The one channel where you already know who saw the ad

A heating and cooling company mailed 40,000 postcards in September, each carrying a code for $150 off a furnace tune-up or replacement. The printing, list and postage came to $22,000. By the end of October the code had been redeemed 61 times, the finance lead divided one number by the other, arrived at $361 per redemption, and put the next mailing on the list of things to cut.

The marketing manager ran a matchback before the meeting: the 40,000 addresses against every customer the CRM had billed in the sixty days after the cards landed. 412 of them had bought. That is $53 per customer, and on that number the postcards were the best money the company spent all year. Two analysts, one mailing, and a sevenfold disagreement — and the uncomfortable truth, which the company only learned because somebody had set aside 4,000 names and not mailed them, is that both numbers were wrong.

Direct mail is not the unmeasurable channel

Direct mail is the only paid channel that gives you a complete list of the people it reached, by name and address, before a single response comes in. Every digital channel estimates its audience; mail knows it. That makes it one of the most measurable channels there is, provided you measure the list rather than the coupon.

The reputation comes from how mail used to be measured, not from what it is. For decades the response device was the measurement: a reply card, a coupon, a code, a dedicated phone line. Whatever came back through the device counted, and everything else did not exist. When digital channels arrived with click tracking, mail looked blind by comparison, and the industry absorbed the idea that it could not be measured properly.

It was the other way round. A display impression is an estimate about a device that may or may not have belonged to a person who may or may not have looked at it. A postcard is a physical object delivered to a specific address, from a list you bought, rented or already own. The recipient file is the audience, and the question “did the people we reached buy more” is directly answerable against a sales file. Almost no digital channel can say the same.

Mail was never unmeasurable. It was measured by the coupon, and the coupon is the least reliable part of the whole exercise.

Promo codes count the people who remembered the code

A promo code measures response that happens through the code, which is a minority of the response mail produces. Most people who act on a mail piece phone, search or visit without it, so code redemptions set a floor on the effect and are routinely mistaken for the whole of it.

• The customer phoned. The receptionist booked the visit and did not ask about a code, because nobody asks about codes on the phone.
• The customer searched the company's name on Google and clicked the first result — so the sale was recorded against branded search, which is how mail's work turns up as search's performance.
• The card sat on the fridge for three weeks, was thrown away during a clean-up, and the customer came back from memory.
• The customer was already in the pipeline, got the card, and it tipped a decision the salesperson had been waiting on. No code, because they were already talking to a person.

Every one of those is a real response, and none of them uses the device. The branded-search case deserves particular attention, because it moves the credit rather than losing it: the mail drives a search, the search gets the click, and the paid search report shows a good month. Cut the mail on the strength of the coupon count, and a few weeks later branded search falls for no reason anybody in the search team can find.

Matchback: the mail file is the lead file

Matchback treats the mailing list as a list of people exposed to the campaign and matches it against the sales file for a window after delivery. It captures the calls, searches and walk-ins that codes miss, because it never depended on how the customer responded.

Mechanically it is the same job as reconciling any lead source against closed sales. One file says who was reached; the other says who paid. The difficulty is not the concept but the keys. A mailing list is names and postal addresses; a sales system is usually names, phone numbers and email addresses, with a billing or service address that may or may not be the one mailed. The match is only as good as the fields the two files have in common, and how carefully they are normalised before anybody compares them.

1. Fix the window before you look at the results, starting from when the pieces arrived — not the day they were posted, and not the day after a good week.
2. Normalise both files the same way: casing, punctuation, apartment and unit formats, abbreviations such as Street and St, and phone numbers to one international format.
3. Match on the strongest shared key first. A house list with phone numbers and emails matches cleanly; a rented list of postal addresses has to match on address and surname.
4. Count every buyer in the window once, however many pieces they received.
5. Apply exactly the same window and the same rules to the holdout group. A comparison between groups measured differently is two numbers, not a result.

Matchback answers “did the people we mailed buy” completely. It does not answer “did they buy because we mailed them”, and the gap between those two questions is where most mail reporting goes wrong.

Addresses decay, and the match has to survive it

A mailing list starts going out of date the day it is cleaned. People move, a household's billing address differs from the one mailed, and unit numbers are written five different ways. Matching on address alone loses buyers to formatting and removals, which quietly lowers the matchback count.

Two practical consequences for matchback. First, match on the address the piece was actually delivered to, after the move update, not the one in the original file — otherwise a household that moved in August and bought in September is a miss. Second, wherever the sales file has a phone number or email address for the customer, and the mail file can be tied to one, match on that instead. A phone number survives a move far better than a street address, and a normalised email survives it completely.

Formatting deserves more suspicion than it usually gets. “Apt 4B”, “#4B”, “Unit 4B” and “4B” are the same door. “North Main Street” and “N Main St” are the same road. A matchback that compares raw strings will report those as different households, and the error only ever runs one way: towards fewer matches, and a mailing that looks weaker than it was.

The window starts at the door, not at the post office

A matchback window should open when the mail reached homes, not when it was handed to the Postal Service. Marketing mail can take days to over a week to deliver, and a window opened on the drop date both starts too early and ends too early.

