Everyone’s CRM Has AI Now. But Is It the Same? And How Much Does It Cost?

If you’ve shopped for a CRM in the last eighteen months, you already know the headline: every product has AI now. Open any pricing page and you’ll find a sparkle icon (guilty), a name that sounds like a helpful coworker (Einstein, Breeze, Freddy, Zia, Chloe), and a list of things the software will supposedly do while you sleep.

Key takeaways

• Every CRM has AI now, so the feature list is useless as a tiebreaker. All ten platforms ship the same six capabilities: lead scoring, email drafting, an AI chatbot, a call/meeting note-taker, record summarization, and next-step suggestions. The real differentiation has moved from what they ship to what it honestly costs once your team uses it.
• “Free AI” almost always means the lightweight assistant layer. The AI that does real work is metered, gated to a higher tier, or sold as an add-on — and the meters run out faster than you’d expect. Insightly’s Copilot caps at 75 queries a month; monday’s entry bucket buys about eight hours of meeting notes.
• Only two platforms include real AI on their cheapest paid plan: Close ($9) and Nutshell ($13). Everywhere else, reaching working AI means climbing to a pricier plan, buying a bolt-on, or feeding a credit meter.

 

Table of contents

• Choosing a CRM because of its AI features
• The most popular AI features in CRM
• Ok, there’s one more key AI capability: MCP server
• How much does AI cost in my CRM?
• The four ways CRMs charge for AI
• What do CRM buyers actually want (and what they don’t)
• What’s coming next, and what is the best AI CRM
• How we continue to approach AI at Nutshell

What’s harder to find is a straight answer to two questions that actually matter.

• How much of this is genuine artificial intelligence, and how much is the same rules-based automation we’ve had for a decade, wearing a new badge?
• How much AI can you actually use before you hit a paywall and have to upgrade or buy more credits?

For transparency’s sake, it’s obvious that I, Will Gordon – Sr. Marketing Director at Nutshell, have a horse in this race. This isn’t a puff piece about why Nutshell is the best AI CRM and you should pick us. It’s an honest inventory of what CRMs are shipping when it comes to AI, how the AI meter runs, and whether any of it lines up with what buyers say they want.

I reviewed ten platforms’ (Salesforce, HubSpot, Microsoft Dynamics, Zoho, Pipedrive, Freshsales, monday, Close, Insightly, and yes, our own, Nutshell) features, pricing and fine print and compared them with the actual desires of the CRM buyer.

The pattern that emerged? It’s less about features than about fine print.

Choosing a CRM because of its AI features

Before we get into the depths of this research, I want to share my own opinion on choosing a CRM because of their AI features. If AI is #1 on your list of needs, you should redo your list.

AI is a feature that should help you use the rest of your CRM more effectively and efficiently. If it’s missing other features or connections that your business needs, having better AI features or pricing should not matter. CRM is the platform, AI is a feature within the platform.

The most popular AI Features in CRM

There’s a fact that few vendors will say out loud: the AI feature lists have almost completely converged. Across all ten platforms, the same six capabilities show up again and again.

• Lead scoring
• Email drafting,
• AI website chatbot,
• Call/meeting note-taker
• Record and timeline summarization
• “Next-step” suggestions.

If your shortlist is two CRMs and you’re comparing their AI bullet points, you’re comparing two nearly identical lists.

I know that I said earlier that you shouldn’t choose your CRM based on AI features alone. However, if the CRM you’re looking at doesn’t have something that fits each of these 6 bullet points, cross them off of your list.

Ok, there’s one more key AI Capability: MCP Server

There is one additional capability worth singling out: the MCP server.

MCP, the Model Context Protocol, lets you connect your CRM directly to an AI assistant like ChatGPT, Claude, or Perplexity, then ask plain-language questions of your own data (“show me the leads we haven’t spoken to in two weeks”) without logging into the CRM at all.

As of mid-2026, Salesforce, HubSpot, Microsoft, Zoho, monday, Close, and Nutshell all offer one. Pipedrive, Freshsales, and Insightly do not: Pipedrive has only third-party connectors, and Freshworks shipped MCP for its IT service product rather than its CRM. It’s an early signal of where this is going. The CRM stops being a place you visit and becomes a source your AI assistant queries.

