FlowsBook a callSign inCreate free accountSign up
AI employees and automation

Explainable AI lead scores: asking why a lead got 82, and what the answer is made of

Explainable AI lead scores show the facts behind each number. How Ask why answers for one result, what it reads, and how to tell a good reason from a weak one.

Part of the guide AI employees

By the xMap team · · 6 min read

Explainable AI lead scores are scores you can question one at a time and get a plain answer built from facts you can check. In xMap Flows every scored result has an Ask why button: press it on a lead that scored 82 and xMap AI answers in a few sentences from that lead's own fields, the reason the scoring step wrote, the instructions you gave it and the filter rules the lead passed, and it names the facts it used. If an answer leans on something that is not in the lead, the score deserves less trust than its number suggests.

This is for sales managers and reps who are handed a ranked list every morning and want to know whether to believe it, and for whoever writes the scoring instructions. It covers why a bare number is not enough, what Ask why reads, how to judge an answer, a made-up exchange, the habits that keep reasons honest, and how to set the flow up.

Why a number on its own does not get used

A rep who sees 82 next to one lead and 64 next to another has two choices: trust the order blindly or ignore it. Both waste the work. What changes behaviour is a reason the rep can check in ten seconds: opened four months ago, independent, on a high street, no booking on its website. With that, the rep can agree, disagree or spot the case the scoring missed.

Reasons also protect you from a quieter failure. A scoring step that has been told to prefer busy places may quietly reward places with many reviews, even when reviews say nothing about whether they will buy. Reading a few reasons is how you notice before a quarter of calls go to the wrong list.

What Ask why reads before it answers

When you ask about one result, the answer is put together from a fixed set of sources, never from the open web:

  • The result itself. Its short fields: name, type of place, city, rating and reviews, opening date, chain or independent, website, and anything earlier blocks added.
  • The score and the reason. The number the xMap LLM step gave and the one-line reason it wrote at the time.
  • Your instructions. What you told the AI blocks of the flow to look for, and any reference notes.
  • The rules it passed. Each Filter and Branch rule in the flow, and whether this result met it.

It then answers in two to four sentences and lists the facts it relied on. It does not claim more than those sources show, and where xMap AI is not available the flow's own rules give the answer instead. You can pick a ready question, such as what made it a strong fit, or type your own: why is this one below the cafe down the road, or why was this one left out.

Telling a good reason from a weak one

  • It cites fields the lead really has. A reason that mentions a website, a rating or an opening date you can see on the card is sound. One that mentions staff numbers the flow never looked up is a guess.
  • It matches your instructions. If you said independents first and the reason praises a chain branch, the instructions or the data need a look.
  • It separates leads. When every reason says roughly the same thing, the score is not telling leads apart; your instructions are too vague.
  • It survives a direct question. Ask what would have lowered the score. A clear answer shows the scoring is anchored in facts.

A made-up exchange about one lead

Everything in this example is invented. Say a supplier of salon products gets a morning list, and the top lead is a beauty salon scored 82 with the reason: independent salon opened in the spring, strong rating from many reviews, no online shop on its website.

The rep presses Ask why and types: why is this above the salon on the next street, scored 70? The answer says the higher one is independent while the other belongs to a chain listed with four branches, that the instructions ask for independents first, and that both passed the rating rule. The facts used are the chain field, the opening date and the rule. The rep can check every one of them on the two cards, and moves on.

The rep then asks about a lead scored 55 that she knows is a good account. The answer says it has no website and no opening date, so the scoring step had little to go on. That is useful: the lead is not bad, it is thin, and the rep adds a note rather than discarding it.

Habits that keep scores explainable

  • Write instructions with a good and a bad example. A sentence such as independent salons on busy streets are ideal; salons inside hotels are not gives the reasons something concrete to point to.
  • Feed the step the facts you care about. If the size of a business matters, add a block that finds it. The AI cannot explain a score with facts it never received.
  • Ask for a short reason in your own words. One line per lead, naming the deciding facts, reads faster than a paragraph.
  • Check the work before it goes out. A Check the work step can hold back any lead whose reason mentions a figure that is not in the lead.
  • Look at what was dropped. Each run keeps what every block passed on and what a Filter removed, with the rule that removed it, so a missing lead can be traced too.

Setting up a list with reasons attached

  1. Schedule: every weekday at 07:00.
  2. Nearby places or Job postings: the businesses or companies you sell to, in your area.
  3. Only new: nothing your reps have already seen.
  4. xMap LLM, Score and rank: a score from 0 to 100 and a one-line reason, with your best and worst examples in the instructions.
  5. Check the work: every reason cites the lead's own facts.
  6. Filter on fit score, then Email the strong ones to the rep.

Each morning the rep gets the leads, each with its score and reason, and every result in the app has Ask why for the ones that look odd. The template Score new leads from a sheet scores a list you already have, and New places every morning, only new builds one. The guide to AI employees covers how much to let them do alone.

Questions

What makes an AI lead score explainable?

Being able to ask about one lead and get a short answer built from facts you can check: the lead's own fields, the reason written when it was scored, your instructions and the rules it passed.

Does Ask why look anything up on the internet?

No. It answers from the result, the scoring reason, the flow's AI instructions and its filter rules, and names the facts it used.

What if the reason mentions something that is not in the lead?

Treat that score with care. Feed the scoring step the facts it needs, tighten the instructions, or add Check the work to hold such leads back.

Can I ask why a lead was left out?

Yes. Type the question on a result, and the run's step view shows what each Filter removed and the rule that removed it.

Templates that do this

IndustriesB2B sales

Read next

More on ai employees and automation · Every article · Templates

Put it to work

An AI employee in xMap Flows does this on live data, on its own timetable. Start from a template or describe it in plain words.