> ## Documentation Index
> Fetch the complete documentation index at: https://docs.leadinfo.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Scoring

> The unified score and the pillars behind it: fit from the ICP Builder, intent from the Intent Builder, engagement from company activity and people from what contacts did. Every number comes with the rules and events behind it.

A single opaque number is not enough to decide whether to act on a company. Leadinfo therefore scores four things separately and combines them into one `score` you can take apart again.

## The four pillars

| Pillar | Default weight | Value (0 to 100) |
| - | - | - |
| **Fit** | 35 | 100 when the company matches every rule of at least one active ICP Builder profile, 0 otherwise. Without an active ICP profile, fit drops out of the score. |
| **Intent** | 35 | The Intent Builder page rules matched by pages the company viewed in the 90 days to `end_date`, in its best-matching active profile: one rule gives about 50, two about 75, three about 85. In stored scores each match fades with a 30-day half-life. |
| **Engagement** | 15 | Company activity: website sessions and extra pages over 90 days, each halving in weight every 14 days. |
| **People** | 15 | What the company's people did: identified visitors, form submissions and Leadbot conversations (stored scores add email and LinkedIn activity), each fading with age. A person matching the persona counts in full, anyone else at half; stored scores keep the company's best persona. |

`score` is the weighted sum of the pillar values. A pillar without a profile drops out and the remaining weights are scaled back to 100; engagement and people always count.

Every signal counts in exactly one pillar. Anonymous company activity goes to engagement; anything a known or identified person did goes to people. A form submission or an email reply therefore raises the people pillar, not engagement.

`meta.scoring` on the companies endpoints lists the effective weights and the active profiles (`icp_profiles`, `intent_profiles`) the scores were computed with.

### Several active profiles

You can have up to five active ICP profiles and five active Intent profiles, for example one per segment you sell to. Each company is scored against every active profile of a kind and keeps its best match, so a small agency is compared with your agency profile and an enterprise with your enterprise profile. Profiles are never mixed: a company's intent comes from one profile's rules, not from rules added up across profiles. The breakdown names the profile that won (`profile`). Adding a profile can raise a company's pillar value, never lower it. Keep only buying-related Intent profiles active: whichever profile scores highest counts as the company's intent.

### ICP: all rules or nothing

A company matches an ICP only when it matches every rule of it; a partial match is no match, and an unknown value (for example no known company size) does not match a rule. Fit is therefore 100 or 0. Without a match, the breakdown shows the closest ICP and the rules it missed. Keep each ICP to the rules that define that segment, and split distinct segments into separate ICPs.

Without an ICP match a company cannot reach the hot tier: fit is 35 of the 100 points.

### Intent: each matched page counts

Intent grows with each Intent rule the company matched, however many rules the profile holds: one fresh pricing view gives about 50, pricing and demo about 75. Every rule counts the same for now; per-rule weights for Intent will become editable in the Portal later.

### Scores per profile

Stored scores carry `profile_scores` on each company row: the fit value for every active ICP, the intent value for every active Intent profile and the people value for every persona. `fit`, `intent` and `people` hold the best of each. To see how a company scores against one particular profile, swap that profile's value in and recombine with the weights in `score_breakdown`.

```json theme={null}
"profile_scores": {
  "icp": [
    { "id": 1, "name": "Core ICP", "value": 0 },
    { "id": 34, "name": "Agencies", "value": 100 }
  ],
  "intent": [{ "id": 1, "name": "Buying signals", "value": 74 }],
  "persona": [{ "id": 12, "name": "Marketing leads", "value": 42 }]
}
```

### The people pillar

| Evidence | Counts |
| - | - |
| A person matching the persona acted (form, Leadbot, email reply or click, LinkedIn reply or connection) | In full, fading with each signal's half-life |
| A person matching the persona was identified on the website | In full, fading |
| Someone else, or an unknown person, did any of the above | At half weight, fading |
| The contact database holds people matching the persona at the company (stored scores) | At least 30, without fading |

Stored scores compute this for every persona and keep the best one; without personas everyone counts in full.

Opt-outs and unsubscribes lower the pillar. Events old enough to be worth less than a point still add their points but no longer appear as evidence.

