By KYN AI Advisory Team — AI implementation specialists, Singapore
To qualify inbound leads with an agent, define what a qualified lead is in rules, let the agent enrich the record and ask two or three questions, score the result against those rules, and route it to a named person with the evidence attached. The agent sorts and prepares. A rep decides. Leads that fail the rules go to a nurture list, not the bin, and anything unclear goes to a person.
This article covers inbound leads from forms, WhatsApp and email. For the cross-border and sector-specific versions, see our pieces on lead qualification for financial services in Singapore and lead qualification and pipeline automation in Malaysia.
What does qualified mean for your business?
Write it down as rules before you build anything. A rule is a test with a yes or no answer. Examples:
- the company is registered in a market you serve;
- the company size or revenue band fits your product;
- the request matches a service you offer;
- the timeline is within a window you set;
- the contact has authority or can name who does.
Weight the rules if some matter more. Keep the list to five to eight tests. More than that and nobody can explain why a lead scored as it did.
The test of a good rule is that two reps would give the same answer. "Seems serious" is not a rule. "Gave a budget range and a date" is.
What should the agent do before a rep sees the lead?
Five steps, in order.
- Capture. Write the raw enquiry to the CRM with source, time and channel. Nothing is lost if a later step fails.
- Deduplicate. Match on email, phone and company name. An existing customer or open deal routes to its owner, not to the new-lead queue.
- Enrich. Look up facts you are allowed to use: the company name against a registry, the website, the sector. Store each fact with its source.
- Ask. If key rules cannot be tested, send one short message asking for the missing items. Two or three questions is enough.
- Score and route. Apply the rules, write the score and reasons, and assign to a named person.
Every step writes to the log with the actor and time. A rep opening a lead sees what was found, what was asked and why it scored as it did.
Which questions should the agent ask?
Ask only what changes the routing. For a services business, three are usually enough:
- What are you trying to get done, and by when?
- How many people, sites or accounts does this involve?
- Who else is involved in the decision?
Rules for the questions:
- ask in the language the lead used;
- put them in one message, not a long chain;
- accept free text and extract the answers, then let the lead correct the summary;
- stop after one reminder if there is no reply.
On WhatsApp, the lead has started the conversation, which changes the consent position compared with cold outreach, but you still need a stop mechanism and a record of the opt-out. Check the rules for the lead's country, such as the PDPC's Do Not Call Registry guidance for Singapore numbers.
How should scoring and routing work?
Keep the logic visible. A table beats a model you cannot explain.
| Outcome | Rule | Route | |---|---|---| | Ready | Meets all must-have rules and has a timeline | Named rep, alert the same hour | | Possible | Meets must-haves, missing one detail | Rep queue, with the open question | | Not yet | Fails a timing or size rule | Nurture list, review on a set date | | Out of scope | Wrong service or market | Polite reply, referral if you have one | | Unclear | Rules cannot be tested | Person reviews |
A lead that fails a rule is not a bad lead. It is a lead for a different time. Keep the record and the reason.
Set a response-time target and measure it. Each alert goes to a named person with a backup if they do not act within the limit you set.
What should never be automatic?
Per our approval rules, outbound messages, money and irreversible actions wait for a named person. For inbound qualification that means:
- an agent may send the clarifying questions from approved templates, but a person approves any message that quotes a price, a term or a date;
- an agent may not mark a deal as lost or delete a record;
- an agent may not decline a lead in a way that carries your name unless the wording is approved and the reason is a clear rule.
Treat the text the lead sends as untrusted. A form field can contain instructions aimed at the agent. See prompt injection in business AI agents.
How do you check that it works?
Run a test before it goes live and then monthly.
- Take 100 past leads whose outcomes you know.
- Run them through the rules. Compare the agent's route with what happened.
- Count qualified leads that were routed as not yet or out of scope. Those are the costly errors.
- Adjust the rules, not the model, when errors cluster.
- Track time from enquiry to first human touch.
Once live, sample 20 leads a week for the first month.
What does a worked example look like?
Take an illustrative accounting practice that receives about 60 enquiries a month by web form and WhatsApp. It sets five rules: located in Singapore, fewer than 50 staff, needs a service it offers, wants to start within 60 days, and the contact can name the decision-maker.
| Enquiry | Rules met | Route | |---|---|---| | Retail firm, 12 staff, wants monthly bookkeeping, start next month | 5 of 5 | Ready: named partner, same-hour alert | | Start-up, 4 staff, wants payroll, no start date | 4 of 5 | Possible: rep queue with the open question | | Manufacturer, 200 staff, wants audit support | Fails size | Not yet: referral if you have one | | Individual asking about personal tax | Fails service | Out of scope: polite reply | | "Need help with accounts" and nothing else | Cannot test | Unclear: person reviews |
The numbers are illustrative. The table shows that every outcome has an owner and a next step, and none is silent.
What goes in the summary a rep sees?
A rep should be able to act without opening the thread. Give them six lines:
- who wrote, from which company, through which channel;
- what they asked for, in their words;
- which rules passed and which did not, each with its source;
- what the agent asked and how they answered;
- any existing record, such as a past customer or open deal;
- the suggested next step and the deadline for the first reply.
The rep can correct any line, and the correction is logged. Over a month, those corrections show which rules to adjust.
Where KYN fits
KYN builds qualification as rules plus agents in the client's own cloud account. For an insurance brokerage, replies and enquiries are sorted at about S$0.01 per reply, each action is logged with the actor, and outbound messages wait for a named person. Agents run on their own restricted accounts, and the database uses row-level security.
The first 14 days are a review of your stack and interviews with your team. A full build ships in under 90 days. See sales systems and integrations.
Frequently asked questions
How many questions should an agent ask a new lead?
Two or three. Each extra question lowers the reply rate and delays the handoff. Ask only what changes the routing.
Should the agent score leads with a model or with rules?
Start with rules. They are easy to explain, test and change. Use a model to read free text and extract facts, and let rules make the decision.
What happens to leads that do not qualify?
They go to a nurture list with the reason recorded. A review date brings them back when circumstances may have changed.
Checked 30 October 2026.
Curious whether Marketing agents fits your business? Talk to KYN on WhatsApp — no forms, just a conversation.
