Skip to content

20sources

CRM data cleansing: stale contact records, data enrichment and the field it overwrites

One invented insurance office runs a data enrichment pass, and the blanks get filled, and so does a field that was already right. How to clean stale contact records in a CRM in an order that protects what you know. What the FTC found when it ordered nine data sellers to explain themselves. Why no honest decay rate exists, and how to measure your own. And the two columns that make every later step possible.

Excelsior Mind39 min readUpdated

CRM data cleansing is the work of finding the records in your CRM that are no longer true and fixing them without overwriting what your customers told you themselves. We build that clean-up for small businesses, and this page gives you the method free: the order to do it in, the questions to put to any data supplier, and the two columns that make every later step possible: where each value came from, and the date it was written. Below, one invented insurance office runs a data enrichment pass, the step a clean-up is tempted to start with, and finds that the blanks got filled and so did a field that was already right. By the end you will know how to clean stale contact records in the right order, what an appended value really tells you, what to decide before any outside data touches your records, and how to check what your last pass did.

Marta's insurance office (invented). Marta owns an independent insurance office with four people: herself, Paul at the front desk and two account managers. Over twelve years its CRM has collected about four thousand records: policyholders, people who asked for a quote and never bought, and people who bought once and moved on. Marta, Paul, their customers and every figure about them are made up; their records run through this page as one worked example of a real method.

Marta's insurance office (invented). Last spring Marta ran a data enrichment pass, because two thirds of the records had no phone number in them. That was a real problem, and enrichment was a real answer to it. By the afternoon the blanks were full. Somewhere in the middle of the file was Ellen Park, whose home and car the office has insured for six years. She gave Paul her mobile number herself, standing at his desk, and the office had texted her renewal reminders on it since. After the pass her record held a different number. Nobody decided that. The pass filled in what was empty, and where a field was not empty it wrote anyway, because that was the default and nobody was asked. Nothing in the row says what used to be there, where the new number came from, or when either of them was true. In September her renewal reminder went to a landline she had given up years before, and in October she phoned to ask why nobody had reminded her.

The blanks getting filled is the half a supplier sells. The overwrite is the other half, and for Marta it was the more expensive one. The idea under everything on this page is that outside data arrives as a suggestion with a label on it, where it came from and when, and never replaces what a customer told you unless a person decides it should. In a real estate office, in an invented example, the same row belongs to the woman you sold a house to three years ago, who gave you that number standing in her own kitchen; real estate data enrichment and property data enrichment are the same pass with property detail added, and they have their own section below.

In short

  • An appended field is someone else's assertion about a person. It arrives without the two things that would let you weigh it: when it was true, and who said so. The Federal Trade Commission put compulsory orders to nine of these companies. It reported that most of their data comes from other companies like them rather than from an original source.
  • There is no honest published rate for how fast contact data goes stale. Figures circulate, they disagree with each other, and none of those we found states what was measured or on how many records. The rate that matters is the one you measure on your own records, two hundred of them, checked by hand.
  • And the expensive failure is not the record that comes back empty. It is the field you already knew, quietly replaced by a value you cannot check. The row does not record what was there before.

Three things this is not

The article next door owns half of what you came here for.

Not the trace

Starting from an address and finding contact details for someone who has never spoken to you is a different act. It has two federal statutes attached to it, and it has its own long article on this site. That question is about acquiring something. This one starts from a person who is already in your database, usually because they came to you. It asks what you are entitled to believe about the row you now hold.

Not the duplicate

Deciding whether two records in your own systems describe one person is a question you can answer. You can open both. There is a published model for it. It has a third outcome worth knowing about. Here the other record is inside a company you have no access to. You cannot see how it decided anything.

Not the old list

Whether you may still contact someone who went quiet three years ago is a question about permission. Permission has dates in it. This article assumes you are allowed to make the call. It is about whether the number you would dial is the one that reaches her. And it is about what happened to the number that used to be in that field.

What is CRM data cleansing, and where does data enrichment fit?

CRM data cleansing is putting the records you already hold back in a state you can trust: one record per person, fields that mean one thing, values that are still true, and a note of where each value came from. Data enrichment, buying the missing details from an outside provider, is one step of it and the last one to run.

Data enrichment is the pass that completes and corrects records you already hold. Three invented contacts show it. One came in with a first name and an email and no phone. Another has a mailing address, and nothing about the property behind it. A third is the same person as a record you took two years ago under a different email. Enrichment sends what you have to an outside provider and writes back what comes home: a number, an email, a detail behind the address, a suggestion that two records are one person.

Two words in that description are doing all the work. "Corrects" assumes the outside answer is better than yours. "Writes back" assumes there was nothing there.

The narrow thing worth understanding is that enrichment is not a lookup and not a fix. It is the arrival of an assertion from a company you have never spoken to, about a person you have. It lands in a system where it will look exactly like something you knew.

Marta's insurance office (invented). Marta's list of what her CRM needed, written after Ellen's call, had five lines on it, and enrichment was the last: records nobody would ever call again; the same person in the records twice, under a married name and a maiden name; a tag field holding forty-one different spellings of eleven ideas; values nobody could say the source of; and, last, the blanks.

Your turn. Write your own list in the same order: records nobody will call, duplicates, fields that mean more than one thing, values with no known source, and only then the blanks.

What does an appended phone number actually tell you?

An appended phone number in your CRM tells you that an outside company's file held that number against a record it decided was your person, on the day you asked; it does not tell you when the number was true, who first wrote it down, or whether it was observed or worked out.

