To the top

How to successfully implement HubSpot Customer Agent

New AI features are now being released in HubSpot at a rapid pace. In this blog series, we'll take a practical look at what you can do with the different functionalities, what needs they address and what you need to consider to get the most out of them from a business perspective. In the second part of the blog series, our expert Antti Nevalainen provides concrete tips based on practical experience to get Customer Agent up and running and integrated into your existing processes in less than a month.

Read also about HubSpot's Customer Agent and who it is suitable for. 

Step-by-step implementation of Customer Agent

Step 1: Content audit (1-2 weeks)

Before you do anything in HubSpot, check out the 20 most common questions your customers ask. Don't guess, find out from your customer service team. They have the data and the practical experience.

Once you have the questions together, check:

  • Do the answers to these questions already exist somewhere?
  • Are the answers up to date?
  • Are the answers written clearly and understandably?

If the answer to any of these is "no", complete the content immediately. This is the most important stage of the whole Customer Agent project, because the agent cannot answer better than its source material.

I recommend starting with a clear FAQ document. Create a simple, easy-to-update Word or PDF file with 20-30 of the most common questions and direct answers that are useful to the customer. This is the quickest way to ensure you have a high enough quality knowledge base on which to build your first working Customer Agent.

Step 2: Setting up an Agent in HubSpot (1 day)

Setting up an agent in HubSpot starts by going to Service > Customer Agent > Set up your agent. You can then go on to fill in more detailed information and configure your agent. This step typically takes one working day, once the groundwork with the content has been carefully done.

Basic information:

  • Give the agent a name for internal use (the client will not see this)
  • Choose the channels where the agent is active (e.g. website chat, in-app, any other service channels).
  • Define the agent's "personality" (Friendly, Professional, Casual, Empathetic, Witty) to match your company's customer service style and brand.

Add a source of information:

  1. Knowledge Base articles: if you have a HubSpot Knowledge Base in place, the articles will be automatically made available to the agent.
  2. Web pages: enter the URL of the content you want the agent to learn about and set the necessary crawling limits to prevent the agent from crawling the wrong or unnecessary pages.
  3. Documents: upload FAQ documents and other key materials (e.g. instructions, process descriptions, terms of service) that you want the agent to use as a knowledge base.

Escalation rules (defines when the agent will transfer the conversation to a human)

Define clear escalation rules so that the customer is directed to the right place at the right time. Typical rules include:

  • The customer directly requests a human.
  • The agent cannot find a response or the confidence threshold is reached.
  • Certain topics or keywords (e.g. "cancellation", "refund", "complaint") are always routed to a human.
  • The customer is frustrated or expressing dissatisfaction (the agent will recognise this from the tone of the messages).

Step 3: Testing in preview mode (3-5 days)

Do not release the agent immediately, but first use the preview mode and test the agent in a controlled environment. This will ensure that a solution you don't yet trust doesn't end up in production. Systematically test the agentwith questionssuch asthe following:

  • Questions for which you know the answer can be found in the source materials
  • Questions for which there should be no answer (e.g. things you do not communicate to customers)
  • Questions that are complex or require more than one step
  • Incorrectly worded, incomplete or unclear questions
  • Questions in a foreign language

Take notes during testing:

  • Which questions are answered accurately and usefully by the agent
  • In which situations it answers poorly, misleadingly or incorrectly
  • Which answers are too long or too short in relation to the client's needs
  • Whether the escalation rules work correctly (when the conversation moves to a human)

Based on your findings, make the necessary corrections to the content, settings and rules. Retest until you achieve a consistent performance. Don't release an agent into production until you have confidence in its responses and that it will support the customer experience and not degrade it.

Step 4: Release and monitoring (ongoing)

When you release an agent, start with a single channel - for example, a live chat on your website. Monitor the agent's performance daily for the first week, keeping the customer service team actively engaged and asking them for systematic feedback.

I recommend monitoring these metrics in particular at the beginning:

  • Resolution rate: how many conversations the agent resolves without human assistance
  • Knowledge gaps: which questions the agent cannot answer well enough
  • Escalation rate: how often and in which situation the agent transfers the conversation to a human
  • Customer feedback: how customers rate the quality and usefulness of the conversation

HubSpot automatically visualizes these metrics on dashboards, so you can see the big picture and progress at a glance.

