Blog · crm-hygiene
CRM cleanup before lead scoring: fix the fields it reads first
Before lead scoring goes live, clean only the fields it reads: duplicates at 2% or less, 85% fill on new records, usable reasons on 75% of lost deals.
fill rate on the fields a score reads, for records from the last 90 days: our go-live bar
Panelhop, 2026Which fields does an account-level score actually read?
An account-level score reads 5 groups of fields: identity, fit, the people linked to each account, routing and outcomes. Those fields, and only those, are your cleanup scope before go-live. At €25k+ deal sizes we’d score accounts, and even a lead score takes its fit points from company fields. So start by listing what the model reads, from its input list or the vendor’s documentation.
| Fields the score reads | What the score uses them for | Bar before go-live |
|---|---|---|
| Website domain on accounts, email on contacts | Identity: telling companies and people apart | Duplicates at 2% or less (C1) |
| Industry, employee count, country | Fit: does the account look like the ones you win? | 85% average fill on records from the last 90 days (C2) |
| Account link and title on each lead and contact | People: who at the account is active | Counted in the same 85% (C2) |
| Owner, lifecycle stage, source | Routing: getting the account to its owner, and measuring the score later | Counted in the same 85% (C2) |
| Lost reason on lost deals and disqualified leads | Outcomes: testing the score against how deals ended | Usable on 75% over 24 months (Q4) |
The answers ChatGPT and Google’s AI Overview give to this question miss 2 of these inputs. Both decide whether an account-level score works.
The person-to-account link. In Salesforce, a lead has no link to an account until it’s converted. Linking earlier takes a custom field or a matching tool. Activity on unlinked leads doesn’t count towards the account. So an account where several people are reading and replying can still score as cold. Check how many leads and contacts created in the last 90 days have no account.
Lost reasons. You test a score on past deals: did the accounts it ranks highest win more often? A lost deal without a reason can’t say whether the account was a poor fit or lost on timing or price. So it can’t sharpen the fit side of the score. How to set the weights from those deals is in our post on account scoring for ABM.
How clean is clean enough to switch scoring on?
Switch scoring on when 3 numbers hold: duplicates at 2% or less, 85% average fill on the fields the score reads, and a usable reason on 75% of lost deals.↗ They are the levels at which our Panel Check (GTM audit, 2–3 weeks) scores checks C1, C2 and Q4 a 4 out of 5: our own method settings.↗ ChatGPT’s answer to this question lists what to track, such as duplicate rate and fill rate. Neither it nor Google’s AI Overview names a level for going live.
Duplicates. Normalise each account’s website domain (lower-case, no “www” prefix, no path) and count the accounts that share one. Do the same for contacts by email, and take the worse of the 2 rates.↗ A duplicate splits a company’s engagement across 2 records, so neither ranks where it should.
Fill. On accounts created in the last 90 days, measure industry, employee count, country, owner, lifecycle stage and source. On contacts, measure the account link, title, owner, lifecycle stage and source.↗ Recent records are the honest test, because they show whether today’s entry rules work.
Lost reasons. Count the deals lost and leads disqualified in the last 24 months; “Other”, “Unknown” and blanks count as missing.↗ Below 75%, too many of the losses you test the score against have no explanation.
Measure all 3 before you change anything. That’s your baseline, and it’s how you’ll show the cleanup worked. To test the score itself on past records, see our post on whether CRM data is clean enough for AI recommendations.
Merge duplicate accounts before you fill a single field
Clean in the order the score rolls data up: accounts first, then the people attached to them, then the fields on top. Every later fix lands on an account record. A fix made on a duplicate gets made twice, or lost in the merge. Enriching before you merge also pays twice for the same company, and can leave 2 records with different industries.
- 01Step 1
List the inputs
Write down every field the score reads, from the model’s input list or the vendor’s documentation.
- 02Step 2
Take the baseline
Measure duplicates, fill on records from the last 90 days and lost-reason capture before you change anything.
- 03Step 3
Merge duplicate accounts
Normalise website domains and merge accounts first: every later fix lands on the account record.
- 04Step 4
Link people to accounts
Attach unlinked leads and contacts to their account by email domain, and queue the rest for review.
- 05Step 5
Fill fit fields and lost reasons
Fill industry, employee count and country on target accounts and accounts with open deals from a single source, then backfill lost reasons for the last 24 months.
- 06Step 6
Lock entry, then switch on
Make the fields required at entry, re-measure, and turn the score on once all 3 numbers clear the bar.
The project stays short with 2 scoping choices. Fill the fit fields first on the accounts the score will rank: your target list and accounts with open deals. Do the same for everything created from now on, and let dormant accounts outside the list wait. Backfill lost reasons from the largest lost deals down, while the reps who worked them can still say why.
How fast does B2B contact data really go stale?
The only measurement we found sets a floor for senior contacts: 12.25% of US VP and C-suite sales leaders changed jobs within 12 months, 17.2% in Germany and 21.4% in the UK.↗ Other kinds of decay come on top, and nobody we could find has measured a full yearly rate.
