Short answer
AI in sales is more than ghost-writing messages. Harry Lui proposes a four-step method — real records, needs judgement, follow-up rhythm and team review — to turn strong sales experience into a repeatable workflow.
When companies talk about AI in sales, the first idea is usually auto-writing WhatsApp messages, generating follow-up emails or building a chatbot. These features save time but do not necessarily improve closing.
That is because the real gap in sales performance is often not whether something is said nicely, but whether you correctly understand what is happening with the customer.
The same “let me think about it” may mean the customer has not yet understood the value, has no internal budget, is worried about execution risk, or is simply declining politely. If AI has no real context and just rewrites that line into three more persuasive replies, the company is only packaging a wrong judgement more smoothly.
What is most worth copying from great salespeople is their judgement process
Experienced salespeople usually pay attention at once to a customer’s explicit needs, unspoken concerns, decision role, time pressure and past interactions. They may not be able to fully explain these judgements, but they naturally decide when to probe, when to provide information and when to pause.
If a company hands the team only its top seller’s best line, it copies only the surface. What is really worth capturing is how that person reads customers, handles incomplete information and adjusts pace in response to new evidence.
AI can help a company turn these tacit judgements into a clearer workflow.
An AI sales system should have four stages
1. Start from real records
Analysis should be based on call records, emails, WhatsApp conversations, customer background and past actions that are authorised for use, not solely on salespeople’s after-the-fact impressions.
The purpose of records is not to monitor staff but to give judgements evidence, and to let colleagues correct transcription errors or add important context. The company needs to define data use, access rights and retention.
2. Separate facts, needs and speculation
AI can organise the customer’s explicit requests, confirmed constraints, unanswered questions and possible hidden needs. But “possible” must remain a hypothesis; it must not be treated as the customer’s real view just because AI wrote it confidently.
A better analysis does not give one conclusion. It lists what is currently known, what is still unknown, and the next question most worth asking.
3. Design a purposeful follow-up rhythm
Follow-up does not mean contacting the customer every day. Each contact should have a clear purpose, such as confirming the decision process, adding risk information, showing a relevant example, involving technical staff, or helping the customer work out the cost of not acting.
AI can draft the next step based on the customer’s status, but the salesperson must still judge timing, relationship and the limits of commitment. In high-value sales especially, automated messages should not replace real understanding.
4. Let every win and loss train the team
After follow-up, the company should record which judgements were supported, which messages made things clearer for the customer, which wording triggered defensiveness, and what finally moved the deal forward or stopped it.
Once a company has accumulated enough good-quality samples, it can compare how strong and average salespeople differ in questioning, judgement and pace, then turn those differences into training, checklists and workflows.
This is more concrete than asking staff to “follow up a bit more”, and closer to real closing ability than simply counting messages.
Management should not turn AI sales into a message factory
If performance metrics look only at send volume, contact counts or the amount of AI-generated content, teams easily produce large numbers of low-quality contacts. A customer receiving more messages is not closer to a decision.
What management should measure is whether needs have been clarified, whether the next step is confirmed, whether key risks are addressed, whether commitments can be delivered, and whether sales forecasts are becoming more accurate.
AI’s role is to lower the cost of organising data and preparing work so salespeople have more time to understand customers — not to let the company disturb more people at lower cost.
Start with one real customer journey
A Hong Kong SME can first choose one important type of customer and compile the full record from first enquiry, needs analysis, proposal and follow-up through to closing and delivery. Management and the front line mark the key judgements together, then let AI help build an analysis template and next-step suggestions.
A trial does not need to connect every system at once. First make sure input data is reliable, suggestions are explainable and colleagues are willing to correct them, and check whether the method works against real closing results.
Only when “data, judgement, action, review” runs as a stable cycle does sales experience move from individual ability to a company asset.
Practical supplement
Three-column notes before follow-up
Use only authorised customer records, let frontline staff correct the AI’s understanding, and then decide the timing of follow-up.
Your entries remain only on this page and will not be saved after you refresh or leave; use Copy worksheet to keep the text.
Discussion question
For a recent opportunity, which unknown most needs clarification in the next follow-up?
Put it into practice
If you want to connect customer analysis, proposal drafting and WhatsApp follow-up into a workable process, see:
AI成交引擎實戰營 (AI Sales Engine Bootcamp, in Chinese) (external site) ↗Source: Original commentary by Harry Lui
Last updated: 2026-10-02（first published: 2026-10-02）