Introducing an AI chatbot in customer service is a manageable project structured in four phases: sources and concept, setup, integration, and go-live with monitoring. For your own team, the effort lies primarily in selecting the right use cases, approving knowledge sources, and making decisions. Setup, system connection, and dialogue design are handled by the vendor or a partner. This article outlines the project phases, who contributes what, and what should be clarified before getting started.
At a Glance
Four phases lead up to go-live: sources and concept, setup, integration, launch with monitoring.
The business department selects the use cases, approves knowledge, and conducts testing.
IT provides access and reviews integration and security.
Data protection must be involved right from the start.
A partner adds value when the chatbot deeply integrates into existing systems and processes.
What Should Be Clarified Before Getting Started
Most delays occur because fundamental questions remain open. Four points should be settled before the initial meeting.
The first three use cases. Choose cases that occur frequently and are clearly defined, such as product inquiries, complaints, or status updates. A chatbot that handles three tasks well is far more compelling than one that does twenty half-heartedly. To see which tasks are suitable, read our article on 15 recurring use cases in customer service.
The knowledge sources. Where is the knowledge for these cases located? Manuals, data sheets, websites, FAQs, internal documents. It doesn't have to be perfectly prepared. What matters is that someone decides which sources are approved.
The key stakeholders. One person from the business department to make decisions, and one from IT to provide access. That is all that is needed initially.
Data protection and security. Involve data protection and information security early on. Data processing agreements (DPA), hosting, and retention periods can be clarified alongside setup if these questions are brought to the table in time.
The Four Phases Up to Go-Live
1. Sources and Concept. Knowledge sources are reviewed, the first use cases are defined, and the handoff to the team is planned. The outcome is a clear scope of what the chatbot should—and shouldn't—do at launch.
2. Setup. The sources are connected and prepared, and dialogues for the use cases are created. The business department tests early on with real-world questions and provides feedback.
3. Integration. Where the chatbot is intended to handle transactions, CRM, ERP, or ticketing systems are integrated, as are channels such as websites, messengers, or phone. Standard connectors are up and running within days; custom integrations take longer.
4. Go-Live with Monitoring. The chatbot goes live, often initially on selected pages or for a specific customer segment. Analytics show which questions perform well and where knowledge gaps exist.
How long this takes depends primarily on the sources and integrations. In the project with Lexie, there were less than four weeks between concept and go-live.
Who Contributes What
Role | Responsibilities |
|---|---|
Business Department | Select use cases, approve sources, sign off on dialogues, test with real questions |
IT | Provide access and interfaces, set up single sign-on, verify integration |
Data Protection and Information Security | Review and approve DPA, hosting, deletion periods, and labeling |
Vendor | Setup, connection of sources, dialogue design for core cases, training up to go-live |
Partner (optional) | Gather requirements, build deep integrations, provide operational support |
The majority of the work lies with the vendor. Your own team primarily contributes domain knowledge about customers and processes—and that is something no external party can replace.
After Go-Live
After launch, the business department maintains the chatbot independently. New content is added via knowledge sources, and dialogues can be adjusted without any coding. Analytics show where questions remain unanswered, pointing the way for next steps.
Scaling is done step by step: additional use cases, more languages, further channels. A chatbot that performs well on the website can operate on messaging channels or the phone using the exact same knowledge base.
When a Partner Adds Value
Many implementations take place directly between the company and the vendor. Bringing in a partner makes sense when the chatbot deeply connects into existing systems, when multiple departments or countries are involved, or when a company is already working with a systems integrator familiar with their IT landscape. Partners then handle requirements, integration, and ongoing operations, while the vendor provides the platform.
How Mercury.ai Structures the Implementation
At Mercury.ai, onboarding is included in all packages: setup, connection of standard sources, dialogue design for core cases, and training up to go-live. Afterwards, the business department maintains knowledge and dialogues independently in the No-Code Studio; separate environments for draft, test, and production ensure that changes only go live after testing. Billing starts only upon go-live. The contract runs for twelve months with a cancellation option after three months.
For projects requiring deep integration, Mercury.ai works with solution partners and systems integrators. To find out whether building in-house is worth it over a platform, read our article Build or Buy: Chatbots.
Conclusion
Introducing an AI chatbot is a manageable project when the key questions are clarified upfront: three use cases, approved sources, two responsible stakeholders, and data protection brought in early. The vendor handles the lion's share of the work. Your own team contributes what only they know: your customers and your processes.
Planning to introduce a chatbot? Talk to us.
