Three honest limits.

We read pages, not the product. We have no basis for judging output quality, and we are not going to invent one. Where this article describes the AI, it describes what the vendor says it does.

The review record does not help here either. The product-level pool we could read is Capterra's, at 4.3 across 7,002 reviews, and its summaries speak about customisation and support rather than about AI quality^[5].

And the undisclosed ceiling means any plan that depends on heavy usage is a plan built on an assumption. If AI volume is central to your business case, get the number first. If the vendor will not give it, treat the AI as a bonus on an already cheap CRM rather than as the reason to buy.

## How the three CRMs package AI, side by side

	Zoho CRM	HubSpot Sales Hub	Salesforce
Where AI starts	Professional, $23	Starter, $7, with named agents from Professional, $90	Core, $195, with the agent layer from $125 extra
Published allowance	none	500 credits at Starter, 3,000 at Professional	500 thousand to 2.75 million Flex Credits per org per year
Agent layer included at	Professional	Professional	Max, $550

All rates per user or per seat per month on annual billing^[1][4][3].

Read across the second row. Two vendors tell you how much AI you may use. One does not. Read across the first row and the order reverses: the cheapest access to agentic features by a wide margin is Zoho's.

A buyer who wants AI cheaply should look at the first row. A buyer who wants to plan a workload should look at the second.

Governance questions, for the procurement review

If your company has an AI review process, six items belong in the questionnaire, because none of them appeared on any page we read.

Model provenance: which models run the agents, and whether that can change without notice.

Training: whether CRM data is used to train anything, with what controls.

Retention: how long prompts, outputs and intermediate data persist, and where.

Opt-out: whether specific modules, fields or records can be excluded, and who can set that.

Region: where inference happens, which matters for data-residency obligations.

Audit: what log exists of actions an agent takes on records, since an agent that writes to a record is making a change somebody will eventually need to reconstruct.

We found no readable vendor page covering certifications or data residency for this product either, so those documents should be requested directly rather than inferred from a feature card.

Where predictive scoring earns its keep

The three predictions on the card suit three different team shapes, and matching them matters more than the model behind them.

Conversion likelihood is for a team with more leads than capacity. If every lead gets worked anyway, scoring them changes nothing; if half get ignored, scoring decides which half.

Close likelihood is for a manager who has to defend a forecast. Its value is not accuracy in the abstract, it is a second opinion that is not the rep's optimism.

Churn likelihood is for any business with renewals, and it is the one most teams never switch on. A renewal saved is worth more than a deal won in most subscription businesses, and the data needed to predict it is already sitting in the CRM.

Pick one of the three to start with. Running all three on day one produces three new dashboards and no behaviour change.

If the vendor will not state a ceiling

Three practical responses, in order.

Ask again, in writing, naming the number you want: agent runs per user per month, or predictions per record per month, whichever matches how you will use it. Vague questions get vague answers.

Failing that, write your own ceiling into the plan. Decide what volume you need, measure your actual usage for a month, and treat any unexplained change in behaviour as the signal you were not given.

And keep the AI off the critical path until you have the number. Use it for work that improves a process rather than work a process depends on. That way an undocumented limit is an inconvenience rather than an outage.

## The comparison that actually matters

For most teams the question is not whether Zoho's AI is better than a competitor's. It is whether the AI changes the edition you should buy.

Here it mostly does not, and that is the point. Professional at $23 per user per month is the tier a growing team buys for automation, reporting and portals. The AI comes with it.

Compare that with the alternative structures. Salesforce sells its agent layer separately from $125 per user per month below its top edition. HubSpot gates named agents by tier with a published credit allowance. In both cases the AI is a reason to move up a rung, and the rung costs money.

So Zoho's AI story is less about capability than about packaging, and the packaging is genuinely favourable. Just do not plan a workload around a ceiling nobody has told you.