The AI workflows that pay off take over a recurring task based on a clear rule, with a person reviewing the result or using it directly. In sales, that mainly means preparation and handoff. In service, it means sorting and answering requests.
The AI features in HubSpot are quick to turn on. They create value where they work inside a process your team runs every day. The eight workflows below, four for sales and four for service, can be set up in HubSpot today. For each one, you see what triggers it, what the AI takes over, where a person approves, and how you measure the benefit.
An AI workflow is a HubSpot workflow in which AI handles one step: summarizing, classifying, or filling in a property. Triggers, conditions, and follow-up actions stay rule-based. The AI delivers a result that the workflow then works with.
In practice, you will come across three levels:
The lines between these levels are fluid. One question helps with planning: How much decision-making do you hand over to the AI? You can read more about agents on our page Breeze & Agents.
In sales, AI workflows pay off where reps spend time searching for and pulling together information. The AI prepares, and the decision stays with a person.
As soon as marketing hands a contact over to sales, the workflow action "Summarize record" (beta) summarizes activities, notes, and the associated company. The summary goes to the responsible rep, for example as an internal notification.
Approval: The rep reads the summary and decides on the next step.
Metric: Time to first contact after the handoff.
When a first meeting is coming up, the same action creates a summary of the company and writes it to a custom property on the record. Your rep reads one paragraph and walks into the meeting prepared.
Approval: The rep reviews the summary before the meeting.
Metric: Preparation time per meeting, based on team feedback.
Many requests come in as free text. The action "Data Agent: Research" (beta) analyzes that field and returns a value, for example whether a concrete project is behind the request. HubSpot uses exactly this case as an example. The workflow then branches: concrete requests go straight to sales, general ones into a nurturing sequence.
From our work: For JUMAG, a manufacturer of steam generators, a custom AI agent in the company's HubSpot portal reviews incoming form submissions before they reach sales. It checks the email domain, company website, phone number, location, and browsing behavior, evaluates the technical project details, and researches the company on the web. It pays particular attention to spoofed company domains and typical language patterns of procurement fraud. Each reviewed submission gets a rating ("Unremarkable", "Review", or "High suspicion") with findings, sources, and a recommendation for the inside sales team. The team makes the decision. The agent never blocks or deletes anything and never contacts the person who submitted the form.
Approval: In the first weeks, sales checks samples and reports misclassifications.
Metric: Share of submissions that sales confirms as a fit.
The same action reads a contact's five most recent call transcripts and fills a property from them, such as visible buying interest. Insights from conversations end up in the CRM and flow into reporting. This requires transcripts attached to the record.
Approval: The responsible rep checks the value before every forecast round.
Metric: Share of open deals with an up-to-date buying interest value.
In service, AI pays off where many similar requests come in and the answers are already documented. The better your knowledge base is maintained, the more the AI can take over.
In the help desk, HubSpot gives every ticket a descriptive name and creates a summary. With Service Hub Enterprise, HubSpot also detects a ticket's category based on the first message. A workflow uses the category to route the ticket to the right team or priority.
Approval: The team corrects misclassified tickets, and you then refine the rules in the workflow.
Metric: Time to first response.
For questions your content can answer, HubSpot suggests a reply in the ticket. It draws on the content you connect to the Customer Agent: knowledge base, website pages, blog, and other sources. Strictly speaking, this is a help desk feature, but it belongs in the service workflow.
Approval: The team uses the suggestion, adjusts it, or discards it.
Metric: Handling time per ticket.
The Customer Agent answers standard questions via chat, email, or WhatsApp and uses the same content. For sensitive cases, you define handoff rules, for example for cancellations, refunds, or login problems. Anyone who wants to talk to a person also reaches your team. HubSpot recommends starting with one channel or a small share of conversations and increasing it step by step.
The quality of the answers depends directly on your knowledge base. That is why maintaining your content comes before the launch.
Approval: Handoff rules for all sensitive topics, plus regular review of conversations.
Metric: Share of requests the agent resolves on its own.
When a ticket moves to second-level support, the AI summary from the help desk is already on the ticket. Whoever takes over knows the status and can go straight to the solution with the customer. If the case goes to sales, "Summarize record" in a workflow also summarizes the history of the associated contact.
Approval: The person taking over reviews the summary when first opening the ticket.
Metric: Time to resolution for escalated tickets.
An AI workflow pays off when four conditions come together:
If one of these conditions is missing, the workflow quickly costs more attention than it saves. Then it pays to work on the process or the data first. For your data foundation, our HubSpot Data Quality Audit is a good fit.
As of October 2026
HubSpot currently sells two pricing models. Which details apply to you depends on when you signed up for your account.
AI actions in workflows require a Professional or Enterprise edition, for example of Sales Hub, Service Hub, Marketing Hub, Data Hub, or Smart CRM. Three more points apply:
In the new pricing model with flexible seats and credits, you choose one edition for the whole account. You license users through GTM seats for marketing, sales, and service and through Ops seats for automation, reporting, and governance. Workflows are available from Professional. Here, every workflow action uses credits: 3 credits per action for the first 10,000 actions per month, fewer after that. An AI action in a workflow costs 10 credits. Which seat HubSpot requires for reply suggestions in this model is not yet documented.
HubSpot frequently adjusts the scope and editions of its AI features. Before you start, check the current HubSpot documentation on AI in workflows.
Start with one workflow for one team, with human approval and a metric you define upfront. Once the workflow proves itself in a trial run, you roll it out to more teams.
In our LOTSE model with the stages Landscape, Orientation, Trial, Scale, and Earnings, AI automation starts in stage T, the Trial: one team, clear guardrails, human approval. Landscape and Orientation clarify beforehand which workflows come first for you. Our roadmap Implementing AI in Marketing and Sales describes how to plan the rollout as a whole.
You can implement AI workflows with your HubSpot admin or with a HubSpot partner. As a HubSpot Platinum Partner, we assess the right workflows with you, build them with AI steps, and support the trial run and scale-up. Our page AI Process Automation describes how this works.