How to Read Intent Scores

Summary: Every intent row scores one person on one topic. score is the tier and perc_score is the strength behind it. Both measure how much recent research we saw from that person, against cutoffs that are the same for every person and topic. This page explains the two fields, how to filter on them, how to tell who is in market, and why the fields do not convert to an account-level surge score.


The Two Fields

FieldValuesMeaning
scorehigh, medium, lowThe tier. Filter on it with min_score.
perc_score1 to 99The strength of this person's recent research on this topic. Sort on it, or filter with min_perc_score.

Each response also carries as_of, the day of intent data the scores come from. Scores are rebuilt once a day.


How a Score Is Built

  1. A person reads pages across the web.
  2. We match each page to the few topics it is most about. The match uses the meaning of the page, not only the words in it.
  3. Each match adds evidence to that person's score on that topic.
  4. We add up the evidence from the last 14 days. A day's weight halves every 7 days, so recent research counts most.
  5. The total becomes perc_score, and fixed cutoffs on the total set score.

The score is absolute. The same amount of research gives the same score for any person on any topic. It is not a percentile, and it is not compared with other people or with the topic's popularity.

A person who stops researching a topic loses strength each day. With no new activity, they leave the topic within 14 days.


What Each Tier Means

scoreperc_scoreWhat we sawUse it for
high62 and upRepeated or sustained research in the last daysSales outreach, alerts, small precise lists
mediumBelow highClear research, with less depth or less recentAd audiences, nurture, wider outbound
lowLowestLight research, often one or two pagesReach, and a second signal next to your own data

Most rows are low. high and medium together are a small share of the people on a topic, and that is by design: a tier is a cutoff on evidence, not a share of the list.

The tier is set from the evidence before perc_score is rounded down, so a row at exactly 62 can be medium or high. To cut on a tier, filter on score with min_score, not on perc_score.

The tier follows perc_score, with one more exception. A person who researches thousands of topics can show medium next to a perc_score above 62. See Score Details.

perc_score compresses towards the top, so many strong rows share a value such as 99. Repeated values are expected.


How to Filter

Choose the filter from the size of the topic, not from habit. A broad topic has many high people. A narrow business topic can have fewer than 100 people in total, with none at high.

  1. Check the count first. The free single-topic /preview endpoints return count for your filters before you pay for a pull. A response with no count means the total is not known, not that it is zero. Read count as an upper limit, because people who opted out are counted but not returned. See Find People Showing Intent on a Topic.
  2. On a narrow topic, send no min_score. Sort by perc_score and take the strongest rows. A high filter on a narrow topic often returns no one.
  3. On a broad topic, use min_score: "high" or "medium". high gives a small precise list for sales tasks. medium gives more people for ad audiences.
  4. Use min_perc_score to cut inside a tier. For example, 80 keeps the stronger part of high.

A low row is still research on the topic. It is light research, so fewer of those people return to the topic in the next days. Use it with a second signal, such as job title or company fit.

If a narrow topic stays too small, add its related topics or use the broader topic above it. See Closely Related Topics on this page.


Who Is in Market

score answers one question: how much did this person research this topic? It does not classify the person. Two more signals do that, and they come from Predictive Signals.

QuestionSignalDefinition
How much research?score, perc_scoreEvidence from the last 14 days, on this page
Is the person in market?in_marketActive on 3 or more of the last 14 days on at least one of your audience's topics. Days on different topics are not added together
Which way is it moving?intent_trendnew, rising, steady or cooling, week over week

Use them together. score says the research is strong, in_market says it repeats across days, and intent_trend says whether it is growing or ending.

  • For an audience: GET /api/v1/signals/audiences/{id} counts the members in market and lists the top companies and buying groups.
  • For one person: POST /api/v1/signals/people/lookup returns the person's high and medium topics that have a 14-day trend, each with its intent_trend. Your topic_ids and min_score filters apply.

Closely Related Topics

We score each topic on its own. A page counts toward a topic only when the topic is one of the best matches for that page, within one category. A passing mention of a subject does not count.

One page can match a few closely related topics, such as one CRM product and CRM software in general. That page's evidence is shared between those topics. It is not counted in full for each one.

We do not assume that research on one product is research on its whole category. The exception is a short list of broad topics, such as Artificial Intelligence and Cybersecurity. Each of these also counts research on child topics that we checked one by one for accuracy.

So research on one subject can spread across several narrow topics, and each one can show mostly low. If a narrow topic looks thin, ask for its related topics together:

curl -X POST "https://api.delivr.ai/api/v1/intent/people/multi" \
  -H "X-Api-Key: YOUR_API_KEY" \
  -H "X-Api-Secret: YOUR_API_SECRET" \
  -H "Content-Type: application/json" \
  -d '{"topic_ids": ["4eyes_105087", "4eyes_516767"], "min_score": "medium", "min_topics": 1}'

Replace the IDs with your own related topics. min_topics: 1 returns a person on any of the topics. A higher value returns only people on that many of them, which is a strict filter for a list that must be precise. The count covers the strongest people read for each topic: 1,000 by default, set with per_topic_limit. On a large topic, a person below that depth is not counted for it.


Compared With Account-Level Surge Scores

If you have used an account-level intent provider, read this section before you reuse its thresholds. The scores measure different things, so no number on one scale equals a number on the other.

DelivrAccount-level surge score
What is scoredOne person on one topicOne company on one topic
Compared withFixed cutoffs, the same for everyoneThat company's own past activity
A strong score meansWe saw much recent research by this personThe company researches more than it usually does
A middle score meansWe saw clear research by this personThe company's activity is normal for that company
No activityNo rowA score at the low end of the scale
Scalehigh, medium, low and a 1 to 99 strengthOften 0 to 100
TopicsOne topic per row, from pages that are about that topic. You choose which topics to combine.Often summarized across a set of topics, as an average score and a topic count

Bombora is the common example. It says Company Surge uses "relative, account-specific scoring" that compares the most recent three weeks with a 12-week baseline (Bombora). Its report API returns scores from 0 to 100 and by default returns only scores of 60 or more (Bombora API documentation).

The difference shows in four cases:

  • Steady, heavy research. A company that reads about a topic every week has a high baseline, so a relative score shows little change. Delivr scores those people high, because the research is there.
  • A new spike from a quiet company. A small rise over a low baseline is a surge. The people behind it can each be low or medium in Delivr, because each one did little research.
  • A middle tier. On a relative scale, the middle band is a company's normal activity and the low band is less than normal. Delivr has no band for normal. Every Delivr row is research we saw, so read medium as active research, not as a baseline.
  • A filter that worked elsewhere. A minimum of 60 on a surge scale is not high, and it is not a perc_score of 60. Set your Delivr filter from the tier table on this page.

To get a company view from Delivr, use POST /api/v1/intent/companies. It ranks companies on a topic by people_count, the number of people there who show intent.


Next Steps


Did this page help you?