How a proposed lead score is tested
The proposed points are built from half of your records and measured on the other half, so the figure shown is what the scoring did on records it had never seen. The scores you already have are read first and, where one of them works, it becomes an input rather than being thrown away. For how a score you already have is judged, read the Is your scoring working? card on the same tab before the proposal underneath it.
Before you start
Choose the outcome at the top of the tab before you read any figure. The same records answer three different questions, and each has its own population.
| Outcome | The question it answers | Who it is measured among |
|---|---|---|
| Became a customer | Does a high score mean somebody becomes a customer? | Everybody read. A person counts when the lifecycle stage says Customer, or they sit on a won deal |
| Is on a sales deal | Does a high score mean sales opens a deal with them? | Everybody read. Counted in people, not deals — one deal can carry several people |
| Is on a won deal | Among people whose deal has closed, does a high score mean it was won? | Only people whose deal has closed. Open deals are left out because nobody knows yet how they end |
Note
An outcome is offered only when at least 30 people reached it and at least 30 did not. Below either floor there is nothing to learn from, and the tab says which side is short rather than showing a rate.
Half to build, half to test
MarqIQ splits your records in two. The points are derived from one half. The score is then worked out for every record in the other half, and the conversion of each band of that score is what the card reports.
This is the whole reason the number can be quoted. A scoring built from some records and then tested on the same records always looks good, because it has memorised them. The card names the size of the untouched half directly — “tested it on the other 4,180 it had never seen”.
The simulation appears only when both of these hold: at least two patterns earned points, and the testing half holds at least 60 records. Below either, no proposal is shown rather than a proposal nobody could check.
Which patterns earn points
Every value of every attribute is counted against the outcome on the building half. A value earns points only when both of these hold:
- At least 30 records in the building half carry that value.
- Its conversion is reliably above or below your average — the cautious end of its own 95% range still clears the average, in whichever direction.
Anything thinner is left out rather than guessed at, and the count of patterns that were real but too thin is printed beside the table. The points themselves are coarse on purpose, because a rubric of fives and tens gets typed into a CRM and a rubric of 7.3s does not:
| How much better or worse, at the cautious end | Points |
|---|---|
| Three times the average or more | 20 |
| Twice the average or more | 15 |
| Half again the average or more | 10 |
| A fifth above the average or more | 5 |
| Four-tenths of the average or less | −10 |
| Seven-tenths of the average or less | −5 |
| Anything in between | 0, so the pattern is listed and carries no weight |
At most 40 values of any one attribute are considered, and at most 24 scored lines are shown. Each line prints the conversion it rests on and how that compares with the average.
A score of yours that works becomes an input
Before proposing anything, MarqIQ judges every score you already have. A score rated Working or Partly working is then fed into the proposal, grouped into quarters of its own range, so the proposal reads as “your score, plus what it misses” rather than a replacement that throws away what works.
- The quarter boundaries are taken from the building half only, so the testing half is untouched by them.
- A score needs at least 120 numeric values in the building half to be used this way, and its lowest and highest cut points must differ. A score that holds nearly the same number on everybody cannot group anybody.
- Lines that came from one of your own scores are marked as such in the table, so you can see how much of the proposal is already yours.
What the headline number means
The proposed score is split into four bands of roughly equal size across the testing half, and the card quotes the top band's conversion. Two readings are possible, and the card says which one you are getting:
- Where you already have a working score, the comparison is against the top band of your best one, and the proposal is described as better only when it converts more than 15% above it. Otherwise the card says your current scoring is already capturing most of what the data can show.
- Where you have no working score, the comparison is against your overall average conversion.
- Where the proposed score did not separate buyers on the records it had not seen, the card says exactly that: the patterns are real but not yet strong enough to score on.
A band whose conversion rests on fewer than 30 records is not quoted as the top band; the next one down that clears the floor is used instead.
Take care
The top band's conversion is not a promise about next quarter. It is what this scoring did on records from your own history that it had never seen. Treat it as the reason to run the points beside your existing score for a few weeks, which is what the advice under the card tells you to do — in an empty scoring field you already have, where one exists.
How a score you already have is judged
Each score is split into four bands of roughly equal size, lowest to highest, and the conversion of each band is compared.
| Verdict | What it takes |
|---|---|
| Working | The top band's cautious end still clears the bottom band's generous end, and the bands climb in order |
| Partly working | The top and bottom separate, but the bands do not climb in order — the middle of the ranking is not reliable |
| Ranking backwards | The bottom band clears the top: the lowest-scored people convert more often |
| Not telling you anything | Even the most generous reading cannot reach twice the conversion from bottom band to top |
| Too early to tell | Fewer than 30 records in a band, or the intervals still allow a working score. No claim either way |
| Mostly empty | Filled in on under a quarter of records |
| Not really a score | Almost the same value on every record, or only two values. A flag, not a ranking |
| Never filled in | The field exists and holds nothing on any record |
Fields are found by name — score, grade, rating, fit, propensity, likelihood, priority and similar — read from the tool's full field list, and then judged on their values rather than their names. The method note prints how many were read, how many behave like a score, and how many were set aside. Two kinds are always set aside and named: a field whose values are text, dates or identifiers, and a satisfaction, survey or social rating, which says nothing about who will buy.
When nothing is claimed
- Nobody is marked as a customer and no won deal is linked to a person: there is no outcome to learn from, and the tab says so and tells you what to set in the tool.
- Fewer than 30 people reached the outcome: the tab names the count and says scoring can be judged fairly once there are 30.
- A scoring field exists and holds nothing: that is reported as a finding of its own. Somebody set it up, or the tool ships it, and it was never switched on.
The first line of How we worked this out names how many people were read and how many of them reached the outcome, and the second says in words how the outcome was decided for your workspace — down to the counts behind each half of it.
If it did not work
The proposal does not appear at all
Why: Fewer than two patterns earned points, or the testing half holds fewer than 60 records.
Fix: Read the count of patterns that were too thin, printed beside the table. Widen the Created range so more records are read, and come back.
The proposal is no better than your current score
Why: The card compares against the top band of your best existing score and claims an improvement only above 15%.
Fix: Nothing to fix. The card is telling you your scoring already captures most of what your data can show, which is the answer you wanted.
A score you rely on reads Too early to tell
Why: A band holds fewer than 30 records, or the ranges still allow a working score, so no verdict is claimed either way.
Fix: Widen the period. The verdict withheld is the claim, not the score.
A field you consider a score is missing from the table
Why: Either its name does not read as a score, or its values were set aside.
Fix: Check the set-aside list under the table: it names every field read and the reason each was set aside. A field with no score-like word in its name is not read at all.
The figures moved after a sync
Why: New records changed the split, the outcomes and therefore the points.
Fix: Compare the population line at the top of How we worked this out between the two readings before comparing the rates.
Next steps
Related articles
Was this article helpful?