In a CRM, AI is worth paying for when it does three plain jobs: drafting replies, writing up calls and notes, and answering routine enquiries while a person keeps the last word. Lead scoring and forecasting are real features, but in our view a firm handling a few dozen enquiries a month gets more from a sorted list and a named owner. The part that is easy to miss is the bill: HubSpot and HighLevel both charge for some of their AI by usage, beyond the subscription.

Updated 12 October 2026. We first published this in October 2025. We have taken out a conversion-uplift figure we could not source, and re-checked every feature and price below against the vendors’ own pages.

What AI does in a CRM today

Strip away the marketing and the useful features come down to four jobs.

Drafting replies and follow-ups. The CRM reads the enquiry, your price list and your past answers, and writes a first reply. A person edits it and sends it. This is the one we would switch on first, because the saving is immediate and the risk is small: someone reads it before it goes. We cover how to build it in answering every enquiry the same day without a new CRM.

Writing up calls and notes. A call is recorded or transcribed and the summary lands on the contact record. The history stays complete without anyone typing it up afterwards. Tell people when a call is being recorded, and read the summary before you rely on it.

Answering routine enquiries. A chat or phone agent handles opening hours, booking a visit, “do you cover my area”. HubSpot sells this as its Customer Agent, and HighLevel sells Conversation AI and Voice AI. It is the most visible AI feature and the one with the most ways to go wrong, so it gets its own section below.

Scoring and forecasting. The CRM ranks leads by how likely they are to buy, or flags deals that have gone quiet. This learns from your past records, so it needs a history of won and lost deals. HubSpot’s lead scoring page says AI scores need a Marketing Hub Enterprise subscription, and a minimum sample of 50 contacts, 25 converted and 25 not. Rule-based scoring, where you set the points yourself, is available from its Professional tier.

The data requirement is modest. The catch is the tier: AI scoring is limited to Marketing Hub Enterprise, so you move up a tier to get it. And the costly failures in a small firm are the enquiries nobody answered, which a scoring model cannot fix.

What the AI costs

AI features are often billed separately from the CRM subscription, and the pricing has been moving. Two examples. Both vendors show prices in dollars, and these were correct at the time of writing.

HubSpot charges for its AI agents in credits, at $0.01 a credit. Its Customer Agent uses 50 credits, or $0.50, each time it resolves a conversation without a human, and HubSpot’s Customer Agent page prices voice calls separately, per minute. In April 2026 HubSpot moved the agent from $1.00 per conversation to $0.50 per resolved conversation. That announcement does not say what counts as resolved. HubSpot’s Customer Agent page defines a resolution as support provided by the agent with no handover to a human rep for 72 hours, so check that definition against how your enquiries actually go. Included credits vary by subscription, and unused ones expire at the end of the month.

HighLevel’s pricing page lists plans from $97 a month and AI Employee add-ons at $50 and $97 a month, and says usage-based charges apply to Conversation AI and Voice AI. Its AI pricing article says the $50 plan includes 1,000 Conversation AI responses and 100 Voice AI minutes a month per location, the $97 plan is unlimited subject to fair use, and phone charges are billed separately whatever the plan.

Here is an illustration, not a quote. If a HubSpot Customer Agent resolved 200 conversations in a month, that would be 10,000 credits, or $100. Whether that is cheap depends on what a resolved conversation was worth to you, and on how many it got wrong or handed over. Ask any vendor for the monthly cost at your volume, in writing, before you sign.

Anything that talks to customers needs a rule

Let the AI draft. Let a person approve anything with a price, a date or a promise in it. A chat agent that answers on its own has no such check, so the rule has to be built in: a short list of what it may answer, a handover to a person for everything else, and a regular read of what it actually said.

The risk is yours. In 2024 a British Columbia tribunal held Air Canada responsible for wrong information its website chatbot gave a customer, and rejected the argument that the chatbot was a separate entity. The tribunal said it should be obvious that the airline is responsible for all the information on its website. That was a Canadian ruling and not UK law, but the point travels: what your website says is yours, whoever or whatever wrote it.

Choosing a CRM: start with the process

The first version of this post named a CRM for each size of business. We have dropped those lead-volume thresholds because we could not source them. These are the questions we would ask instead.

  • Who will run it? Someone has to keep the pipeline stages, forms and automations in order. If that is an agency or one person who might leave, say so now.
  • How do enquiries get in? Web form, phone, email, WhatsApp? A CRM that cannot see your main channel just adds typing.
  • What do you need out of it? The two or three numbers you check every week. If the CRM cannot produce them without a spreadsheet, it is the wrong one.
  • What would leaving cost? Ask how you export contacts, notes and workflows before you commit. Our checklist for moving off GoHighLevel shows what that looks like in practice, and the same questions apply to any platform.

We have compared GoHighLevel and HubSpot in detail in GoHighLevel vs HubSpot, including where each suits.

If your enquiries still live in a spreadsheet and an inbox, you may not need a CRM yet. A tracker with an owner and a reminder can fix many missed enquiries, and our note on signs your business has outgrown spreadsheets covers when that stops being enough.

Before you switch anything on

  1. Write down who owns each enquiry. AI does not rescue an enquiry that nobody is responsible for.
  2. Clean the records. Merge duplicates, fill in missing statuses and mark what was won and what was lost. Anything the AI drafts from or learns from is only as good as this.
  3. Turn on one thing. Reply drafts or call summaries, for one person, for four weeks. Add a second feature only when the first has earned its place.
  4. Write down what it may not do. No prices, dates or promises without a person.
  5. Set a spending limit. HighLevel’s help article describes an AI usage limit you can set for each location. For any other platform, ask what the equivalent is.
  6. Measure two numbers. Time to first reply, and how many enquiries have no reply after a full working day. Compare them at four weeks against where you started.

Where to start

If you are weighing a CRM or an AI add-on and want a second opinion before you buy, the Systems and AI Review is half a day with us, on site or on a call. You get a two page plan covering the three places your admin costs you most and what to fix first, and a fixed quote for the next step. Our CRM automation page explains how we build this kind of setup, or you can book a 30 minute call first. Every piece of work is agreed in writing before it starts.