Think about what has to happen for a phone number to appear in that column. You send an identifier: a name, an address, an email. Someone else's system then decides which of its own records describes the same human being. It decides on partial information, with no way to ask anyone. Then it returns the value it holds against whichever record it picked.

So the number in your CRM is the end of a chain of at least two guesses. The first is that this record is about your person. The second is that this number belongs to that record. Neither guess is shown to you. What arrives is a bare string in a field, formatted exactly like the numbers your customers typed in themselves.

That matching problem has its own long piece on this site, duplicate contacts in a CRM. It is worth reading, because the published model for it has a third outcome that a build can throw away. One difference changes the whole picture here. In a sync, both systems belong to you and you can open both. In an enrichment response the other system is a black box. The threshold it used, the fields it weighed and the confidence it settled on are all facts about that company, and you are a customer of it.

What arrives in the column

Four things an appended value does not tell you.

It does not say when

A phone number in an enrichment response is the best answer in someone's file on the day you asked. The file does not say when it was assembled. Two numbers can look identical in your CRM. One can be a fact confirmed last month, and the other a fact that was true in a different decade. Nothing in the row tells them apart.

It does not say who

The name on your invoice is the company you bought from. That company mostly bought it as well. So the truthful answer to where a value came from is the last name in a chain, rather than the name of whoever first wrote it down. And nobody publishes the chain. That is not a criticism of any particular supplier. It is the shape of the market, and a federal regulator has said so on the record.

It does not say observed or inferred

Some of what sits in these files was written down by someone who saw it happen. Some of it was worked out from other things, by a company that has never met the person it describes. Both come back through the same interface, in the same shape. So a conclusion that has been rounded into a value looks exactly like a measurement by the time it reaches your CRM.

It does not say what it replaced

This is the one that costs money, and it is entirely under your control. If the pass wrote into a field that already held something, the row should say so. Whether yours does is a thing you can check this afternoon. The answer decides whether a bad pass can be undone. Where it does not, the old value is simply gone. Nobody chose to discard it. And the one person who could have confirmed it was right is the customer whose number you no longer have.

Marta's insurance office (invented). The number that replaced Ellen's mobile was a landline at an address she had left. The outside file had matched her name and her old street to a record of its own, and returned the number it held there. Both guesses were reasonable, and both were invisible from Marta's side: the row showed a phone number and nothing else.

Your turn. Open one record that a pass has touched and ask of each appended field: when was this true, who said so, and was it seen or worked out? Write down which of the three questions the record can answer.

Looking steeply up a red brick wall at a tall painted sign in white capitals on a dark panel, its letters running vertically one above the next, the upper ones painted over each other so the name cannot be resolved and only the last of them standing clear and reading TOR, with FOUNDRY and CO in horizontal capitals below, a corbelled brick cornice along the top, a window in the brickwork to the right and pale blue sky along the left edge
Someone painted that, and it was true. Someone painted over it, and that was true too. What is left is a wall carrying two claims at once, one of them readable and neither of them dated. There is no way to tell from the wall which one you are looking at. That is what a contact record looks like after a pass that overwrote a field. And unlike the wall, your CRM shows you only the top layer. Photograph by A Continuous Lean, CC BY 2.0.

Where does appended contact data come from?

Appended contact data comes, in the one study made under compulsion, from companies that buy from and sell to each other more than they observe anyone: the Federal Trade Commission ordered nine of them to explain where their data came from, and reported that they get most of it from other companies in the same trade.

The primary document this page relies on has held up for twelve years because of how it was made.

In December 2012 the Federal Trade Commission issued compulsory orders under section 6(b) of the FTC Act to nine named companies that collect and sell information about consumers. The orders required them to file special reports on where their data comes from, what they do with it and what tools consumers have to control it. The Commission published the result as its report of May 2014, A Call for Transparency and Accountability, covering their practices from January 2010. An older source, kept for the argument: nine companies, named, under compulsion, describing themselves to a regulator. This page found nothing else like it in this subject.

The evidence

Where nine data sellers got their data, under compulsory FTC orders

Where nine data sellers got their data, under compulsory FTC ordersPurchase information about individuals from commercial sources: 8 of 9. Buy data from, or sell it to, each other: 7 of 9. Obtain information directly from federal government sources: 6 of 9. Counts out of the nine named companies the Federal Trade Commission ordered to file special reports in December 2012, under section 6(b) of the FTC Act. The orders covered their practices from January 2010. The three rows overlap: most of the nine do all three. The axis is nine because that is the whole population studied, not a sample of the industry.Purchase information about individuals from commercial sources8 of 9Buy data from, or sell it to, each other7 of 9Obtain information directly from federal government sources6 of 9

Counts out of the nine named companies the Federal Trade Commission ordered to file special reports in December 2012, under section 6(b) of the FTC Act. The orders covered their practices from January 2010. The three rows overlap: most of the nine do all three. The axis is nine because that is the whole population studied, not a sample of the industry. Source: Federal Trade Commission, Data Brokers: A Call for Transparency and Accountability, May 2014.

This report is from 2014, and the industry may have changed since. So do not read the bars as a description of whoever supplies your own enrichment today. Read the middle one as a structural fact, because that is what it is. When companies in a market mostly buy from each other, no single one of them can tell you where a value originally came from. The chart cannot show you the thing that follows from it. The answer to "where did this come from" stops being a name and becomes a direction of travel.

