If you are researching Cognigy conversational AI, you are probably looking at contact centre automation and deciding how much conversation design your team wants to own. We do not describe or assess that product here. This article answers the question underneath: should a mid-sized firm design conversations as flows, or configure an agent that works from knowledge and rules, and how do you set up the second option on the levelpartner.ai platform?
Flow design versus agent configuration
There are two broad ways to build conversational automation. In the flow approach, a designer maps each conversation path, with steps, conditions and integrations. It gives precise control and suits large operations with dedicated designers. In the agent approach, you write a goal, rules and knowledge, and the AI decides how to phrase each reply within them.
Neither is better in general. The flow approach wins where every step must be exact and auditable. The agent approach wins where conversations vary and nobody has time to maintain hundreds of paths. Most firms of 10 to 250 people are in the second group.
Questions to settle before choosing
- How many distinct conversation types do you handle, and how often do they change?
- Who will maintain the automation, and what skills do they have?
- Must every reply follow approved wording, or is approved meaning enough?
- Where should the conversation's outcome land: a CRM, a ticketing system, both?
- What channels matter: phone, chat, SMS, WhatsApp, social?
If you have a handful of conversation types, no dedicated designers and a CRM at the centre of your sales and service, configuration usually beats flow design.
Before you start
- A list of your top ten conversation types with the outcome you want for each.
- A knowledge base built from current FAQs and process notes.
- Calendars, pipelines and tags in place, so the agent has somewhere to record outcomes.
- Hand-off destinations for each conversation type that needs a person.
- One owner for upkeep.
How to set it up
- Go to AI Agents > Conversation AI and click Agents List.
- Click the agent, then Bot Goals.
- In Information Collection, select Name, Email and Phone.
- Add custom questions for each conversation type's required details.
- Open Update Contact Field and choose where each answer goes.
- Turn on Skip if Already Filled for details the contact has.
- Select Trigger Workflow so each outcome starts its automation.
- Set up Email Notification Settings to alert your team.
- Turn on Enable Conversation Summary for the account owner.
- Click Save and test each type in the test panel.
Where flows still belong
Even in the agent approach, some steps are better as fixed workflows. Booking confirmations, reminders, document requests and internal alerts run more reliably as automation in Workflows than as AI decisions. Let the agent handle the conversation and let a workflow handle what happens after it. That split gives you flexibility in talking and predictability in follow-through. It also makes changes safer, because editing a reminder never alters how the agent speaks, and rewording an answer never breaks a reminder.
Suppose a brokerage handles eight conversation types: new enquiries, document questions, appointment changes, rate expiry, referrals, complaints, payment queries and general questions. Flow design might need around 120 mapped paths to cover them. Configuration would need three agents with eight goals and a hand-off rule for complaints. With one part-time owner to keep it current, configuration is the option that person could realistically maintain.
Running time
Configuring three agents takes two to three days. Upkeep is about an hour a week per agent in the first month, falling once the knowledge base settles. Flow-based systems usually need more design time for each change; ask any vendor to show you how long a typical edit takes.
Teams that move from flow tools to agents try to recreate every path as an instruction. The agent then becomes rigid and confused. Describe outcomes and limits, not paths.
When flow design is the right call
If you run a large contact operation with strict scripts, compliance-reviewed wording for every reply and many back-office integrations, a dedicated flow-design platform may suit you better. For background, see contact center artificial intelligence and conversational AI technology. For the agent approach in depth, read the conversational AI contact center guide.
Troubleshooting
Exact wording is paraphrased. The instruction to use it verbatim is missing or weak. State it as a must-never-paraphrase rule and keep the wording short.
Agents overlap on the same conversation. Two agents share a channel without clear goals. Give each channel one agent, or route by tag.
Outcomes are not recorded. Actions are missing for some goals. Map every goal to at least one action.
Topic changes confuse the agent. Its goals are too narrow. Allow it to answer a second question and then return to the first.
Frequently asked questions
Does this compare the platform with Cognigy?
No. We do not describe another vendor's features or pricing. Use the questions above with each vendor and judge on your own requirements.
Can the agent use approved scripts?
Yes. Store approved wording in the knowledge base and instruct the agent to use it unchanged for those topics. Test those topics more often than others.
Do we need developers?
Not for configuration on the platform. You need a clear writer who knows your conversations. Custom integrations beyond the platform are quoted to scope.
Is the phone handled the same way?
A voice agent works from the same knowledge and rules, with its own instructions for spoken pace. Many firms launch chat first and add phone a month later.