There are 2 things to keep in mind when looking at a CRM’s MCP server:

1. You will likely want a paid plan for your chosen LLM (ChatGPT, Claude, etc.). Some allow connections with the free plan when you use the desktop app. However, free plans can get pretty limited with token usage. Your budget + your needs = your call.
2. Is the MCP server Read-only or is it Read and Write. “Read-only” means that the MCP has full access to your data so that it can answer questions related to your data. If the MCP has the ability to write, that means you can ask your LLM to change the data in your CRM on your behalf. Read is standard and Write is being viewed with a bit more caution by some CRMs. If the CRM you’re considering has MCP with Write access, make sure to ask if you can toggle that feature on/off for certain users. The last think you want is for your Sales intern to update all the records in your database.

So the useful question isn’t “does the CRM have AI?” All of us will say “yes”. The real questions are:

• Does it have AI features that I’ll actually use? 
• How much does it cost to access the CRM’s AI features?
• What are the limits before I’ll have to spend more?

How much does AI cost in my CRM?

This is the question the pricing pages bury. Most CRMs have different pricing models that allow you to “pay for what you need” with pre-built packages. We all essentially do the same thing when it comes to AI availability and usage. Some of these costs are going to seem HIGH while others are going to seem really LOW. The patterns you’ll see are that “free AI” almost always means the lightweight assistant layer, while the AI that does real work is metered or gaited, and the meters run out faster than you’d think.

CRM	AI at noextra cost?	Lowest cost toaccess AI	How much AI you can use before upgrading / paying more
Salesforce	Partial	$0 to start, then $500 / 100k credits	Foundations includes a starter credit pool, then a standard action costs 20 credits ($0.10). Salesforce’s own “Where is my order?” example burns 40 credits ($0.20). ~200k credits is roughly 10,000 actions. No rollover.
HubSpot	Partial	$0 assistant; ~$20/seat for agents	Free Breeze Assistant; agents draw monthly credits: Starter 500 / Pro 3,000 / Enterprise 5,000. A resolved support conversation costs 50 credits ($0.50), so Starter is roughly 10 resolutions per month. No rollover.
MS Dynamics 365	Partial	$105/user; $150 to get credits	1,000 Copilot Credits per user per month, but only on the $150 Sales Premium tier; Enterprise and Professional include zero. A grounded generative answer is about 12 credits, so roughly 83 per month.
Zoho CRM	Partial	$23/user (Zia)	Predictive Zia from Professional ($23). Zia Agents run on tokens: 30M free tokens per month (Standard) then $1 per 1M, roughly 10,000 free agent actions a month before you pay.
Pipedrive	Partial	~$59/user (email AI)	No credit pool. AI email summarization is capped at 50 per user per day (10 per 30 min). Basic “assistant” insights come earlier, but the generative tools sit at the Premium tier.
Freshsales	Partial	500 free sessions (Growth+)	500 one-time complimentary AI Agent sessions (not recurring), then $49 per 100 sessions, up roughly 5x from the old price. Buying any pack permanently removes the free 500.
Monday CRM	Partial	~$12/user (incl. 1,000 credits)	Monthly credit buckets: Basic 1,000 / Standard 2,000 / Pro 3,000. An AI Notetaker costs 120 credits per meeting-hour, so Basic’s 1,000 buys roughly 8 hours of notes a month. Credits are now a mandatory paid add-on for new accounts.
Close	Yes	$9/user (incl. 500 credits)	Chloe is on every plan; AI credits per user per month run 500 / 1,000 / 1,500 / 2,000 by tier. Voice also burns separate telephony minutes. Close doesn’t publish a per-action credit cost.
Insightly	No	$49/user	Copilot is gated to the $49 Professional tier and capped at 75 queries per month (100 on Enterprise). Roughly 4 questions a workday exhausts it in under four weeks.
Nutshell	Yes	$13/user (included)	AI is on every tier from $13. AI Outcomes per user per month: 10 / 20 / 40 / 100 / 150 by tier, where 1 outcome equals 1 use (for example, one call summary). Lightweight “AI Assists” (transcription, reply suggestions, smart reports) are unlimited.

“AI at no extra cost?” reads Yes (included on the entry plan), Partial (a free assistant layer but real AI is gated or metered), or No. Prices are per user per month, annual billing. Full sourcing in the companion brief and matrix.

Salesforce (Einstein and Agentforce)

AI at no extra cost? Partial. A free Salesforce Foundations tier switches Agentforce on, but every action is metered in Flex Credits, and the paid Starter Suite doesn’t bundle the agents at all.

According to Salesforce’s own pricing page, credits sell at $500 per 100,000, a standard agent action costs 20 credits ($0.10), and the company’s sample “Where is my order?” interaction burns 40 credits ($0.20). Credits don’t roll over. It also runs three-plus parallel pricing models that can’t be mixed, a confusion signal in itself. Salesforce does offer a genuine MCP server (GA April 2026).

