People who search for Druid conversational AI are usually looking for a conversation that finishes a process, not one that only answers a question. That is the right instinct. This article does not review Druid or list its features. It explains, for a business of 10 to 250 staff, how to design customer conversations that complete real work, what to check in any process-led conversational product, and how to build the same pattern with AI agents and workflows on the levelpartner.ai platform.
Answering versus completing
An answering agent tells a client which documents you need. A completing agent asks for them, sends the upload link, checks what arrived, chases what is missing and tells your team when the file is ready. The second one removes work. The first one moves it.
Most of the cost in a growing service business sits in these small processes: onboarding a new client, collecting documents, rescheduling, confirming details before a meeting. Each one involves several messages and someone remembering to send them. A conversation that completes the process removes that memory burden.
What to check in any process-led product
- Can the conversation start a multi-step process and track where it is?
- Can it pause for days while waiting for the customer, then resume with context?
- Where is the process state stored, and can your team see it on the client record?
- Who changes the process when your business changes, and how long does that take?
- What happens when the customer goes off script mid-process?
If the process state lives in one system and the client record in another, your team will check two screens every day.
Before you start
- One process you want to automate end to end, written as a numbered list of what happens today.
- The forms, documents or payments the process needs.
- Pipeline stages or tags that show where each client is.
- The point at which a person must step in.
How to set it up
- Open Automation and select Workflows.
- Choose Create Workflow to start from scratch.
- Click Add New Trigger and pick the event that starts the process.
- Add trigger filters so only the right contacts enter.
- Click Add Action and select the first fixed step.
- Repeat Add Action for each wait, check and reminder on your map.
- Test with a new contact that has never entered the workflow.
Onboarding needs an engagement letter, identity documents and last year's accounts. Picture an administrator sending each request by email and tracking replies in a spreadsheet, at about 45 minutes per client. With the workflow sending and chasing, and the agent answering questions like "which bank statements?", the administrator would only review completed files. At ten new clients a month, that is roughly seven and a half hours of chasing handed over.
Why the split between workflow and agent matters
Let the workflow own the order of steps, and let the agent own the conversation. A workflow is predictable: it sends the chase on day three every time. An agent is flexible: it understands "I sent that yesterday, did you get it?" Putting the fixed logic in the agent makes the process unreliable. Putting the conversation in the workflow makes it robotic.
In practice the agent hands over to the workflow by updating a field or a tag, and the workflow hands back by moving the client to the next stage. Keep that handshake small and named, so anyone on your team can read the client record and say exactly where the process stands without opening the conversation history.
For more on this division, read AI business process automation. The AI follow up article covers the chasing pattern on its own.
Teams often automate the whole process at once. Start with the step that eats the most staff time, usually chasing, and add the rest once that runs cleanly.
Weekly time to run it
Budget half an hour a week to review stuck clients and an hour a month to refine wording. The process itself runs without supervision.
When not to use this pattern
If a process happens a few times a year, automating it costs more than doing it by hand. If every client needs a bespoke path, keep a person in charge and use the agent only for questions. For high-volume service teams with dozens of agents, a contact-centre product may suit better; see enterprise conversational AI.
Troubleshooting
Clients receive a chase after they have replied. The workflow is not checking for the upload before chasing. Add a condition on the form submission.
The agent says a document is missing when it has arrived. The agent cannot see the stage change. Make sure the workflow updates the stage or a field the agent reads.
Clients reply to the chase with questions the workflow ignores. Route replies to the agent, not to a dead end.
Frequently asked questions
Is Druid conversational AI similar to AI agents on the platform?
Both aim to let conversations do real work. The platform's agents sit inside the CRM, calendars and workflows, so the process state and the client record are in one place. Compare any product on that basis.
Can the agent collect payments as part of a process?
Yes, through the workflow rather than the agent. A workflow step can send a payment link at the right point in the process, and your team sees on the client record whether it has been paid before the next stage begins.
What process should we automate first?
Pick the one with the most chasing: document collection, onboarding or appointment confirmation. Those give the clearest saving.