The error from a drop-date window is not symmetrical, which is why it matters. Purchases in the first few days are attributed to a mailing that had not arrived, inflating the early count with unrelated sales; purchases at the far end are cut off by a window that closed a week before it should have. On a sixty-day window the distortion is modest. On a fourteen-day window for a time-limited offer, it can be most of the answer.

Informed Delivery adds a click, not a sale

Informed Delivery gives a mail piece a digital twin: a preview in the recipient's email or app, with a clickable image beneath it. It adds an impression and a measurable click to the campaign, and it does not tell you who bought.

It is worth running. A clickable image under a preview is a second chance at a response for no extra postage, and the click data is useful for comparing creative. The risk is the familiar one: a campaign report full of impressions and click-through rates starts to feel like an answer, and the question it was supposed to serve — did this mailing produce customers who would not otherwise have bought — is left exactly where it was.

Matchback overclaims unless you hold some names back

A holdout group is a random slice of the list you deliberately do not mail. Its purchase rate over the same window is the baseline — what the mailed group would have done anyway — and only the difference above that baseline is what the mail caused.

Back to the heating company. Its list had 44,000 names, and before the mailing somebody set 4,000 aside at random and did not mail them. Over the same sixty days, measured the same way, 27 of the 4,000 bought: 0.68%. The mailed group bought at 1.03%. Applied to 40,000 mailed households, the holdout rate predicts 270 purchases with no mailing at all. The mailing produced the rest.

142 is the honest point estimate, and it deserves a range beside it, because 27 purchases is a small count. The natural variation on a count that size is roughly plus or minus ten, which moves the baseline between about 170 and 370 and the incremental effect between about 40 and 240. That range is wide, and saying so is not a weakness of the method. It is the method telling you that a 4,000-name holdout can distinguish “the mail worked” from “the mail did nothing”, and cannot yet tell you precisely how well.

The company kept mailing. It also doubled the holdout on the next drop, so the range would narrow, and stopped reporting promo code redemptions as the result — they now appear as a response-channel breakdown, which is what they are.

Matching a mailing against sales is a reconciliation, and it is the reconciliation CloseRev runs. It takes a closed-sales export and a source file, normalises phone numbers and email addresses, matches at the confidence each piece of evidence supports, and shows what it could not match as Direct / Unknown. It matches on phone and email, not on postal addresses — so it suits a house list or any mail file your CRM holds contact details for, run once for the mailed group and once for the holdout, and it does not suit a rented list of addresses alone or Every Door Direct Mail.

Promo codes say the mail barely worked, matchback says it worked brilliantly, and both are describing the same mailing. The holdout is the only way to find out which one is less wrong, and it has to be set aside before the mail goes out.

What to report after a mailing

Report the incremental customers with their range, the matchback count beside it for context, and response devices as a breakdown of how people responded — never as the result. Cost per incremental customer is the number that belongs in a budget decision.

The fifth line is where the codes belong, and it is useful there. If most incremental buyers phoned, the card should carry the number larger than the logo. If a large share arrived through branded search, the search team needs to know that part of its good month was a postcard. That is information a coupon count can give. What it cannot give, and was never able to, is the verdict on whether the mailing paid for itself.

The best-measured mailing is not the one with the cleverest response device. It is the one with a holdout set aside before printing and a sales file matched honestly afterwards.

Questions people actually ask

What is direct mail matchback?

Matchback compares the list of people you mailed against the list of people who bought within a set window after the mail arrived, and counts the overlap. It needs no code, coupon or special phone number, because the mail file already records exactly who received the piece. Its weakness is that it counts everyone on the list who bought, including the customers who would have bought without the mailing.

Why do promo codes undercount direct mail response?

Because most people who act on a mail piece do not use the code. They phone the office, search for the company by name, type the web address from memory, or keep the card on the fridge for three weeks and lose it. A code measures the people who remembered it at the moment of purchase, which is a small and unrepresentative share of the people the mail moved.

What is a holdout group in direct mail?

A random sample of the list that you deliberately do not mail. Because the holdout is chosen at random, its purchase rate over the same window is what the mailed group would have done without the mail. The difference between the two rates, applied to the mailed group, is the incremental effect of the mailing: the part of matchback's number the mail actually caused.

How long should a direct mail matchback window be?

Long enough to cover the purchase cycle and short enough that unrelated purchases do not swamp it — commonly 30 to 90 days, measured from when the piece arrived rather than when it was posted. A heating replacement quoted in October and signed in November needs the longer end; a restaurant offer needs the shorter. Whatever the window, use the same one for the holdout group, or the comparison means nothing.

Can you attribute Every Door Direct Mail?

Not to individuals. Every Door Direct Mail is addressed to “Postal Customer” on the routes you choose, so there is no list of recipients to match against. You can compare sales in the mailed routes or ZIP codes with similar routes you did not mail, over the same period, which is a geographic holdout rather than a person-level one — coarser, and still far better than counting redeemed codes.

Does USPS Informed Delivery tell you who bought from your mail piece?

No. Informed Delivery shows opted-in recipients a preview of their mail with an optional clickable image beneath it, and campaign data tells you about those notifications and clicks. It adds a digital touch and a measurable click to a physical piece, but a click on a notification is not a purchase, and the sale still has to be matched against your own records.

Found this useful?

• Share on X
• Share on LinkedIn
• Share on Reddit
• Share on Email