AI AGENTS

AI agents at the wheel of your commercial engine.

Not another chatbot or prompt library. We build agents that run always-on on your CRM data, in HubSpot Breeze, OpenAI or Claude, with a human team lead as owner. One measurable outcome per agent, no hype demo.

Schedule a strategy call Take the Maturity scan first

What every Addmark agent always has

01
Grounded on your CRM data No loose ChatGPT prompts, but grounding on HubSpot objects, properties and knowledge base
02
A human as team lead Every agent has an owner within your team who reviews and steers the output
03
A measurable outcome Reply rate, deflection, content throughput or churn signals, not vague productivity

AI work delivered for

HubSpot has positioned itself since May 2026 as an agentic customer platform. That fits how we build agents: grounded on the CRM, with a human team lead and a dashboard where you see the effect. Not running ChatGPT, but running HubSpot.

Five things every Addmark agent always has

A specific job, not "everything with AI"
Grounding on your CRM data, knowledge base or products
An owner within your team who reviews and steers output
An outcome you can measure in HubSpot, not in tokens
An escalation path to a human for cases the agent cannot handle

What an agent explicitly is not

A chatbot on your website with loose prompts
A ChatGPT licence distributed across your team
A classifier that is correct in 60 per cent of cases
A tool your RevOps lead has to explain afterwards

Per engine

AI agents touch four jobs

AI Agents is not a standalone pillar, it is a layer on top of Marketing, Sales and Service. Per job the agent looks different, runs on different data and is managed by a different team lead. Click a job to see the setup.

Pipeline

AI on outbound prospecting and deal momentum

In Sales, the Prospecting Agent runs on outbound and pipeline monitoring. It finds and qualifies new accounts in your ICP, enriches via Apollo or Clearbit, places them in HubSpot and prepares the first outreach step. Your SDR starts the day with a filled queue, not an empty field.

Owner: Carel. Stack: Apollo, Clearbit, HubSpot Sales Hub, Breeze Prospecting Agent or a custom OpenAI flow.

More about the Prospecting Agent

Demand

AI on content throughput and lead qualification

In Marketing, the Content Agent produces blog drafts, LinkedIn posts, ad copy and email cadences grounded on your positioning, industry and case database. No generic output, but material that fits your voice.

Owners: Dante and Sander. Stack: HubSpot Breeze, Claude or OpenAI, with grounding on your content library.

AI on deflection, ticket triage and knowledge

In Service, the Customer Agent answers repeat questions in your customer portal or chat, grounded on your knowledge base. A classifier sorts and routes tickets in Breeze Ticket Triage. Sentiment detection escalates to a human as soon as it falls outside scope.

Owners: Carsten and Kim. Stack: HubSpot Service Hub, Breeze, knowledge base articles as grounding source.

More about the Customer Agent

Retention

AI on early churn signals and save flows

The Churn Agent looks beyond the cancellation. It combines CRM engagement, product usage and support sentiment into a health score per customer, signals risk before a renewal is in danger and automatically triggers a save flow with a task for the CSM, a targeted email sequence or a retention offer. For B2B teams that want to predict retention rather than react.

Owner: Carsten. Stack: HubSpot Service Hub, custom properties on contact and company, Breeze or a custom OpenAI scoring flow, with grounding on your usage data.

More about the Churn Agent

Case, Visma Raet · Visma Raet

Van losse tools naar een pipeline die de directie vertrouwt

A Content Agent and a light Prospecting Agent alongside the same team.

Visma Raet already had a strong HubSpot foundation. The question was not whether AI, the question was where AI genuinely adds value without weakening the brand voice. We started with a scoped Content Agent, grounded on the existing positioning and their case database, with the marketing lead as owner.

Alongside that, a light prospecting flow that enriches accounts in the ICP and places them in HubSpot, with the SDR as reviewer. No full automation, but measurable throughput. After six weeks both agents were in a fixed monthly cycle with grounding update and KPI review.

Who builds and runs the agents.

Per agent you assign a team lead on your side. On our side, Carel sits in as architect at every kick-off, after which Dante, Sander, Carsten and Kim take over execution and maintenance.

Carel Schrier

RevOps Lead / Agent architect

Owner of the Prospecting and Customer Agent engagements. Writes the scope, chooses the stack and stays involved in reviews and KPI checks.
Dante Zwanenburg

Content & Ads Agent owner

Builds the Content Agent together with Sander. Grounding, prompts, output review and connection to your content calendar and LinkedIn cadence.
Carsten Huiskamp

Customer Agent & data owner

Works on signals, data models and the Customer Agent. Data quality and grounding on product and ticket data are his domain.

Frequently asked questions

Frequently asked questions about AI Agents.

Questions we hear at every introduction. Honest answers, no vendor talk.

Is this the same as ChatGPT?

No. ChatGPT is an interface; an agent with us is a unit of work that runs always-on on your data, in your stack, with a specific job and a measurable outcome. Under the bonnet an agent can use a large language model, often Claude or OpenAI or HubSpot Breeze, but that is the engine, not the agent itself. What makes the agent is grounding, prompts, escalation paths and an accompanying dashboard.

What if I have no AI strategy?

In our experience that is the rule rather than the exception. We then start with the AI Readiness Scan. Three to four weeks, we look at data readiness, score use cases on impact and feasibility, classify against the EU AI Act and deliver a 90-day roadmap with the top-3 agents to start with. Only then do we build.

How do you measure whether an agent is successful?

Per agent we agree upfront on a handful of hard KPIs, not vague productivity. For a Prospecting Agent those are reply rate, quality score of sourced accounts and SDR time saved. For a Content Agent those are throughput, editing time per piece and organic reach. For a Customer Agent those are deflection rate, CSAT of AI conversations versus human, and escalation rate. We build the dashboard as a deliverable of the agent.

Which tools run under the bonnet?

Per agent we choose the stack. The default choice is HubSpot Breeze, because in the B2B mid-market it integrates best with the rest of your CRM. For content work we often fall back on Claude (Anthropic), for structured tasks on OpenAI. For data enrichment Apollo or Clearbit. For classification and sentiment sometimes a light custom flow. What we do not do: choose a tool because it is in the hype. We choose on grounding quality, integration with HubSpot and cost per action.

Where does our data live?

By default all CRM data sits in HubSpot, region EU. For grounding files (knowledge base, positioning, case database) we decide per agent whether they land in HubSpot, in an EU Claude context or in a configured vector store. We document the data flow per agent and classify against the AI Act, so your IT or compliance team can assess it. We do not use consumer AI tools that can be trained on your production data.

What if the agent says or does something wrong?

Every agent has a human team lead who reviews output before it reaches the customer. For content work, nothing goes live without a final editor. For service work we have an escalation path: as soon as a conversation falls outside scope, it goes to a human. For data work the agent leaves an audit trail; you can always trace back to the source conversation and the prompt. No black box.

Ready for your first agent?

Which agent do you switch on first?

Schedule a 30-minute strategy call. We look at your current stack, your data readiness and which agent delivers the fastest measurable gain in your situation. Not a sales call, just scope work.

Want a baseline first? Take the Maturity scan, then we use the outcome as the starting point for the call.

Schedule a strategy call Take the Maturity scan first

AI agents at the wheel of your commercial engine.

What we mean by an AI agent

Five things every Addmark agent always has

What an agent explicitly is not

One agent per commercial job, not one mega-AI for everything.

Prospecting Agent

Content Agent

Customer Agent

Churn Agent