5 levels of agentic maturity

This might be an area you might already be using. These tools are used to automate repetitive tasks using predefined rules; they don’t have reasoning/learning capability. These models also need to be first set up by a human. They need to be provided with the information and rules.

• A chatbot is essentially someone having a conversation with AI. They are commonly seen on websites with a pop-up at the bottom of the screen where you can ask questions. They can advise you where to go and what processes to follow.

• A CoPilot is a tool that is integrated within apps, like Microsoft CoPilot, and it can make changes with permission.

More advanced tools are now learning from their processes, for example, when someone marks that the answer provided wasn’t helpful.

As the name suggests, these agents find information. Like ChatGPT, they can find answers to questions and suggest a course of action, but they can’t complete the action for you. A member of staff will then have to make the final decision on what to do.

Here is an example (provided by ChatGPT):
An agent that listens to a staff member typing “Which members attended our 2023 conference and haven’t renewed yet?”

The agent:
• Queries the CRM
• Returns a list with insights
• Suggests: “You could send a re-engagement email or call these 10 lapsed members”
• The member of staff then decides whether to send the email or not

By this step, AI agents are moving away from needing input from a member of staff. This step means AI are starting to complete simpler tasks that humans could do whilst focusing on one area.

For example, you might have an event registration agent that:

• Interacts with a member via chat
• Recommends an event based on interests
• Registers them
• Sends confirmation
• Updates their CRM record with session attendance

These agents can coordinate actions across multiple systems or business areas. The tasks at this point are becoming more complex.

A member writes: “Can you register me for this Friday’s event and check if I’m still eligible for the mentoring program?”

The agent:
• Pulls your membership status from CRM.
• Checks event availability from an event platform.
• Looks up mentoring eligibility from a third system.
• If everything checks out, it:
o Registers the member
o Books a mentoring session
o Updates their contact record with tags
o Sends a follow-up email with a calendar invite and links

AI agents in different departments or systems autonomously collaborate, hand off tasks to each other. They can even supervise other agents, regardless of which tools they’re built on.

Example:
• A support agent identifies a member complaint that relates to a billing issue
• It escalates to a finance agent, who pulls payment records and identifies the error
• Then a CRM agent updates the member’s record and sends an apology email with a discount
• A compliance agent logs the incident for reporting
This happens without humans being involved, even though these agents might live in different tools (e.g., Zendesk, Stripe, HubSpot)

Whilst the pace of AI is rapidly evolving and it can complete more tasks, it can be important to reassure your team that their jobs are safe. It can be helpful to set boundaries for what AI can and cannot be used for.

Finally, we wanted to make a note on the importance of considering the ethical use of AI. Zentso supports sustainable development and are passionate about the work of non-profits and membership organisations driving social change. There has been much discussion on the climate effects of using AI tools. We encourage you to consider this when choosing to use AI.