By KYN AI Advisory Team — AI implementation specialists, Singapore
Score every account on three separate numbers: fit (does the company match who you sell to), intent (is it showing interest) and timing (is there a reason to act this month). An AI agent can compute all three from data you already hold, but it should not blend them into one score. A single number hides which part is weak, and it hides when the data has gone stale. Keep three columns, write the rule for each, and let a person pick who gets a call.
Why three scores and not one?
A blended score answers no question. An account scoring 70 might be a perfect fit with no interest, or a poor fit that opened four emails. The next step differs for each.
- High fit, low intent: nurture, or an introduction through someone you know.
- Low fit, high intent: check the fit before spending time. Often it is a student, a competitor or a vendor.
- High fit, high intent, no timing signal: queue it and watch for a trigger.
- All three high: a person calls this week.
Many AI business development tools return one ranked list. A ranked list still needs a rule behind each rank, or nobody can tell why account 14 beat account 15.
How do you score fit?
Fit is the slowest-changing number and the easiest to make explicit. Write it as a short checklist a person could apply by hand.
- Country and entity type. For example, a Singapore private limited company with a registered address.
- Size band. Headcount or revenue band you serve well. Use ranges, not a point.
- Industry. The sectors where you have shipped work and can show proof.
- Systems in use. Evidence of the tools your offer replaces or connects to.
- Disqualifiers. Public sector, companies under a certain age, competitors.
Give each criterion a weight of 0, 1 or 2. Sum them. Store the reason for every point, not only the total. When a sales lead asks why an account scored 7, the answer is a list of five lines.
Fit is where an agent helps most with research. It reads a website and a registry record and fills the checklist, citing the page it used. A person spot-checks 20 accounts a week. If more than two are wrong, fix the rule, not the account.
How do you score intent?
Intent is what the account has done, that you can see and are allowed to see. Use first-party signals:
- visits to pricing or case study pages (from your own analytics);
- replies to outreach, sorted by type: interested, not now, question;
- form submissions and WhatsApp enquiries;
- attendance at your events or a webinar;
- repeat visits from the same company domain.
Weight by effort. A reply that asks a question outweighs three page views. Count each signal once per week so a single noisy day does not dominate.
Third-party intent feeds exist. Treat them with care: their match to your account list is often loose, and the data is hard to audit. If you cannot trace a signal to an event you can show a colleague, do not let it move the score.
How do you score timing?
Timing answers a different question: is there a reason this account would act now? It is the number that goes stale fastest, and the one most agents get wrong. We cover the failure modes in lead scoring decay in seasonal markets and the account sequencing trap.
Useful timing signals, each with a date:
- a new director or finance lead joined in the last quarter;
- a contract or licence renewal date you know of;
- a financial year end, when finance teams plan spending;
- a new filing, office move or funding announcement;
- a reply of "not now, ask in March".
Every timing signal needs an expiry. "New hire in the last quarter" lapses when the quarter ends. "Ask in March" becomes live on 1 March and lapses at the end of April. An agent that stores the signal without the expiry will keep ranking dead leads.
How should the agent combine the three?
Do not multiply or average them in a hidden formula. Use a visible rule table.
| Fit | Intent | Timing | Action | |---|---|---|---| | High | High | Any | Person calls this week | | High | Low | Live | Draft a personalised first touch for review | | High | Low | None | Watch list, re-score monthly | | Low | High | Any | Person checks fit before any contact | | Low | Low | Any | No action |
The agent proposes the action and writes the reason. A person approves anything that sends a message under your name. This matches the pattern in what an AI sales agent can and cannot do.
How do you know the scoring works?
Check it against outcomes, not against how it feels.
- Take the last 30 won deals and the last 30 lost ones. Score them as they stood three months before the outcome. Do the wins score higher on fit?
- Count how many of the accounts the agent put in "call this week" a person agreed with. If a person rejects many of them, the rules need work.
- Look at the rejected accounts. Pick 20 from "no action" and ask a salesperson if any should have been called.
- Re-run monthly. Rules drift as your offer and market change.
Small samples mislead. With 30 deals, a three-deal gap is noise. Use the exercise to find broken rules, not to claim a precise accuracy.
What does a scored account look like?
Here is one illustrative account, not a client result. A 60-person Singapore wholesaler is scored by an agent.
| Dimension | Score | Reasons recorded | |---|---|---| | Fit | 7 of 10 | Singapore private company (2), headcount in band (2), sector shipped before (2), uses a spreadsheet for stock (1), no disqualifier (0) | | Intent | Medium | Visited pricing twice this month; asked one question by WhatsApp | | Timing | Live until 30 November | New finance manager listed on its site on 3 October; signal expires after 8 weeks |
Rule table result: high fit, medium intent, live timing. The agent drafts a first message that cites the new finance hire and puts it in the review queue. A salesperson edits and sends.
Three rules keep this honest:
- Every point has a reason line and a source. A score without reasons is deleted.
- Timing signals carry an expiry date in the same row.
- Anyone can override a score, and the override and the name are logged. Overrides are the clearest test data you will get; read them monthly.
Where KYN fits
KYN builds account scoring inside the sales system it ships, running in your own cloud account. The agent computes fit, intent and timing with rules your team writes, stores the reason and the expiry with each signal, and proposes an action. Outbound messages wait for a named person, and every action is logged with the actor. For an insurance brokerage we built outreach with one checkable first line and a reply desk that sorts each reply.
The first 14 days are a review of your CRM fields and how your team picks who to call. The scoring agent, with rules your team can edit, ships in under 90 days. See sales systems and integrations.
Frequently asked questions
Should I buy a lead scoring tool or build rules?
Start with rules written on paper. If a person cannot explain why an account ranks where it does, a tool will not fix that. Buy or build once the rules are stable and the volume makes hand scoring slow.
How many accounts do I need before scoring is worth it?
Scoring pays off once a team has more accounts than it can review weekly, which for many SMEs is a few hundred. Below that, a salesperson's judgement and a shared sheet work.
Can the agent use LinkedIn for intent signals?
Be careful. LinkedIn's rules restrict automated collection; see LinkedIn data and scraping risk for AI prospecting. Prefer signals from your own site, email and WhatsApp.
Checked 30 October 2026.
Curious whether Marketing agents fits your business? Talk to KYN on WhatsApp — no forms, just a conversation.