The finding that changes how you should think about an enrichment response is the middle bar. These are not nine companies each independently observing the world. They are, to a substantial degree, one market trading the same records among themselves. The Commission's report states the consequence plainly: the nine get most of their data from other companies in their own trade rather than directly from an original source, and one of them draws consumers' contact information "from twenty different sources".

That is the honest picture behind the field in your CRM. Not a company that knows something about your customer, but the last company in a queue, passing on what it was passed.

It would be virtually impossible for a consumer to determine how a data broker obtained his or her data; the consumer would have to retrace the path of data through a series of data brokers.

Read that as a statement about you rather than about consumers. If the person the record is about cannot retrace it, neither can you, and you are the one who is going to be asked.

Marta's insurance office (invented). Marta asked her supplier where the landline had come from. The answer named a second company the supplier buys from, and that company was the end of what anyone could tell her.

Your turn. Ask your own supplier, in writing, where one appended value on one named record came from, and write down how many companies the answer names before it stops.

Is an appended field a fact or a guess?

An appended field can be either a fact someone observed or a conclusion someone drew, and the two arrive in the same column in the same shape, so the only way to know which you bought is to ask the supplier, field by field.

There is raw data, which the report describes as things "such as a person's name, address, home ownership status, or age". And there is derived data, "which they infer about consumers". The report gives its own examples of how inference works: a boating licence taken as an interest in boating, a subscription to a technology magazine taken as an interest in technology, and two cars of the same make taken as loyalty to that brand.

Those are marketing categories rather than contact details, and it would be a mistake to say your appended phone number is an inference. What is not a mistake, and is the point, is that a field in your CRM has no way of recording which kind of value it holds. If a supplier's file holds observations and guesses in the same layout, both land in your database looking the same, and nothing in the row says which is which.

The practical version of this is a question with a short answer, and it is worth putting in writing to whoever supplies you: for each field you buy, is this something the source observed or something the source concluded? A provider who can answer that field by field is telling you a great deal about how carefully the product was built.

Marta's insurance office (invented). Marta's supplier offered, beside phone numbers, a "likely homeowner" flag and an "estimated household income" band. She asked which of the three were observed. The answer was that the phone numbers came from records and the other two were estimates, and she bought only the phone numbers.

Your turn. List every field your supplier sells you and write beside each one: observed, concluded, or no answer.

Why is the same contact data sold under three names?

The same contact data is sold under three names, as a marketing list, an identity check and a people search page, because the companies the FTC studied make all three out of the same raw and derived data; what separates the products is the purpose of the buyer.

The evidence

What the same nine companies sell the same underlying data as

What the same nine companies sell the same underlying data asMarketing products, for reaching people: 5 of 9. Risk products, for checking that people are who they say: 4 of 9. People search sites, for looking a person up: 3 of 9. Counts out of the same nine companies, from the same report. They add up to more than nine because several of them sell in more than one category. That is the finding rather than an error. The report describes the same raw and derived data being made into a marketing list, an identity check and a public lookup page.Marketing products, for reaching people5 of 9Risk products, for checking that people are who they say4 of 9People search sites, for looking a person up3 of 9

Counts out of the same nine companies, from the same report. They add up to more than nine because several of them sell in more than one category. That is the finding rather than an error. The report describes the same raw and derived data being made into a marketing list, an identity check and a public lookup page. Source: Federal Trade Commission, Data Brokers, May 2014, on the three product categories the nine sell.

The counts are from 2014, and the industry may have changed since. So read them as a shape rather than as a market share. Note that the three overlap, because several of the nine sell in more than one category. What the chart cannot draw is the part that matters. The report describes all three shelves as stocked from the same raw and derived data. And the description is picked at the moment of sale rather than at the moment anybody wrote the data down.

One pool of data, three shopfronts. The identity check, the marketing list and the people search page are built out of the same material. What separates them is the purpose the buyer had.

What the law allows you to do with contact information is a question for the provider and for your own counsel, and this page does not answer it. Before buying any appended contact details, ask the provider in writing which permitted purpose your account sits under, and whether anything it supplies comes from a consumer reporting agency.

Marta's insurance office (invented). Marta put both questions to her supplier in one email and kept the reply with the contract, so that the answer is on file before the next pass, not after it.

Your turn. Send your supplier the same two questions and file the answer where the next person to run a pass will find it.

How fast do contact records go stale?

Nobody can tell you how fast your contact records go stale before measuring your own: a rate measured on someone else's records says nothing certain about yours, so the useful number is the one you measure on two hundred of your own records, checked by hand now and again later.

Figures for how fast contact data goes bad circulate, and this page uses none of them: a figure is only worth quoting with the sample it was measured on, the method and what was counted, and a rate measured on someone else's records says nothing certain about yours.

So there is no decay rate on this page, none in the calculator, and none on our service page. There is something more useful: the reason a rate cannot be a single number in the first place, and a way to measure your own.

The measurement is plain. Take two hundred records at random from the part of your database you would really work, among the people you are allowed to contact. Check each one by hand and mark it confirmed, wrong or unknown: a value is confirmed only when the person confirms it, by answering the number, replying from the email or saying the address is theirs, and wrong when someone else answers or the person says it is not theirs. Anything else counts as unknown, not as right. It needs no tool and no purchase, only the time it takes to reach two hundred people. That gives you the share that is wrong today. Check the same two hundred again in six months and count only the values that were confirmed the first time and are wrong now: that count, out of the confirmed ones, is your own rate. What comes out is measured on your own people, in your own market, at their own ages, which is what a quoted figure cannot be.