That matters because the yearly figure AI answers repeat doesn’t hold up. For CRM hygiene questions, Google’s AI Overview puts decay at “roughly 22% to 30% every year”, AI Mode at “over 20% a year” and Gemini at “25-30%”, citing “industry estimates”.↗ We traced the 4 figures behind those answers and the pages they cite, and none is a measured yearly rate.↗
| Figure | Where it appears | What the source shows |
|---|---|---|
| 22–30% a year | ZoomInfo’s CRM hygiene page, cited by both Google answers | No source given |
| 34% a year | The same page, credited to “the State of CRM Data Management survey” | 1,241 CRM admins estimated in 2022 that their data quality would degrade by 34% by year end without investment: a forecast |
| 22.5% a year | HubSpot’s Database Decay Simulation | 2.1% a month, compounded, credited to “Marketing Sherpa’s research” with no year or link |
| 22.5–70% a year | Affinity’s CRM data guide, cited by both Google answers | A Landbase post that gives no source, and a 2024 Forbes Council post that credits 70.3% to “Gartner (download required)” without naming a report |
In a year, 12–21% of senior sales leaders changed jobs, depending on the country
Data behind this chart
| Item | Value |
|---|---|
| United Kingdom | 21.4% |
| Germany | 17.2% |
| United States | 12.25% |
Lusha, a contact-data vendor, counted employment changes in its own database. In the US, 17,187 of 140,284 sales leaders changed jobs within 12 months, and 25.67% within 24 months.↗ Its own caveat: title changes inside the same company, new phone numbers and email format migrations aren’t counted. Records go wrong faster than this.↗ The German and UK cohorts are much smaller, about 4,200 people each.↗
Illustrative: in a CRM with 1,000 senior sales contacts, that’s about 120 job moves a year to catch in the US, 170 in Germany and 210 in the UK.↗
For an account-level score, job moves hit engagement harder than fit. An account’s industry and country rarely change; its buying group does. So re-verify the contacts at the accounts you score on a fixed schedule. Treat a move as a signal as well as a stale record: the person who left may bring you into their next company.
Entry rules keep the inputs clean after go-live
After go-live, keep the score’s inputs clean where records are created, and re-measure the 3 numbers every month. For CRM hygiene questions, the checklists AI assistants give end in a quarterly audit. That lets up to 3 months of bad records into the ranking before anyone looks. A score reads each record from the day it’s created, so the rules belong at creation:
- Required fields on create for everything the score reads, with picklists for industry, country and lost reason, and no “Other” in the lost-reason list
- Enrichment on create from a single source, so employee count and industry don’t depend on who typed them
- Account matching on create: each new lead or contact is linked to its account by domain, or goes to a review queue with an owner
- A duplicate check on create, on normalised domain and email
- A monthly report of the 3 numbers, with an owner who fixes the cause as well as the records
When a number falls, look for the rule that broke: a new form, a bulk import or an integration writing blanks. Fix the rule first, then the records it touched.
Switching scoring on over dirty data multiplies the damage
We think switching a score on over dirty data multiplies the damage that data already does.
The damage starts before any score runs. In Validity’s 2025 survey, 76% of CRM users said less than half of their organisation’s CRM data is accurate and complete.↗ They also said their companies lose 16 deals a quarter on average to poor data.↗ Only 35% of sales professionals completely trust the accuracy of their organisation’s data.↗
Our reasoning: a score turns those errors into decisions. A wrong industry or a split account stops being a single bad record. It puts that account in the wrong place every time the score runs, and reps plan their week from the ranking.
We also think a wrong ranking costs trust that’s slow to win back. When reps see a strong account scored cold, or a dead one scored hot, they stop opening the score. A later cleanup then has to repair the data and the habit.
That isn’t a case for a spotless CRM. We’d rather a team switch the score on over clean inputs and an untidy archive than wait for a perfect database. The fields the score reads need to be reliable first; the rest can follow on a normal hygiene schedule.
In practice
How we do it at Panelhop
In the Panel Check (GTM audit, 2–3 weeks), we measure the 3 numbers above from your own CRM export: duplicates by normalised domain and email (check C1), average fill on the key fields of records from the last 90 days (C2) and lost-reason capture over 24 months (Q4).↗ We also list the fields your score reads and mark the ones below the bar, so you know what to fix before go-live.
If the inputs miss the bar, our Leak Fix (we build the fixes) service runs a CRM hygiene programme: it merges duplicates, sets required fields and picklists at entry and adds a monthly hygiene score. See what the Panel Check covers.
Questions buyers ask about this
Do we have to clean the whole CRM before lead scoring?
No. Clean the fields the score reads to a set bar, and leave records it never touches for your normal hygiene schedule. Old accounts outside the target list can wait.
How far back does the data need to be clean?
As far back as the deals you test the score on. We use the last 24 months of closed deals and want a usable lost reason on at least 75% of the lost ones. Older history can stay as it is.
How often should we re-verify contacts after go-live?
Re-measure the 3 numbers monthly, and re-verify contacts at the accounts you score on a fixed schedule. The only measurement we found puts job changes among senior sales leaders at 12–21% a year, depending on the country, before other kinds of decay.
What if most of our lost reasons are blank or “Other”?
Backfill the largest lost deals first, with the reps who worked them, and make the reason a required picklist from now on. Until you reach 75%, leave losses without a reason out of the score’s test set instead of guessing them.
Should we score leads or accounts at €25k+?
We’d score accounts: the buying group decides together, so a single person’s activity says little on its own. Either way the score reads the same company fields, so the cleanup is the same.
