How AI lead scoring works, the data Dynamics, Salesforce, HubSpot, Zoho and Meta need before it pays off, and a scoring worksheet for Kerala and Gulf teams.
- Published
- Updated
- Reading time
- 13 minutes
Drafted with AI tools; researched, edited and approved by Althameem Ali. Our editorial policy
On this page · 10 sections
- What is lead scoring, and what does AI add?
- Rules-based or predictive: which fits your lead volume?
- Which signals predict a sale for Kerala and Gulf leads?
- How do you build a first scoring model in a spreadsheet?
- How do you check the score is right?
- How do scores reach Meta and Google bidding?
- What can go wrong with lead scoring?
- Frequently asked questions
- What to do next
- Changelog
Full disclosure: we build CRMs with routing and follow-up rules on Xale (opens in a new tab), the CRM platform by Xale Private Limited (reported by our team), so we have a stake in this. Our position is still the cautious one: start with rules, and switch to a predictive model only when your volume clears the vendors' own minimums.
This guide explains what scoring is, how much data each AI option needs, which signals matter for Kerala and Gulf leads, how to build and test a first model in a spreadsheet, and how scores can feed your Meta and Google bidding.
What is lead scoring, and what does AI add?#
A lead score is a number that ranks enquiries by how likely they are to buy. It adds up two kinds of evidence:
- Fit: does this person look like your past buyers? Budget band, location, the project or course they asked about, whether a phone number works.
- Intent: what have they done? Replied on WhatsApp, asked for a site visit, opened the brochure, come back a second time.
A rules-based score is points you assign by hand: +20 for a referral, +15 for asking about a site visit, −30 for an invalid number. It is transparent, cheap and quick to change.
A predictive or AI score is learned from history. The software looks at leads you won and lost and works out which signals separated them. Odoo, for example, documents its method as a naive Bayes model on won and lost opportunities (opens in a new tab), always using the pipeline stage and sales team, with optional signals such as country, phone and email quality, source, language and tags (Odoo 20.0 documentation, read 4 October 2026). HubSpot's version predicts the probability that a contact closes as a customer within 90 days (opens in a new tab) (updated 11 January 2026).
What AI adds is the ability to find patterns you would not have guessed, and to keep re-learning as your market changes. What it needs in return is volume and clean outcomes. Our guide to AI in CRM puts scoring alongside the other AI features worth considering.
Rules-based or predictive: which fits your lead volume?#
Most vendors publish the data their model needs before it will train, and a small team can sit below several of them. Here is what each asks for, read on 4 October 2026:
Sources: Microsoft Learn, Configure predictive lead scoring (opens in a new tab) (7 August 2025); Salesforce Help, Considerations for Setting Up Einstein Lead Scoring (opens in a new tab); the HubSpot and Odoo pages above; Zoho's edition comparison (opens in a new tab); Meta's Conversions API for CRM integration (opens in a new tab) (updated 17 December 2025).
Read the minimums against your own numbers. A Kozhikode builder getting 300 leads a month creates about 2,000 in 200 days, comfortably past Salesforce's 1,000. But 120 of those must be converted (in Salesforce, turned into an account and contact), which needs about 6% of them to reach that stage; at 2%, the builder stays on Salesforce's global model until its volume roughly triples. Microsoft's 40 and 40 counts qualified and disqualified leads, not sales, which is easier to reach if your team marks leads honestly.
Our rule of thumb: if every new lead already gets a call on the day it arrives, a score has no queue to sort. Xale makes the same argument for admissions teams in its post on lead scoring for admissions (opens in a new tab) (read 4 October 2026): status, source, intake and each role's follow-up date usually rank students better than points. Scoring earns its place when leads outnumber the calls your team can make in a day.
Which signals predict a sale for Kerala and Gulf leads?#
Many scoring guides are written for email-led B2B sales, where opens and webinar sign-ups count. When most conversations happen on WhatsApp and the phone, these are the signals worth recording:
- Source. A referral, a walk-in or a Google search usually behaves differently from a Meta instant form filled in by accident. Store the campaign and form on every lead the moment it arrives; Xale, for instance, saves the campaign and form tags on each Meta lead (opens in a new tab) and syncs ad spend daily, so cost per lead stays current. Our post on building a CRM that scales explains why the source can't be rebuilt later, and our Xale case study shows the CRM the team behind Marketlube built.
- Reply speed, theirs and yours. A buyer who answers your first WhatsApp message within an hour is a different lead from one who never replies. And leads your team answered slowly will look "cold" in the data even if they weren't.
- Form answers. Budget band, timeline, preferred location, course or intake. Two or three well-chosen questions on the form often tell you more than any behavioural signal.
- WhatsApp replies. A two-way conversation, a request for a floor plan or a fee structure, a voice note: intent you can log.
- Location. For Kerala property, an enquirer in the same district or a Malayali in the Gulf is a different buyer from someone in another state. For admissions, the student's home district matters for walk-ins.
- Budget band against what you sell. A ₹45 lakh budget for an ₹80 lakh villa is not a hot lead, however many times they message.
Gulf leads need two adjustments: score on the lead's local time, so a reply at 11 pm IST from Dubai counts as prompt, and don't penalise NRIs for not booking a site visit they can't attend until their next trip home.
How do you build a first scoring model in a spreadsheet?#
Start with points you can explain to the team. Here is a worksheet we would use as a starting point for a Kerala real estate sales team; the points are a template to calibrate against your own deals, not a benchmark.
For an admissions team, swap in course fit (marks meet the entry requirement: +20), the next intake (+15), a parent involved in the conversation (+10), documents ready (+15) and the same source and reply signals.