Marta's insurance office (invented). Paul took two hundred records from the policyholders and recent quotes, the part of the database the office really works, and checked them over two afternoons. In Marta's invented example, thirty-one of the two hundred had a phone number that someone else answered, nineteen more stayed unknown, and twenty-one of the thirty-one wrong numbers belonged to people who had asked for a quote and never bought. Paul put the same two hundred in the calendar for April, to check again.

Your turn. Pick your two hundred and write down, before you start, what share you expect to be wrong. Then check them and compare.

Why does the rate depend on who your customers are?

The rate at which a CRM record goes stale depends on who your customers are, because a field stops being true when something happens to a person, a move, a new job, a new phone, and those things happen at different speeds to different people.

Contact details do not rot on their own. They stop being true when something happens to a person: a move, a job change, a marriage, a new phone, a business closing. So the speed at which a field goes wrong is the speed at which that thing happens to the people in your database, and different people are not the same.

Which means the useful question was never how fast data decays. It is what sits underneath a particular field, and how often that thing changes for the people you hold. One government survey measures exactly that shape, for exactly one field. The field is employment.

The evidence

How long people had been with their current employer, by age, January 2024

How long people had been with their current employer, by age, January 202425 to 34 years old: 2.7 years. 35 to 44 years old: 4.6 years. 45 to 54 years old: 7.0 years. 55 to 64 years old: 9.6 years. Median years of tenure with the current employer. It comes from a supplement to the Current Population Survey, a monthly sample of about sixty thousand households. Median means the point at which half of all workers had more and half had less. So half of the youngest group had been in the job under two years and nine months.25 to 34 years old2.7 years35 to 44 years old4.6 years45 to 54 years old7.0 years55 to 64 years old9.6 years

Median years of tenure with the current employer. It comes from a supplement to the Current Population Survey, a monthly sample of about sixty thousand households. Median means the point at which half of all workers had more and half had less. So half of the youngest group had been in the job under two years and nine months. Source: U.S. Bureau of Labor Statistics, Employee Tenure in 2024, Table 1, released September 2024.

This is about employment, not about the mobile number of a customer. It is here for one reason. It is the only measurement in the whole of this subject that names its survey, its sample and its definition. Use it as an argument rather than as a number. Whatever underlies a field decides how fast that field goes wrong. The rate is different for different people. So a single percentage covering everybody in your database is describing a population that does not exist. This page found no equivalent measurement with a stated method for a personal mobile number, a personal email address or a mailing address. That is worth holding next to the way the figures circulating about data decay are quoted, as though someone had done exactly that study.

The Bureau of Labor Statistics release on employee tenure for January 2024 puts the overall figure at "3.9 years", the lowest since 2002. An older source, kept for the argument: it is the edition this page was written from, and the Bureau has since published one for January 2026. The chart above is Table 1 of the same release, and the spread across it is the whole argument.

Now put, in an invented example, a twenty-six-year-old and a sixty-year-old in one database with a work email address each. Run a single percentage across the pair of them. Whatever that percentage is, it describes neither of them, and it will be quoted as though it described the database.

Marta's insurance office (invented). Paul's sample showed the same shape: the wrong numbers sat among the younger quote requests, and the long-standing policyholders, Ellen aside, had barely moved. One rate across the whole database would have described neither group.

Your turn. Split your two hundred by one thing you already know about each person, such as how long they have been a customer, and count the wrong values in each half separately.

How do you clean stale records in a CRM, step by step?

You clean stale records in a CRM in an order that protects what you already know: choose the records worth keeping, merge the duplicates, make each field mean one thing, add a source and a date to every value, measure a sample by hand, decide the rule for disagreements, and only then fill the blanks from outside.

Before the first step, export the whole CRM to a file and keep it, and try every bulk change on ten records first, including restoring one of them from the export. A mistake can then be repaired from the copy; a merge or a deletion is still hard to undo, which is why those are tried on the ten before anything else.

  1. Choose the records you would really work. Not the whole database. The rows someone in your business would call or write to this quarter. Everything after this step is done to those rows first.
  2. Merge the duplicates. Two records for one person will each be cleaned and enriched separately, and come back disagreeing. The method for deciding which records are the same person is in duplicate contacts in a CRM.
  3. Make each field mean one thing. A tag field with forty spellings of eleven ideas cannot be cleaned by any outside file. Write the short list of values each field may hold, map every existing value onto it, and move any value that does not fit into a note on the record rather than deleting it.
  4. Add a source and a date to every value. Two columns beside each contact field you care about, the same design everywhere on this page: where the value came from (the customer, a form, a call, a named supplier) and the date it was written, which is not the date it became true. An outside value that disagrees goes into a third, suggestion field; nothing already there is deleted. Fill them for what you already have as far as anyone knows, and mark the rest "unknown".
  5. Measure a sample by hand. The two hundred records from the section above. Now you know the share that is wrong today; the second check, months later, gives your rate.
  6. Decide what happens when two sources disagree. The four choices are in the next section. Write down one per field.
  7. Only then fill the blanks from outside. Ask your own customers first where you can, at a renewal, a visit or a booking; then an enrichment pass, run on the chosen records, writing into empty fields only, with the source and the date written beside every value it adds.