## Reading the breakdown

Every company row carries `score_breakdown`, one entry per pillar:

```json theme={null}
{
  "pillar": "intent",
  "source": "intent_builder",
  "weight": 41,
  "value": 64,
  "points": 26.2,
  "matched": ["Pricing", "Integrations"],
  "unmatched": ["Demo or trial", "Customer cases"]
}
```

`weight` is the effective share in percent, `points` is what the pillar added to the score, and `matched` and `unmatched` are the labels of the rules the company did and did not satisfy. `profile` names the profile the company matched best (for fit without a match, the closest ICP; for people, the best persona). Keep the questions apart when you act on it: fit answers "is this an account we want", intent and engagement answer "are they interested now", people answers "are the right people involved".

## Explaining one company

`GET /v1/trackers/{tracker_code}/companies/{company_id}/score` returns the row, the pillars, the pages viewed in the window, and two lists:

* `known`: observed facts such as sessions and pages, identified visitors, form submissions, CRM status, tags, and warnings such as a shared-network traffic pattern.
* `unknown`: what the data cannot tell, such as nobody identified, no CRM record or incomplete firmographics.

On stored scoring the response also carries `events`: each signal with the points it adds today and the pillar it counts in (`engagement` or `people`), with the matched persona for people events.

The window defaults to the 90 days up to `end_date`.

## Score tiers

Every company row carries `score_tier`, so you can act on bands instead of exact numbers:

| Tier | Score |
| - | - |
| `hot` | 70 and up |
| `warm` | 50 to 69 |
| `cool` | 30 to 49 |
| `cold` | below 30 |

## What changed

Accounts on stored scoring, which is rolling out, keep a history of every meaningful score change. Two endpoints turn it into signals:

* `GET /v1/trackers/{tracker_code}/score-changes` lists the companies whose score moved in the last `days` (1 to 30): who rose (`direction=up`), who fell (`direction=down`) and who moved up into a tier (`became=hot` or `became=warm`). Each company comes with the score and tier before and after, the causes merged per pillar with the points each added or removed, and a plain `summary` such as "Quillis rose from 58 to 72 and is now hot. Viewed Pricing (NL)". `relationship` works as on the companies endpoints and defaults to `prospects`. `tag_ids` keeps only companies carrying any of those tags and `exclude_tag_ids` drops companies carrying any of them, for example your own Partner or Login tags. Both filters apply before `limit`; `meta.has_more` says when more companies matched.
* `GET /v1/trackers/{tracker_code}/companies/{company_id}/score/history` shows one company's changes over time, newest first, with the same summaries and the events that arrived.

Changes caused by the company itself count: new activity, activity fading, updated company data or new persona matches in the contact database. Changes caused by your own edits to profiles, personas, signal settings or tracker roles, and changes from a new version of the scoring method, are left out of the movers feed unless you pass `include_setting_changes=true`.

Intent fades gradually: a buying page viewed today counts in full, one viewed 30 days ago counts half, and after 90 days it no longer counts. The breakdown says when each matched page was last viewed.

## Tracker roles

When an account has more than one tracker, each tracker has a role that decides how its traffic counts:

| Role | Counts toward the score |
| - | - |
| `website` | Fully. Only website pages feed the intent pillar. |
| `product` | Not at all. Visits to your own app or portal are product usage, not buying interest. |
| `careers` | Not at all. Visitors to a jobs site are applicants. |
| `content` | At half weight, without feeding intent. For an academy, docs or help center. |

The role is suggested from the domain (`app.`, `portal.` and `my.` are product, `jobs.` and `careers.` are careers, `academy.`, `docs.` and `help.` are content) until you set one. `GET /v1/trackers` shows each tracker's `role` and `role_source`. Set a different role in the Portal. Every company is rescored at the next scoring run.

## The other scores

* `engagement_score` and `engagement_tier` (cold below 30, warm, hot from 71) describe recency-weighted company activity (sessions and pages) over 90 days.
* `people_score` (0 to 100) is the people pillar on its own.
* `lead_score` (0 to 100, all time) is the Portal's behavioural score from visit days, sessions and pages.

## The scoring model endpoint

`GET /v1/scoring-model` describes the method for your account: the configured weights, the pillars with their effective weights and descriptions, the active ICP and Intent profiles (`icp_profiles`, `intent_profiles`), the engagement model and a reading guide. Use it to explain scores to users or to check which profiles are active before you tune them.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.