HubSpot (Breeze)

AI at no extra cost? Partial. Breeze Assistant, the chat copilot, is free on every plan including the free CRM and burns no credits. The agents that do real work are Professional and Enterprise only, and metered.

Per HubSpot’s product catalog, monthly credit allowances run 500 (Starter), 3,000 (Pro), and 5,000 (Enterprise), and a resolved support conversation costs 50 credits ($0.50), so a Starter plan gets roughly ten resolved conversations a month before paying more. To its credit, HubSpot’s 2026 shift to outcome-based pricing (pay only when the agent resolves a ticket or surfaces a lead) is the most buyer-aligned move any major vendor has made. HubSpot also ships an official Claude and MCP connector on all plans.

Microsoft Dynamics 365 Sales (Copilot)

AI at no extra cost? Partial. Embedded Copilot is bundled only from the $105 Sales Enterprise tier, not the cheaper Professional plan, and you only get usable agent credits one tier higher.

According to Microsoft’s Dynamics 365 Sales pricing, just 1,000 Copilot Credits per user per month are included, and only on the $150 Sales Premium SKU; Enterprise and Professional include zero. A grounded generative answer runs about 12 credits, so the allowance covers roughly 83 such responses a month. Microsoft offers an MCP server via Copilot Studio, though Claude Desktop isn’t supported and ChatGPT is “coming.” One more cost worth naming: Microsoft’s naming maze (Viva Sales, then Sales Copilot, then Copilot for Sales, then “Sales agent”) is its own kind of tax.

Zoho CRM (Zia)

AI at no extra cost? Partial. Predictive Zia is included once you reach the $23 Professional tier (full suite at $40 Enterprise); the free and Standard plans don’t get it.

The newer agentic layer is the interesting part. According to Zoho’s Zia Agents pricing, agents run on a token wallet with 30 million free tokens a month on the Standard tier, then $1 per million tokens, roughly 10,000 free agent actions a month before the meter starts. Zoho also built a proprietary large language model it runs in its own data centers, and offers an official MCP server. It’s the deepest AI suite in the set. The catch is that predictive Zia is data-hungry and needs dozens of closed deals before it stabilizes.

Pipedrive (Pipedrive AI)

AI at no extra cost? Partial. Basic “Sales Assistant” insights appear on lower tiers, but the genuinely generative tools, the AI email writer and summarizer, sit at the Premium tier (about $59).

There’s no credit pool; instead there are hard throttles. Per Pipedrive’s own help docs, AI email summarization is capped at 50 per user per day (and 10 per 30 minutes). The conversational assistant and the much-promoted AI agents remain in limited beta, and Pipedrive has no first-party MCP server, only third-party community connectors.

Freshworks Freshsales (Freddy AI)

AI at no extra cost? Partial. Every paid tier from Growth up gets a one-time pool of free AI Agent sessions; the Freddy Copilot assistant is a separate $29 per agent per month add-on.

According to Freshworks’ session FAQ, you get 500 complimentary AI Agent sessions (one-time, not recurring), after which it’s $49 per 100 sessions, a roughly 5x per-session increase from the legacy price introduced in November 2025. Buying any session pack permanently removes the free 500. Freddy’s predictive scoring and forecasting are genuine ML, and to Freshworks’ credit it doesn’t brand its ordinary workflows as “AI.” There’s no MCP server for the CRM, only for its IT service product.

monday CRM (monday AI)

AI at no extra cost? Partial, and tightening. For accounts created after monday’s May 2026 “AI Work Platform” relaunch, AI credits are a mandatory paid component bought alongside seats.

According to monday’s AI Credits documentation, monthly credit buckets run 1,000 (Basic), 2,000 (Standard), and 3,000 (Pro), an AI Notetaker costs 120 credits per meeting-hour, and an AI block costs 8, so the Basic bucket buys about eight hours of meeting notes a month. monday offers a hosted MCP server. The honest knock, per Gartner’s first Magic Quadrant placement for the platform, is “limited AI and ML depth” relative to established players, and the classic no-code “recipes” are deterministic automations frequently shown in AI framing.

Close (Chloe)

AI at no extra cost? Yes for access, then metered. Close puts Chloe on every plan including the $9 Solo tier and frames it as “you only pay for the work Chloe does.”