Marta's insurance office (invented). Marta kept twelve hundred of the four thousand records: every policyholder, and every quote from the last two years. Merging found ninety people who were in the CRM twice, which left eleven hundred and ten records. The tag field went from forty-one spellings to eleven values. Paul added the source and date columns and marked every existing phone number "unknown" except the ones a customer had confirmed to the office since the spring. Ellen's was one: she had given her mobile again on her October call, and Paul had put it back in her record that day with "the customer, by phone" beside it. The sample was already done. For phone numbers Marta combined two of the four behaviours in the next section: both values kept, the outside one in a separate suggestion column, and for current policyholders a person decides. Then she ran the pass again, on the eleven hundred and ten, set to write into empty fields only and to return, for every field that already held a number, the value it would have written.

Your turn. Write the seven steps on one page with your own numbers beside each: how many records you keep, how many duplicates you expect, which fields need a short list.

What should happen when your record and the outside file disagree?

When your CRM record and an outside data file disagree, one of four things happens, and every build has one of them: the newest value wins, yours wins, both are kept side by side, or a person decides. Find out which one yours does before a pass runs, not afterwards.

The first is that the newest value wins. Newest means the value that arrived most recently. It does not mean the value that most recently became true. A number your customer confirmed to you last month gets replaced by a number a provider has held since some date nobody recorded, because the query happened today. For a value a customer gave you, this is the one to rule out.

It is worth seeing how that behaviour is presented by the software, because it tells you which way the tooling leans. HubSpot's documentation for importing records describes protecting what you already hold as something you switch on. There is an advanced option called Prevent property overwrite. The page explains it as: "if you're updating existing records, prevent the import from overwriting records' existing property values for the row". When it is selected for a property, "the import will update the property for new records and existing records that have never had a value for the property. It won't update the property for existing records that have a value or had a value in the past, even if currently empty."

What matters there is the shape and not the detail. On one major CRM, per property, at import time, keeping your own value is a checkbox. That is one platform and not a survey of the field. It is exactly why the four choices are worth settling in writing before a pass runs, instead of discovering afterwards which one you got.

The second is that yours wins and the response is discarded. Safe, and it quietly turns enrichment into a fill-the-blanks exercise. That may be exactly what you wanted, and it should be a decision rather than an accident.

The third is that both are kept, in separate fields, with the outside one clearly marked as a suggestion. This is the one that costs least to be wrong about. It costs one suggestion field beside each contact field.

The fourth is a review queue. Disagreements go to a short list, and a person settles them. That is right when the field matters and the volume is small. The four can be combined, field by field or by kind of customer. It is the same shape as the clerical review step in the published record-linkage model, which the duplicate contacts article covers, and that step exists for exactly this reason.

Whichever you pick, one habit does more than the choice itself. Write down, on every enriched row, where the value came from and when it was written. Not in a log somewhere: in the record, beside the value, where the person about to dial it can see it. Those two columns are the difference between "I do not know where that number came from" and a one-sentence answer, and it is easiest to add on the day the build is done.

A stained parchment leaf in ochre and olive, closely covered in Greek script in dark brown ink with several lines and one large initial picked out in red, fainter marks visible between and behind the writing, the ragged left edge lit blue where a rod holds it against a pale mount
The page was scraped clean and written on again, which is what people did when parchment was expensive. The older text is still there, faintly, under the newer one. The only reason anybody can read it now is that someone went looking for it with the right equipment. Keeping what you overwrote costs a database column. Recovering it afterwards costs a research project. Photograph by Walters Art Museum Illuminated Manuscripts, CC0 1.0.

Marta's insurance office (invented). Of the three hundred and seventy chosen records that already held a phone number, the second pass returned a different number for eighty-three. Under Marta's rule none of them was overwritten: each went into the suggestion column, and the twenty that belonged to current policyholders went onto Paul's review list. He settled them by phone over a week, four or five a day. Ellen's record was among the eighty-three: the pass returned the old landline again, and under the rule it went into the suggestion field, beside the mobile she had given in October.

Your turn. For each field you would enrich, write which of the four behaviours you want, and find the setting in your own CRM that makes it happen.

Who is responsible for a wrong record about a customer?

The business that holds a wrong record about a customer in its CRM is the one a correction law addresses, not the company that supplied the value: California's consumer privacy law gives consumers the right to ask the business that maintains inaccurate information to correct it.

There is one place where a legislature has written down what happens when a business is holding something wrong about a person. It is not a federal statute, and it may not apply to you. It is worth knowing anyway, because it says whose problem this is.

California's consumer privacy law, at Civil Code 1798.106, gives a consumer "the right to request a business that maintains inaccurate personal information about the consumer to correct that inaccurate personal information, taking into account the nature of the personal information and the purposes of the processing". A business receiving such a request "shall use commercially reasonable efforts to correct the inaccurate personal information as directed by the consumer".

Read who that is addressed to. Not the data supplier. The business that maintains the information, which in the story at the top of this page is Marta's office.

Whether it applies to you is a separate question. The thresholds are published, so a first reading is a matter of three lines against your own year; whether an exception applies to your business is a question for counsel. The same law defines a covered business at Civil Code 1798.140 as one doing business in California that also meets one of three: annual gross revenues over twenty-five million dollars in the preceding year, a figure the law adjusts over time, or annually buying, selling or sharing the personal information of a hundred thousand or more consumers or households, or deriving half or more of its revenue from selling or sharing personal information. Read those three against your own year and you will know where you stand.

So this is not a compliance duty for a small business below those lines. It is a description of the right shape, written by people who thought hard about it, and it says two things worth adopting whether or not anyone is making you. The duty attaches to whoever holds the record rather than to whoever supplied it. And answering it requires knowing where a value came from, which is the column this page keeps coming back to.