Remember also the agent's continuous improvement

This is not a "set and forget" tool, but a constantly evolving part of the customer service process. To get the best business value out of Customer Agent, you should build a light but systematic maintenance routine for it. On a weekly basis, I recommend going through the knowledge gap report, adding missing information to the Knowledge Base or other documents, updating outdated answers and clarifying unclear points.

At a monthly level, it is important to analyse resolution rate trends. In addition, try to identify situations where the agent is successful and where it needs support, check the credit cost against the time saved and the customer experience, and discuss with the customer service team where they continue to patch the agent and what needs to be improved in terms of content, rules or escalations.

Improve agent customer encounters through integration and personalisation

CRM integrations

One of Customer Agent's most powerful features is its ability to leverage CRM data directly in the customer service experience. You can give the agent access to up to 10 contact properties, such as:

  • Customer name
  • Company
  • Products or services purchased
  • Order or contract status
  • Customer segment

In practice, this means that an agent can:

  • Greet the customer by name ("Hi Matti, how can I help you?").
  • Check the status of an order or contract directly from HubSpot CRM.
  • Provide personalised responses based on the customer's history, segment and buying behaviour.

It is important to note that the agent can also update CRM fields. This is very useful for updating the "last contact" field or refining the customer segment, for example, but requires careful planning to avoid accidentally overwriting critical data or important historical information. So plan in advance which fields the agent can update and under what conditions.

Other integrations and custom actions

If the basic Customer Agent is not enough, you can extend its capabilities by building custom actions through API integrations. This allows you to truly connect the agent to the underlying systems and automate steps that would otherwise require customer agent time.

Practical examples of implementations I have done:

  • Retrieving the order number and displaying the order status from an external system
  • Automatically sending a password reset link
  • Booking an appointment directly during a conversation from the calendar
  • Retrieving product information from the product register and displaying essential information to the customer

These require development and careful integration planning, but when implemented correctly, they make Customer Agent a much more powerful and business-valuable tool, especially when you want to integrate HubSpot with ERP, order management or other mission-critical systems.

If you want to build a similar level of automation or integrate Customer Agent into your existing processes, contact us - we can talk through what makes technical and business sense to implement in your environment.

What are the most typical challenges and how do you solve them?

Problem Cause Solution
Agent gives wrong answers Source material is unclear, contradictory or outdated.
  • Write the content clearly, concretely and in a way that helps the customer
  • Update documents regularly as part of an ongoing maintenance routine
  • Remove outdated or incorrect documents completely to prevent them from becoming an agent source
  • Systematically test the agent after each major content update
The agent cannot answer common questions Information is not available to the agent in any format.
  • Build a proper, up-to-date FAQ document.
  • Do not assume that the information can be found on the website. Often critical information is only in internal systems or in people's heads.
  • Writethe answers explicitly open-ended so that any Customer Service Representative, and Customer Agent, can understand them without background knowledge.
The agent is passing on too much of the conversation to the human Escalation rules are too strict or confidence threshold too high.
  • Loosen escalation rules and confidence threshold gradually, not all at once
  • Monitor reporting to see which conversations are escalated and why. Especially topics, channels and times.
  • Add and refine content on the topics that escalate most often, for example with the latest FAQ or Knowledge Base articles.
Responses are too long or too short The source material does not match the expected use case or the agent's tone of voice settings do not match the brand.
  • Chop and edit source documents into clear, short enough chunks
  • Adjust the agent's personality setting to better match your company's customer service style
  • More examples of good branded response templates (tone, length, level of concreteness)

Summary

A Customer Agent is a truly effective tool when the groundwork is done carefully and the agent is integrated into the overall customer service process. It is not an automatic solution to everything, but if designed and implemented correctly, it can take care of a significant part of routine customer service, freeing up customer service agents' time for more demanding and valuable customer encounters and providing fast and consistent responses 24/7. The tool pays for itself quickly, both in terms of saved working time and improved customer experience.

Critical to success is doing your content work properly and ensuring that information is up-to-date and unambiguous. You should test the agent thoroughly before release in different scenarios and languages, monitor its usage and continuously improve the solution based on reporting and team feedback. It's also important not to assume that the agent will work perfectly straight away, but to see it as a constantly evolving part of the service process.

If you are considering implementing Customer Agent and want to spar how it would fit into your architecture, processes and HubSpot environment, drop us a line. We'll go over your data, integration needs and goals. I'll be honest about whether or not it makes sense to pilot right now.

Book a free HubSpot demo meeting with us!

Book a free demo