AI customer support on the levelpartner.ai platform answers the questions your existing clients ask every week, on chat, SMS and WhatsApp, from a knowledge base you control. Clients get an answer in seconds at any hour, and your account managers stop spending their mornings on "where is my document" messages. This guide covers support for people who already pay you, not new leads, and shows how to build it so it escalates properly.
What support automation changes for a firm your size
In a company of 40 to 150 people, client service usually sits with the same people who are meant to grow accounts. Every repeat question pulls an account manager away from renewal and upsell work. That is a quiet reason revenue goes flat: the people who could expand accounts are busy answering logistics.
An AI support agent takes the predictable layer. Opening hours, document checklists, how to reschedule, what happens next in the process, where to upload a file. It answers from your written material, logs the exchange on the contact record and passes anything unusual to a named person. Your team keeps the conversations that need judgement.
Conversational AI for customer support versus a help page
A help page waits for the client to find the right article. Conversational AI for customer support meets the client in the channel they already use and answers the question as they phrased it. The trade-off is control: a static page says exactly what you wrote, while an agent paraphrases. That is why the knowledge base and the escalation rules matter more than any other setting.
Before you start
- The channels clients already use, connected to your account: web chat, an SMS number, WhatsApp if you use it.
- Your thirty most common client questions, pulled from the shared inbox, with approved answers.
- A list of topics the agent must never handle, such as complaints, refunds, contract changes and anything regulated.
- A named person or small team who receives escalations, with notifications switched on.
- Existing clients tagged in Contacts so the agent can tell a client from a prospect.
How to set it up
- Go to AI Agents > Conversation AI and open Agent List.
- Select an existing bot or create a new one.
- Open the Bot Training tab.
- Choose an existing Knowledge Base from the selector, if you have one.
- Select All URLs with this Path to crawl one site section.
- Enter the full URL, including https://, and click Get Data.
- Review the pages found and select only the relevant ones.
- Click Train Bot and wait until each URL shows Trained.
- Click + Add Q & A and enter a question and answer.
- Save the Q&A and test variations of the question.
- In Bot Goals, set up the Human Handover action for every topic on your never-handle list.
- Enter a question in Test your Bot and check the reply, including one that should escalate.
Where it fits in the client journey
Support questions cluster at predictable moments: the first week after signing, the days before a deadline and just after an invoice. Map those moments before you write the knowledge base. A mortgage broker hears the same document questions after every application. A real estate team hears the same viewing and offer questions. Consultants and agencies field status and scheduling questions during delivery.
Say the firm signs 25 new clients in a quarter, and each sends around eight logistics questions in the first month. That is 200 messages landing on six account managers, or about 33 each. If an agent answered the predictable 150 and escalated the rest, each manager would handle roughly eight a month instead.
Weekly running time
Plan one focused day to build, mostly writing answers. After launch, the support owner spends about 45 minutes a week reading escalated threads and a sample of resolved ones. Every escalation that the agent should have handled becomes a new knowledge base entry. Every answer that went wrong becomes a stricter instruction.
The usual error is pointing the support agent at prospects and clients alike. A prospect needs qualifying and booking. A client needs accurate answers about their own account. Run separate agents for each, and route by tag. For the prospect side, the AI sales agent guide covers the selling job.
When not to automate support
Do not let the agent answer anything that depends on the client's specific account data unless that data lives in contact fields the agent can read reliably. Do not let it handle a complaint, even a mild one, because an unhappy client wants a person. And if your firm has fewer than a dozen repeat questions a week, a well-organised inbox may be enough. The AI receptionist article covers the same thinking for phone calls.
Troubleshooting
Clients get answers meant for prospects. The agent cannot tell them apart. Tag clients in Contacts and add an instruction that changes behaviour by tag, or split into two agents.
Escalations sit unread. The assigned user is not checking Conversations. Add an internal notification by email or SMS in a workflow and assign to a team rather than an individual on leave.
Answers drift from your current process. The knowledge base still holds an old document. Remove superseded files rather than adding new ones on top, and date each entry.
The agent loops on the same question. The client is asking something the knowledge base half covers. Add a direct answer and an instruction to escalate after two unresolved replies.
Frequently asked questions
Can the agent see a client's account details?
It can read contact fields and conversation history held in the CRM. It cannot see systems that are not connected to the platform. Keep account-specific answers to fields you trust and escalate the rest.
Does it replace our support inbox?
No. Conversations remains the inbox, and the agent works inside it. Your team sees every thread, can take over at any point and can read the full history before replying.
What should we measure?
Track the share of threads closed without escalation, the time to first reply and the number of escalations per week. A falling escalation count means the knowledge base is improving. Our wider guide to conversational AI explains how these conversations are structured.
Will clients know they are talking to AI?
They should. Say so in the first message and offer a person on request. Clients accept quick, accurate automated answers far more readily than a bot pretending to be human. See AI text messaging for the SMS side of that disclosure.