The same FTC report is blunt about how rarely that was possible. Of the companies it studied, it found that only two allowed consumers to correct their personal information for marketing purposes. The report's sentence does not make its denominator explicit, so this page gives no share and draws no chart.

Your turn. Take one field a customer could ask you to correct, and write down how you would find out where its value came from, using only what is in the record today.

Does New York's SHIELD Act cover the fields enrichment adds?

New York's SHIELD Act does not cover a phone number, a mailing address or an email address on its own, which is what an enrichment pass adds to a CRM: it protects "private information", a narrow defined term, and a phone number, a mailing address or an email address on its own is not on its list.

The New York Attorney General's own description of the law it enforces says the Act requires "any person or business that maintains private information to adopt administrative, technical, and physical safeguards". No revenue threshold is attached, which is a wide obligation. But "private information" is a defined and narrow term. On the Attorney General's page, it means personal information combined with a Social Security number, a driver's licence number, or an account number with its security code. And the Act extended it to biometric information, and to a username or email address together with a password.

A telephone number is not on that list. A mailing address is not on that list. An email address on its own is not on that list.

So the New York statute people mention in this context is about security and breach notification for a specific set of high-risk identifiers. A phone number or an address on its own is not one of them; a record that already holds one of the listed items, such as an account number with its code, is covered because of that item, not because of the phone number beside it.

Nothing above is quoted from the statute itself. It all comes from the Attorney General's published summary of the law that office enforces. That is a step further from the text than this page would like to be, and it is worth saying out loud, because the whole finding here is what one defined term does and does not include.

The system

From your own record back to your own record.

Six hops, and it is a loop rather than a line. That is the thing that makes this different from every other data project in your business. The output lands on top of the input. Three of the six happen inside a company you have no access to. The only two you control are the first and the last. That is where all of the available honesty lives.

The path from a contact record you already hold, out to a matching decision made inside somebody else's system, and back onto the same row: Your record, The query, The match, The claim, The write, The rowYour recordWhat you hold nowThe queryWhat you send outThe matchTheir guess, unseenThe claimA value with an ageThe writeWhat it lands onThe rowWhat it says after

Scroll to follow the chain

  1. Your record: What you hold now
  2. The query: What you send out
  3. The match: Their guess, unseen
  4. The claim: A value with an age
  5. The write: What it lands on
  6. The row: What it says after

In your numbers

How many fields would a pass replace that you already knew?

The number you would actually send, which for most people starts as the whole database and should not be.

2,000

Not the share you would call. The share where the field you are about to enrich is not empty. Sort by the column and look.

40%

This one is yours to supply, and it is worth measuring rather than guessing. Run two hundred records where you already know the answer, and count the disagreements. Nobody can quote it to you honestly before seeing your list.

25%

Opening the record, seeing where your own value came from, deciding. Seconds when you remember the person and much longer when you do not.

3

Fields where somebody has to decide which answer is yours

200disagreements

Records in the passyour 2,000
2,000records
Where the field is not emptyyour 40%
800filled fields
Where an outside file says something elseyour 25%
200disagreements
At your settling timeyour 3
600minutes
In hours60 minutes in an hour
10hours

The headline is the third row rather than the hours. The hours are the affordable half, and the count is the one nobody asks about before buying. Every one of those is a moment where something will choose on your behalf if nobody has chosen deliberately. The choosing happens silently. Shares of records produce fractions, and half a disagreement is not a thing. So read anything with a decimal in it as a rough count. Three things this deliberately refuses. There is no rate of decay anywhere in it, because none of the circulating figures states a sample or a method, as the section above explains. There is no figure for how often the appended value turns out to be the correct one. The only person who can measure that is you, on a couple of hundred of your own records where you already know the answer. That measurement is worth more than anybody's benchmark. And there is no money in this calculator at all. A reachable customer is not a sale. The loss from a number you can no longer reach is not payable on any date. And the sums that turn either into a dollar figure do not exist.

What does a data enrichment pass take, and what decides its size?

A data enrichment pass takes two kinds of work: the supplier's pass over your records, whose size grows with how many rows you hand over, and the work around it, choosing the records, writing the source and date, settling the disagreements, which is where the value is and which gets skipped.

The supplier's part is set by the supplier, and two questions decide its size: how many records you hand over, and how the supplier counts them.

Which makes the first decision how many records to run. The number worth having is the count of rows someone in your business would call this quarter, not the whole database. Producing it takes thinking rather than a purchase, and whatever it comes to is the honest size of the job. The gap between it and the whole database is money about to be spent making people reachable that nobody was going to reach.

The second thing to settle before signing is how the supplier counts: every record you submit, or only the ones that come back with something. That difference changes the sums completely: in an area where not much resolves, the first arrangement can spend a budget and leave you with little to show for it.

What we would build is the work around the pass, and it is where the value is: deciding which records qualify, writing the source and the date onto every enriched row, making the disagreement behaviour explicit rather than inherited, and putting the flagged rows somewhere a person will look. They are easy to skip when nobody puts them on the list at the beginning. The pieces sit beside everything else we build, on the home page.

On time, the part to plan for is the conversation about the four disagreement behaviours above, and about which fields you want touched, the same kind of conversation that sits at the heart of an AI assessment. That conversation is worth having before the first record moves. Adding a source column to a database that has already been enriched twice is a miserable job. If your contact records already carry a source and a date, this is straightforward. If they do not, adding that column is the project, and the enrichment is the easy part bolted on afterwards.

