SERVICE

Breeze agents on the work that is suited for automation.

HubSpot has been an agentic customer platform since 2026. For service this means: ticket deflection via AI chat in your customer portal, FAQ automation on your knowledge base, ticket classification and sentiment detection on incoming requests. We do not build experiments, we deploy Breeze agents where they make a measurable difference. The human stays ultimately responsible, the agent handles the repetitive work.

Book an AI-for-service call Take the Maturity Scan first

What you deliver

01
AI chat in customer portal Grounded on your knowledge base
02
Ticket classification and routing Via Breeze Ticket Triage
03
FAQ automation On repeat questions, shorter response time

The observation

Everyone talks about AI, few service teams have anything that actually runs.

Service teams are flooded with demos and hype decks. A chatbot that occasionally says something unexpected, a classifier that is 60 per cent accurate, a sentiment tool nobody knows how to use. We do it differently: small, grounded agents on work that demonstrably lends itself to customer support automation. Knowledge base in order first, then agents on top, then a measurement loop on deflection and CSAT. A grounded data layer, often built on a clean CRM architecture, is the foundation for that. Not an experiment, a working model.

Knowledge base not AI-ready

Articles are written ad hoc, duplicated, or so outdated that an AI can no longer use them. A Breeze agent on a messy knowledge base produces answers your service team has to correct themselves.

Ticket classification requires manual effort

Every new ticket is manually categorised by a team lead. Error-prone and expensive. AI classification via Breeze Ticket Triage gets 90 per cent right, provided it runs on a clean category set.

Repeat questions 40% of volume

How do I reset my password, how do I download an invoice, how does feature X work. These questions are often resolved with FAQ automation or AI chat in the portal without a service rep ever seeing them.

Missing the sentiment tipping point

A customer grows frustrated across three tickets, the service team only notices at the fourth. Sentiment detection on the tone of incoming messages triggers an early escalation and a personal conversation.

Approach

From hype demo to working Breeze agents.

Six to ten weeks to roll out the first agents to production. Knowledge base clean-up first, then agent configuration, then a pilot with a sub-team before we go to the full service team. Then the monthly cycle to refine, so agents and humans keep steering together towards a healthy customer success loop.

01 Phase 01

Knowledge base audit

Which articles are used, which are dead. Top-20 questions versus top-20 articles, what overlaps, what is missing. For AI grounding it must be clean, otherwise agents give answers your service rep has to correct.

Wat je krijgt Knowledge base audit with clean-up plan and grounding set

02 Phase 02

Choosing the agent set

Which Breeze agents are valuable in your context: Customer Agent for AI chat in the portal, Knowledge Agent for FAQ automation, Ticket Triage for classification and routing, Sentiment for escalation. We do not choose everything, only what fits the volume and the business case.

Wat je krijgt Agent set with business case and expected deflection per agent

03 Phase 03

Build and grounding

Configure agents in HubSpot Breeze. Grounding on knowledge base, ticket history and product context. Escalation path on uncertainty or sentiment shift, ticket goes directly to a human. Pilot with a sub-team, two weeks running before we go broader.

Wat je krijgt Working agent set, escalation workflows and pilot evaluation

04 Phase 04

Rollout and measurement loop

Full rollout to service team and customers. Training in working methodology: agent as colleague, human as ultimately responsible. Deflection dashboard, CSAT comparison AI versus human, escalation quality. Monthly cycle for prompt tuning.

Wat je krijgt Live agents, trained team, deflection dashboard and monthly owner

Who builds it with you

Two hands on the AI rollout.

You sit with Carsten and Carel from the first session. Carsten handles the Breeze configuration, the grounding and the measurement loop in HubSpot. Carel oversees the business case, the change approach and the accountability that stays with the human.

Carsten Huiskamp

Consultant & online marketer

Configures the Breeze agents in HubSpot, does the grounding on knowledge base and ticket history, builds the escalation workflows and the deflection dashboard. Prefers to work directly in the tooling with a service lead alongside.
Carel Schrier

RevOps Lead / Strategist

Oversees the business case per agent, leads the change approach for the service team and ensures that ultimate responsibility stays with the human. Writes the measurement loop and remains involved until the deflection figures are stable.

Case · B2B mid-market

Deflection rate from 0 to 22% with Customer Agent on a grounded knowledge base

A B2B SaaS team was handling high volumes of password resets, invoice questions and feature explanations through the service team. Tickets were piling up, CSAT was declining, and the team was not growing at the pace of the customer base.

In eight weeks we set up Customer Agent in the portal with grounding on the knowledge base, activated ticket triage and sentiment detection for escalation. Deflection rate went from 0 to 22 per cent, CSAT on AI-resolved tickets stayed within 4 per cent of human-resolved. The service team now has time for the work that really matters, including early signalling for churn prevention on red-zone customers.

What clients usually ask.

Honest answers to the questions we hear in almost every AI-for-service engagement.

What is a realistic ticket deflection with AI?

15 to 35% of tickets in B2B is achievable if the knowledge base is sound and the AI chat is well grounded. In high-touch business it can be lower because customers expect personal contact, in tech-touch it can be higher. Start with a realistic 15% and grow from there.

Which Breeze agents are ready for production?

As of May 2026, Customer Agent (AI chat in portal), Knowledge Agent (FAQ automation) and Ticket Triage Agent (categorisation and routing) are mature enough for production. Sentiment detection and summarisation on ticket history are in beta with good results. We only deploy what demonstrably works in your context.

How long does an AI-for-service engagement take?

Six to ten weeks for the complete rollout: knowledge base clean-up, agent configuration, escalation path to a human and the measurement loop on deflection and CSAT. After that, a monthly cycle to refine the prompts and grounding based on what agents do and do not handle well.

What does AI for Service cost?

Bespoke per situation. Depends on the scope (ticket classification only, or also AI chat in the portal and sentiment detection), the state of your knowledge base and your HubSpot tier. Many clients land this in a RevOps-as-a-Service retainer where strategy and execution come together. Book a call and we will make a proposal for your situation.

What happens if the AI makes a mistake?

Three safety layers. One: agents are grounded on your knowledge base and data, not on the open internet. Two: escalation path on uncertainty or sentiment shift, ticket goes directly to a human. Three: measurement loop on errors, with weekly review and prompt tuning. Ultimate responsibility stays with your service team, not with the agent.

Do we need a mature HubSpot already?

For the AI layer it works better if your data model and knowledge base are in order. Agents run on your data, so property quality and knowledge base quality are the foundation. With teams where that is not yet right, we start with a short CRM architecture action or knowledge base clean-up before we roll out agents. Our maturity scan shows in one session what is still shaky and where the first agents can already run safely.

Do you also work with AI tools other than Breeze?

Our preference is Breeze because we are a HubSpot Platinum Partner and the agents run natively in the platform on your data. For clients with Zendesk we work with Zendesk AI or an Intercom Fin integration. At architecture level we are platform-agnostic: the pattern of grounding, escalation and measurement loop remains the same.

How do you measure whether the AI delivers value?

Four KPIs. One: deflection rate (tickets resolved without a human). Two: CSAT on AI-resolved tickets versus human-resolved. Three: average handling time. Four: escalation quality (how many of the escalated tickets should the AI have handed over earlier). The monthly cycle revolves around these four.

Ready to start?

Ready for AI on the work that is suited for automation?

Book a call. We look at your knowledge base, your ticket volume per type and your HubSpot status. After that you get an honest scope, an agent set that fits and realistic deflection expectations.