Then set bands: for example 60 and above is hot (call within the hour), 30 to 59 is warm (call today), below 30 goes into a WhatsApp nurture sequence. Build it in a sheet first, with one column per signal and a total, before you touch your CRM. A score your team can recalculate by hand is a score they will trust.
How do you check the score is right?#
Test it against deals that have already closed, before it decides anything:
- Take the last 100 to 200 closed leads, won and lost, and score them with the worksheet as they looked when they arrived.
- Sort by score and split into bands. If the score works, most of the won deals sit in the top band. If wins are spread evenly, the points are wrong.
- Compare conversion rates by band. Hot leads should convert several times better than cold ones. If they don't, change the weights and test again.
- Re-check monthly. Markets drift. Onam, Gulf school holidays and admission seasons all change who enquires and why. Xale's admissions post makes the point that intake calendars can make a quiet enquiry the urgent one.
Predictive tools do a version of this for you. Microsoft's, for example, flags a model as not ready to publish when its accuracy (its area under curve score) falls below a threshold, shows the top factors behind each score, and can be set to retrain every 15 days. But you should still run the back-test yourself: a model trained on a year of mis-marked leads learns the mistakes.
How do scores reach Meta and Google bidding?#
A score becomes far more valuable when your ad platforms see it, because they can then buy more leads that look like your hot ones. This is the closed loop our performance marketing team sets up, and it works in two ways:
- Meta. Conversion Leads optimisation takes your CRM stages back through the Conversions API, so campaigns optimise for the stage you choose, such as "qualified" or "site visit booked", instead of raw form fills. Meta's guidance asks for 200+ leads a month, uploads at least daily, a target stage reached within 28 days of the lead at a conversion rate between 1% and 40%, and the 15 to 17 digit Meta lead ID stored in your CRM.
- Google. For leads qualified or closed offline, Google recommends enhanced conversions for leads, sending the values from your CRM (opens in a new tab). For value-based bidding you must report two or more different values, which "can be real economic values, like revenue or proxy values, like a lead score", according to its value-based bidding guide (opens in a new tab) (both read 4 October 2026).
The practical version: send one stage that happens often enough (qualified, or site visit booked), give stages different values, and upload daily. The full set-up, from click IDs to upload windows and Google's June 2026 move to the Data Manager API, is in our guide to offline conversion tracking. To judge the result, compare cost per qualified lead rather than cost per lead; our cost per lead benchmarks explain why.
What can go wrong with lead scoring?#
- Bias you didn't intend. A model that learns "leads from this district rarely buy" will deprioritise everyone from that district, including the ones who would have. Review which signals drive the score, and keep sensitive traits out of it entirely.
- Stale models. A score built on last year's launch will misjudge this year's buyers. Retrain or re-weight on a schedule, and after any big change in product, price or campaign mix.
- Ignoring low scores. Low means "later", not "never". Put low-scoring leads into an automated nurture sequence and let a reply bring them back up the list.
- Garbage outcomes. If salespeople close leads as "lost" to clear their queue, the model learns that good leads are bad. Make the loss reason a required choice from a short list.
- Purpose and accuracy under DPDP. India's Digital Personal Data Protection Act (opens in a new tab) requires a business to ensure the "completeness, accuracy and consistency" of personal data likely to be used to make a decision that affects the person, or shared with another business (section 8(3)). From May 2027 the Rules also require consent notices that spell out each purpose. Make sure your notice covers using enquiry data to prioritise follow-up and, if you send it to ad platforms, to improve advertising.
Frequently asked questions#
How many leads do you need for predictive lead scoring?#
It depends on the vendor. Microsoft Dynamics 365 needs at least 40 qualified and 40 disqualified leads created and closed in the training window. Salesforce Einstein needs 1,000 leads created in the last 200 days with 120 converted, and uses a global model until then. Meta's Conversion Leads optimisation asks for 200 or more leads a month. Below these numbers, a rules-based score is usually more reliable.
Does Zoho CRM have lead scoring?#
Yes. Zoho's edition comparison lists scoring rules from the Standard edition, covering 10 fields a module and up to 50 in Ultimate, while Zia's prediction features, including the Prediction Builder, are in the Enterprise and Ultimate editions. Check the current comparison page before buying, as editions change.
What is a good lead score?#
There is no universal good score, because points are relative to your own model. What matters is that leads in your top band convert clearly better than those in your bottom band. Set bands by testing the score on your own closed deals.
Is lead scoring useful for real estate in Kerala?#
It helps when enquiries outnumber the calls your team can make in a day, which happens around launches and big campaigns. The most predictive signals are usually the budget band against the project price, the timeline, the source and whether the buyer asked for a site visit or price list.
What to do next#
Export last quarter's closed leads with their source, budget, timeline and outcome, and score them with the worksheet above. If the top band holds most of your wins, put the score into your CRM and routing; if not, adjust the weights before anyone relies on it. In Xale, routing is an automation: Auto-distribute hands each new lead to the least-loaded person in a role or branch, as xale.in describes it (opens in a new tab). If you want a second pair of eyes on the numbers, share the export with us and we will tell you whether a score, a better pipeline or simply faster first replies would help you most.
Changelog#
- 2026-10-10: The source signal now links our Xale case study, which shows the CRM the team behind Marketlube built.
Related services
Industries
Written by
Co-founder, Marketlube
Althameem Ali researched, edited and approved this post. Marketlube is a growth partner in Calicut building brands, websites, CRMs and campaigns for businesses across Kerala, India and the Gulf.