What does data enrichment not do?

Data enrichment does not verify a person, tell you how old a value is unless you ask for it, promise a match, make a record legal to contact, tell you anyone wants to hear from you, or clean up after itself.

It does not verify a person. What a phone and email check establishes, as our own service page describes it, is that a number is well formed, is in service and is not a duplicate. Those are useful checks. None of them is the check people assume, which is that this number reaches this person. Our own service page uses the word "verify", and this is the sentence that qualifies it.

It does not tell you how old a value is unless the provider is asked for it and passes it through. Freshness is the attribute that answers the first question on this page. It is worth asking for by name, because a response that does not carry it looks exactly like one that does.

It does not promise a match. A rate quoted before anyone has seen your list is a rate about someone else. Two hundred of your own rows, run as a sample, settles the question for your own records.

It does not make a record legal to contact. Do-not-call registrations and consent rules attach to the call rather than to the data, and they are not affected by how the number was obtained. They are covered in detail in the article on reactivating an old database.

It does not tell you anyone wants to hear from you. A complete record is not proof that the person can be reached, and interest is a different question again, and it is not in any file anyone can sell you. A pass that makes four thousand people reachable has not produced four thousand conversations, or any evidence about whether they would be welcome.

And it does not clean up after itself. If a pass writes a value that turns out to be wrong, the wrongness is now yours. It sits in your system of record with your business's name behind it. The provider's involvement ended at the response.

Three ways a pass that worked leaves you worse off

All three happen after the data arrives.

The database looks finished

Before the pass, the gaps were visible. Everybody handled the list accordingly. Afterwards every row is full. In the interface, a number a customer typed in herself looks exactly like a number a file suggested. The database now carries the confidence of its best row and the accuracy of its worst. And the only person who could tell the two apart has stopped being able to.

Nobody can undo it

Six weeks later someone notices a run of wrong numbers. They want to know which pass introduced them. If the write did not record the previous value, the date and the source, that question has no answer. The only remaining option is to run another pass over the top of it and hope.

It is spent on people you will never call

Enrichment is charged per record. A database that has been collecting for years holds a great many rows nobody has any intention of calling. Running the whole thing is the default, because it is one click. It turns a budget into fullness rather than into conversations. The version worth buying is the one where someone chose the rows first. Choosing them costs an hour.

The honest read

Export twenty contacts you would swear you know, phone numbers and all. Send them to us with the numbers removed. We will run them. We will send back what an enrichment pass would have written into those fields. You can see for yourself how many agree with what you already had, how many disagree, and how many come back empty.

Ask us what your last pass did

It is twenty rows, and it costs nothing. We do not need access to your CRM. You keep the only copy that has the real numbers in it.

How do you check what your last enrichment pass did?

You check what your last enrichment pass did on the records you already have, with no tool: open contacts you know, ask where each value came from and when, find someone you know moved, and ask what the overwrite setting was.

  1. Open five contacts you personally remember. People you have spoken to. Not the first five, and not the ones you have been working this week.
  2. For each one, ask where the phone number came from. Not whether it is right. Where it came from. If the record cannot tell you, you have already found the finding, and you have found it on the five people you know best.
  3. Look for a date beside any enriched value. Anything at all: an appended-on date, a source field, a note. If there is nothing, then every value in the database is the same age as every other value, which is to say unknown.
  4. Find a contact you know moved. Someone who left the area. Look at what the record says now, and decide whether the record knows they moved, or is quietly still describing the person who lived there.
  5. Sort by the phone column and look at the gaps. Count how many rows are empty. That number is the real size of the problem enrichment solves, and it is worth holding next to the number you were about to send.
  6. Ring one number that you did not put there yourself, on a contact you are allowed to call. Not to sell anything. Just to find out who answers. One call tells you more about your own data than any report, and if the answer surprises you, that is the finding.
  7. Ask whoever ran the last pass what the overwrite setting was. Not what should have happened. What the setting was, per field. If nobody can answer, treat that as an open question rather than as a reassurance.

Marta's insurance office (invented). Marta ran this check in October, the week of Ellen's call and before any of the seven clean-up steps, and it is how she learned what the spring pass had done: of her five remembered customers, three had a phone number nobody could explain, and nobody could say what the overwrite setting had been.

What does data enrichment add in a real estate office?

In a real estate office, data enrichment adds the same contact fields as anywhere else, plus property data: which property a contact owns, what the public records say about it, and the detail behind a mailing address. The cleansing method above is the same, in the same order.

Real estate data enrichment and property data enrichment services sell two different things that arrive in one response. The contact half is the phone number and the email, with every question on this page attached to it. The property half is detail drawn from public records behind an address: the owner of record, the property's description, the last sale. How much of it comes back for your own addresses is a question a sample of your own rows answers, not a quoted match rate.

The overwrite in this trade has a familiar face, in the invented example from the top of this page: the woman you sold a house to three years ago gave you her number standing in her own kitchen, and a pass that writes over non-empty fields will replace it with whatever an outside file holds against her old address. A step-six choice that keeps your value and puts the outside one in a suggestion column protects her record the same way Marta's protected Ellen's.

Tagging cleanup and standardization in a real estate CRM is step three of the method: a tag field that holds "buyer", "Buyer", "buyers 2023" and "past client - bought" is four labels for three different things: who the person is to you (a buyer), when (a year), and a stage (a past client). Write the short list of tags (for example, in an invented office: buyer, seller, past client, renter, investor), keep a stage or a year in its own field, map every old tag onto the list, and move what does not fit into a note before any outside data arrives, because an outside file cannot tell which spelling meant what.