Per Close’s pricing page, included AI credits run 500, 1,000, 1,500, and 2,000 per user per month by tier. The asterisk: Chloe Voice, the genuinely impressive autonomous outbound caller, consumes AI credits and separate telephony minutes, and Close doesn’t publish what a single Chloe action costs in credits, which makes the real per-call cost hard to predict. Close offers an official MCP server and even publishes an llms.txt file to court AI-engine citation.

Insightly (Copilot)

AI at no extra cost? No. Copilot is absent from the $29 Plus tier entirely; you reach it at the $49 Professional plan.

According to Insightly’s pricing table, Copilot is capped at 75 queries per month on Professional and 100 on Enterprise, where one query is a single conversational prompt. A rep asking four questions a workday burns through 75 in under four weeks. Insightly’s labeling is unusually clean (it keeps automation visibly separate from AI), but there’s no free AI, no MCP server, and the tightest usage cap of any platform here, on its closest “simple CRM” peer set. 

Nutshell (Nutshell AI)

AI at no extra cost? Yes. AI is included on every tier starting at the $13 Foundation plan. No separate AI SKU, no top-tier gate.

Each of our plans includes a pool of AI Outcomes per user per month (10, 20, 40, 100, and 150 from Foundation to Enterprise), where one outcome equals one use, like a single call summary or lead score. Lightweight “AI Assists” such as note transcription, reply suggestions, and smart reports are unlimited on every plan. We also have an MCP server connecting the CRM to ChatGPT, Claude, etc. Any negatives? Our AI Outcomes pool is still a cap, just a transparent, predictable one. We do restrict some of our AI Assists on higher plans, but again, the AI assists come with unlimited usage.

The four ways CRMs charge for AI

Step back from the individual products and the pricing sorts into four architectures. Most vendors blend more than one.

• Tier-gating (included, but only on upper plans). No separate AI price, but you must climb to a pricier plan to reach it: Zoho’s predictive Zia ($40 Enterprise), Freshsales’ Freddy ($39 Pro), Insightly’s Copilot ($49 Professional).
• Add-on or bolt-on. A separate line on top of the seat. Salesforce’s Agentforce add-ons at $125 per user per month are the archetype, with editions climbing to $550 per user per month.
• Consumption, credits, or tokens. Pay per unit of work. Salesforce Flex Credits, Microsoft Copilot Credits, Zoho tokens, monday AI credits, and Close AI credits all live here. It’s the fastest-growing model.
• Outcome-based. The 2026 twist: pay only when the AI completes a task. HubSpot now charges $0.50 per resolved conversation and $1.00 per qualified lead. More aligned to value, but still usage-variable.

The takeaway:  Only two platforms in this set, Close ($9) and Nutshell ($13), include AI on their cheapest paid plan by default. Everywhere else, “AI” on the lowest tier means the assistant layer, and the AI that does real work waits behind a higher plan, an add-on, or a credit meter.

What do CRM buyers actually want (and what they don’t)

AI adoption in CRMs is real and rising. According to G2’s State of AI in CRM, 45% of sales pros now use AI weekly, most often inside the CRM, and 44% say AI makes them more likely to use the CRM at all. But enthusiasm is uneven and specific. Buyers want a few things badly and treat the rest with open skepticism.

What CRM buyers want, and why

• Call and meeting note-takers, because the value isn’t the transcript, it’s the action items landing in the CRM record automatically. It’s the single most-loved, most-used AI feature in the category. The catch that buyers name repeatedly is that it only works if it syncs natively. Anything that makes them copy-paste gets abandoned.
• Email drafting and reply starters, because they kill the blank page. Reps edit-and-send rather than send raw, so it saves time without ceding control.
• Quiet data capture and auto-logging, because the highest-value AI is the kind that means a rep doesn’t have to type. It’s unglamorous, and it’s exactly what they want.
• Lead ranking and enrichment, because prioritization saves real time when the underlying data is clean, with scoring accuracy running a solid multiple above manual guesswork.
• Querying the CRM from their AI assistant (MCP), because people increasingly live in ChatGPT and Claude, and asking “which deals are stalling?” without logging in removes a step from the day.

What CRM buyers don’t want, and why

• AI sales forecasting, because they don’t trust it. According to
  • ASLI’s 2026 analysis, forecasting is accurate at 85 to 95% only on clean, milestone-based pipelines and collapses to 50 to 60% on messy data, with only about 7% of organizations hitting 90%-plus accuracy. A confident number built on bad data is worse than no number.
• Fully autonomous agents acting on their own, because the trust isn’t there yet. According to
  • Gartner (May 2026), 69% of B2B buyers still turn to a human rep to validate AI-generated insights. The rep isn’t being replaced; the rep is becoming the fact-checker, which means an agent that acts unsupervised makes buyers nervous.
• “Me-too” feature parity, because every CRM now ships the same six AI features, and buyers have stopped believing any vendor’s AI is meaningfully different from the next.
• Confusing credit meters and top-tier gating
  • Metronome documents, credits are “useful but not loved.” Expiration rules and surprise overages erode trust, and being charged the same credit for a trivial query and a complex one feels unfair, the phenomenon the industry now calls “meter shock.” Salesforce’s Agentforce, at $125 to $550 per user per month, is the lightning rod for the gating complaint.

