AI lead qualification: what it can read in an enquiry and what it cannot

Forty messages arrived overnight on WhatsApp, the website form and a trade portal, and you have one hour before the shop opens. You need to know which five to ring first. An AI tool can read those forty and sort them by what is written in them: product, quantity, place, timing, a drawing attached. It cannot read what is not written: whether the buyer has the money, whether they decide alone, whether they are collecting prices for a boss. Let it decide who gets called first. Keep the decision to quote, visit or walk away with a person.

Five enquiries, read the way a tool reads them

The quickest way to see the limit is to look at real-shaped messages. Imagine, as an invented example, a furniture maker in Jodhpur who sells mango-wood and iron pieces to shops and to overseas buyers. Five messages arrive one morning.

"Need 40 dining sets, container, Dubai showroom, sample by Diwali, send CIF." Rich in signals: quantity, destination, a date, a trade term. A tool ranks this at the top, and it is right to. A person still has to check one thing the message cannot say, whether this buyer has imported before.

"Price list pls." Four words. A tool puts it at the bottom, as it should for a stranger. But the number it came from is a repeat buyer in Bengaluru who writes this way every time. The tool cannot know that unless your records say so.

A long, careful message from a design firm, with a mood board attached. Reads as high effort, so it ranks high. In fact the sender wrote the same note to nine workshops. Effort in the message is not intent to buy.

"Reference: Mr Shekhawat of Jaipur told me to contact you." A tool sees a name. Only you know whether Mr Shekhawat is your best customer or someone who met you once at a fair.

"Quote best price for 500 chairs urgently." A flag word and a quantity. The owner knows this sender writes "urgently" on everything, so the tool is misled by a habit it cannot see.

Look across the five. The tool reads signals well and intent badly, and the signals that matter most, budget and who decides, often do not appear in a first message.

What should it do, then?

Give it the jobs that need patience, not judgement.

• Pull product, size, quantity, city and date out of a rambling message and put them in your fields.
• Spot the job-seeker, the vendor pitching you and the duplicate of yesterday's enquiry.
• Apply your written rule at midnight and on Sunday. If your rule is "a container quantity, to a port we ship to, with a date inside three months, goes to the top", it applies it without drifting.
• Label each enquiry in plain words: call today, call this week, reply by message only. Nobody acts on a score of 63.

Keep a column beside the label called "my call", where the salesperson writes what they decided. It is your best training material.

How do I write rules that deserve to be used?

Do not accept a vendor's default. Build them from your own history.

1. Pull your last fifty to a hundred enquiries, each marked order, lost, no reply or never qualified.
2. Look for the clues that separated orders from the rest. Perhaps buyers who named a destination converted more than those who asked for a catalogue.
3. Keep the three or four clues that held up. Those are the rules.
4. Run them on the same hundred and see how many orders landed in the "call today" group.

If step one is impossible because outcomes were never recorded, your first project is the record, not the tool. A scorer fed on untidy outcomes learns your confusion.

What went wrong in the first week?

Back in Jodhpur, the owner's first rules favoured long messages with attachments, because past orders had often come with a drawing. Within days the top group filled with design firms collecting prices. The shortest messages slid to the bottom. One was a line from a Muscat hotel group: "Need 60 beds Jan, Jodhpur ex-works?" A salesperson spotted it in the "my call" column and rang, but a day late. The group had already had a quote from a workshop in Saharanpur, and the order that came was for 20 beds, not 60.

The owner rewrote the rule. A named destination or a named month now lifts an enquiry whatever its length, and attachments no longer count by themselves. The low group is also read by a person for ten minutes each morning, since that is where the short, serious messages hide.

Where must a person decide?

On whether to quote at all, because declining a buyer who may grow is a business call. On price, credit and advance. On sensitive cases: a complaint, a repeat buyer with history, a referral from your best customer. And on overrides: any salesperson may ignore a low label and say why.

When is it not worth the trouble?

If you get a handful of enquiries a day, reading them yourself takes minutes and teaches you more. Scoring pays when the pile is too big to read with care, or when several people share it and rank it differently.

Do not start before the enquiries have one home. If they sit in three phones and a notebook, no tool can score what it cannot see. And never score on a name, a language or a spelling. Those are poor guides to who will buy, and they shut out good customers. Stay with business signals.

Run it in parallel before it runs your day

For a few cycles, let the tool label while the team works as before, then compare. Count the orders that came from the group the tool called low. If the answer is "some", change the rules before you trust them.

Before any tool, put your last fifty enquiries in one sheet and mark each as order, lost or no reply. Then write down the questions a person asks on the first call; a qualifying framework for a small team is a place to start. AI lead qualification and scoring is the service for the sorting itself, the AI for sales and marketing hub shows where else AI fits, and if you want to talk it over, tell us what you need.