And the line with the neighbouring topic matters more here than in any other trade. Starting from a property and finding contact details for an owner who has never spoken to you is a different act from completing a record for someone already in your database, with its own legal questions, and this page does not cover it.

Common questions about CRM data cleansing and data enrichment

What is data enrichment, in plain terms?

Data enrichment is a pass over contact records you already hold that sends what you know to an outside provider and writes back what it holds: a missing phone number or email, a detail behind an address, a suggestion that two records are one person. It is sold as completing your database. What it does is add someone else's assertions to your database. That is a useful thing, and it is not the same thing.

What is the difference between CRM data cleansing and data enrichment?

CRM data cleansing is the whole job of making your records trustworthy again: removing duplicates, making each field mean one thing, recording where values came from and fixing what is no longer true. Data enrichment is one step inside it, filling blanks from an outside source, and it belongs at the end, after the steps that protect what you already know.

How is enrichment different from skip tracing?

On this page, enrichment means completing what you hold about someone already in your database. Finding contact details for someone who has never dealt with you, starting from an address, is a different act, the one called skip tracing, and this page does not cover it. Whether there was ever a relationship, and what that means in law, is a question for the provider and for your own counsel.

Will enrichment overwrite the data I already have?

Whether enrichment overwrites the data you already have depends on how the pass is set up, and it is worth settling rather than assuming. On the CRM whose documentation is quoted above, not overwriting a value you already hold is a per-property checkbox selected at import time, which tells you which way the tooling leans. Decide it deliberately: fill blanks only, keep both values in separate fields, or send disagreements to a person. Whichever you choose, insist that the previous value, the source and the date are written to the record. Without those, the only way back from a bad pass is the export you made before it, if you made one.

How accurate is appended contact data?

The accuracy of appended contact data on your own list cannot be known before running it. What you can do is measure it: take two hundred records where you already know the answer, run them, and count how many agree, disagree and come back empty. It is a number about your own customers rather than someone's benchmark.

How fast does contact data go stale?

How fast contact data goes stale on your own records is a number only your own sample gives you. What is knowable is the mechanism: a field stops being true when something happens to a person, so the rate is a property of the people in your database rather than of the data. It will be different for a first-time customer in her twenties and a couple who have been with you for decades.

Does an enriched list mean I can call it?

An enriched list does not mean you can call it, and the two questions are entirely separate. Whether a record is reachable is a data question. Whether you may contact it is a rules question with dates and registries in it, and the answer does not change because the number was appended rather than volunteered. The neighbouring articles on this site cover that ground.

What should I ask a provider before signing?

Before signing with a data provider, ask five short questions. Does the charge land on every record submitted or only on the ones that resolve? Does the response carry a date or an age for each value? For each field, is it observed or inferred? If a person tells you never to contact them again, what changes in your system? And will you run two hundred records from my own area before I commit to anything? Written answers to all five tell you what kind of company you are dealing with.

What is the one thing that makes CRM data cleansing manageable?

The one thing that makes CRM data cleansing manageable is a source column and a date column beside each contact field, filled in from the day the record is created, whether the value came from a form on your website, a conversation, or a pass. It is easy to add before you have data and close to impossible to rebuild afterwards. The hard questions on this page become answerable once those two columns exist.

Do you work with businesses outside New York?

We are based in New York, and the call works for an owner anywhere. The California and New York rules on this page are those states' own; the method applies to any business's records.

What to do first about stale contact records

Nothing on this page argues against enrichment. In Marta's invented example, two thirds of a database with no phone number in it was a real problem; asking your own customers fills some of it, and for the rest the only source is someone else's assertions.

What it argues is that the assertion should arrive wearing a label: where it came from, when, and what it landed on. Those three facts turn an appended value from something you have to trust into something you can weigh. They are a small part of a build to capture, and they cannot be recovered once a pass has run without them.

Pick one contact you are certain about, someone whose number you have used, and open their record. Then find the answer to a single question: where did that number come from, and when. If your CRM cannot tell you that about the one person you are sure of, it cannot tell you about the four thousand you are not. And every argument in this article is an argument about a column you have not added yet.

There is no figure here, because the largest part of the bill is paid to someone else, per record. It moves with how many records you send, and with whether you are charged for attempts or for successes. What we would build around it is the smaller and duller half. Deciding which records are worth running at all. Writing the source and the date onto every row. And making sure a value that lands on a field which already had one goes somewhere a person can see.

Start with the page on data enrichment for small business: what it appends, what it checks, and what it leaves to you.

Sources

  • Federal Trade Commission, A Call for Transparency and Accountability, the report of May 2014 on nine companies that collect and sell consumer data. An older source, kept for the argument: a study of the trade made under compulsory orders.
  • U.S. Bureau of Labor Statistics, the release on employee tenure, January 2024 survey, and its Table 1. An older source, kept for the argument: the edition this page was written from; a January 2026 edition has since been published.
  • HubSpot, the knowledge base page on importing records, the Prevent property overwrite option. Live documentation.
  • California Civil Code 1798.106 and 1798.140, on the California Legislative Information site. The law in force.
  • New York State Office of the Attorney General, its page on the SHIELD Act. The office's own summary of the law it enforces.

Know somebody who would argue with this? Send it to them.