What they wish existed but can’t rely on yet

• Forecasting they can trust, explainable AI that shows why a deal is at risk, not a black-box score.
• Proactive next-best-actions that are genuinely smart, not templated reminders dressed up as insight.
• AI that works on day one with messy data, because the prerequisite-data problem is exactly what smaller teams can’t staff for, and it’s the clearest unmet need in the market.

Let me summarize what CRM users want in plain language

CRM users want AI that will enter data, log calls, get started on a blank page, tell you what to do next, and surface trends and pattern in their data automatically or with minimal prompting.

Managers, reps, marketers and the rest of your team want to save time on the logging and the list building and spend more time on the money making activities.

What’s coming next and what is the best AI CRM

My bet is that more and more CRM users will get over the hump of “AI will take my job” and start to realize “AI makes me better at my job.”

The big CRM players will likely continue to focus on the hype and the flair of AI, while users will continue to simply want AI features that help them do their job just as well by using the CRM less (note-taking, auto-logging, next-step suggestions, etc).

If you look at the AI industry, separate from CRM, there are no signs of slowing down. I expect that to be the same when it comes to implementing more and more AI into CRMs.

What’s the best AI CRM? There will never be a 100% answer to this question – sorry. When you’re looking at choosing your first, or your next, CRM, ask yourself these questions:

• Does it have all of the features we need?
• Is it in our budget (if not, can we make it pay for itself with how we use it)?
• Is it easy to use? Will my team use it?
• Does it have the 6-core AI features?
• Am I going to run out of AI credits? Do I have to pay top dollar for AI access?

There’s always more questions you can ask, but that’s a good start!

How we continue to approach AI at Nutshell

I promised this article wouldn’t be a sales pitch, and I meant it. AI has been a big focus at Nutshell for the past two years, and I’m not going to tell you we did it better than everyone above. That’s genuinely for you to decide. What I can do is be transparent about the choices we made and why, because they were direct reactions to the frustrations in this very article.

We made three calls.

First, we put AI on every plan, starting at our entry tier, instead of gating it behind a top-tier upgrade, because the loudest complaint in the market is that the useful AI is always one expensive plan away. Yes, there are some features we have gaited on some higher tiers, but we packed A LOT of AI value into our most affordable plan.

Second, we tried to make the usage meter legible.

• AI Assists – Lightweight everyday helpers (lead recaps, reply suggestions, smart reports) are available to all users with unlimited usage.
• AI Outcomes – Heavier features that produce a real result (call summarization, lead researcher, etc.) drawn from a clearly stated monthly pool, one outcome equals one use. We did that because not knowing what a given click will cost is corrosive to trust.

Third, we leaned into the two things buyers actually told the market they wanted: note-takers that sync natively into the record without copy-paste, and an MCP server so you can query your CRM straight from ChatGPT, Claude, etc.

Now the honest part, held to the same standard I applied to everyone else. An Outcomes pool is still a cap. Ten outcomes on our entry plan is a limit, and a heavy user will feel it. We think a transparent, predictable cap plus unlimited everyday assists beats an opaque credit meter, but you should weigh that against your own team’s usage, not take my word for it. I’d rather you make that call with clear information than with a marketing slogan.

Which is really the whole point of this piece. The AI feature lists have converged; the differentiation has moved to honesty about what is genuinely AI and honesty about what it costs once your team starts using it. So if you’re the one choosing, skip the feature-list bake-off and ask three questions of any vendor, mine included.

1. Which of these features genuinely reasons or generates, and which are automations in an AI costume? Make them answer specifically.
2. How much AI can my team actually use at our real usage level before we pay more, not on the headline plan, but in month three? If they can’t tell you, that’s your answer.
3. Does the AI work on day one with my messy data, or does it need months of cleanup and a pile of closed deals first? The “when does it actually work for me” question is a far better filter than whether a box is checked.

In 2026, every box is checked. That’s exactly why it stopped being useful, and why the meter, not the feature list, is where you should be looking.

BACK TO TOP