The quick read
  • Promo codes, scans, and tracking numbers give you a floor, never the whole picture.
  • Matchback against sales records gives you a ceiling that includes sales that would have happened anyway.
  • A random holdout of five to ten percent shows the true incremental lift in between.
  • Wait four to eight weeks and watch branded search, because mail sends people to search.
  • Save the mailed file and version assignments so you can answer questions later.

The tracking problem nobody warned you about

Digital marketing spoiled a generation of marketers. Click an ad, land on a page, buy something, and a tidy chain of tags records every step. Then direct mail arrives and the chain breaks. A person reads a postcard on Thursday, forgets about it, remembers the brand on Saturday, searches by name, and buys on Monday. The sale shows up as organic search. The mailer gets zero credit. The next quarterly review concludes that mail does not work, and the budget moves to a channel that is better at taking credit than at creating demand.

This is the central attribution problem in direct mail. The piece does its job in the physical world, and the buyer completes the purchase in the digital one, often through a different door. UTM parameters, the tags that make web campaigns traceable, depend on someone clicking a link. A postcard has no link. People do not type a long web address with query strings from a card. They scan a code, or they search, or they walk in, or they call. Some respond within an hour, and some a month later.

The good news is that direct mail is among the most measurable offline channels, because you control who gets the piece and when. You know the exact names and addresses. You know the in home window. That control makes methods possible that digital channels cannot match, such as clean holdout tests. This guide walks through what works, from simple to sophisticated, with the strengths and weaknesses of each, and ends with a practical stack you can set up in a week.

What you are actually trying to measure

Before choosing tools, be clear on the question. There are three different things people mean by attribution, and they need different methods.

Direct response. Which sales came from people who used the response mechanism on the piece, such as a code, a QR scan, or a dedicated phone number? This is the easiest to count and always an undercount, because many people respond without using the mechanism.

Matched response. Which of your mailed recipients later bought, whether or not they used a code? This requires matching your mailing file to your sales records by name and address. It is much more complete, but it still includes people who would have bought anyway.

Incremental lift. How many sales happened because of the mailing that would not have happened without it? This is the true value of the campaign, and the only measure that survives a skeptical finance team. It requires a comparison group.

Most small businesses stop at the first. Better teams do the second. The best do all three, and use the third to calibrate how much to trust the first two.

Method one: unique promo codes

The oldest method remains useful. Print a code on the piece, and tell the customer to use it at checkout, at the counter, or on the phone. When the code comes in, you know the source.

What makes it work. The code must be worth using. A real discount or bonus gets entered. A code that gives nothing gets ignored, and your data gets worse. Make the code memorable and simple to type, and make it unique to the campaign, or better, unique to the segment or version. A code like SPRING-A and SPRING-B tells you which version of the piece did better in a split test.

What goes wrong. Codes leak. A good one ends up on coupon sites within a week. Customers who did not receive the piece find it and use it, and your attribution inflates. Mitigate this by using single use codes tied to a recipient, which variable data printing can generate for each name. You can also limit codes to a start and end date and a cart minimum.

The undercount. Many buyers never enter the code. They forget, they decide at the register that the discount is not worth the effort, or they buy through a channel where the code does not work. Expect the code count to capture somewhere between a fraction and half of the true response. Do not treat it as the total.

Method two: QR codes and personalized landing pages

Phones read QR codes natively, and people are now used to scanning them. A code on a postcard is the closest direct mail gets to a click. It is also the best bridge to your digital analytics.

Point to a unique URL. Each campaign or segment should have its own address, with the usual tracking tags attached inside the destination so that your analytics sees the source. The code itself can encode a short redirect so you can change the destination later without reprinting. Check that the redirect is fast and that the page works well on a phone, because nearly all scans come from one.

Personalize the code. With variable data, every piece can carry a different code that identifies the recipient. A scan then ties a specific person to a visit, even if they never fill out a form. Use this carefully. People are sensitive about being tracked, so say plainly what the page does with their information and make the page useful on its own.

Make the scan worth it. A code that leads to the home page is wasted. Send people to a page that continues what the postcard started: the same offer, the same image, the same headline, and one action to take. A scan without a reason is a curiosity. A scan with a reason is a lead.

Know the limits. Scan rates are low compared with the number of people who see the piece. Many recipients who intend to buy will search the name instead. Treat the scan count as a floor, not a measure of interest. Print also a short, memorable web address as a fallback for people who prefer to type, and watch traffic to it.

Method three: tracking phone numbers and vanity addresses

For businesses that sell by phone, such as home services, legal, medical, and B2B, a unique phone number printed on the piece is still the strongest single signal. Call tracking services assign a number to each campaign or segment, forward the call to your regular line, and record the source, the duration, and often a recording. You can see how many calls each version of the piece produced, how many turned into booked jobs, and how long it took.

The same logic applies to vanity web addresses. A short address that exists only on the mailer, such as a campaign specific page on your domain, tells you that anyone who visits it by typing probably saw the piece. It is not perfect, because some people find it by sharing, but it is a clean directional signal with no tagging required.

Train your staff to ask. Whoever answers the phone or greets a customer should have a simple question: how did you hear about us? Record the answer in a field, not in a free text note that no one will read. Self reported source is imperfect, because people forget and misremember, but it catches the walk ins and the people who did not use any of the mechanisms above. Pair it with the other methods.

Method four: matchback analysis

A matchback takes the list of people you mailed and compares it with your sales records afterward. Anyone who appears in both lists, within a defined window after the in home date, counts as a matched response, regardless of whether they used a code.

How it works. Gather the mailing file with names and addresses. Gather orders for the period, with the billing and shipping names and addresses. Standardize both so that "123 Main Street Apt 4" and "123 Main St #4" match. Then match on address first, and use name as a secondary check. Report the matches by day after the in home date, to draw the response curve.

What it tells you. Matchback captures the shopper who read the postcard on Thursday and ordered on Monday through a search. It also shows the timing, which helps you pick the right measurement window. Many campaigns produce responses for four to eight weeks, and the tail is often longer than people expect.

What it cannot tell you. It credits the mailing with every sale from a recipient, including those who would have bought anyway. For a house list of recent customers, a large share of matches may be sales that would have happened regardless. For a cold prospect list, the match is closer to true response. Matchback is a good ceiling and a poor measure of lift on its own.

Practical notes. Do the matching yourself or ask a data service. Keep a clean record of the mailed file with an identifier for each recipient and the version of the piece. Without a stored record of who got what, you cannot do any of this after the fact.

Method five: the holdout test

This is the method that settles arguments. Before you mail, randomly remove a group of eligible recipients, usually five to ten percent, and do not send them anything. Everything else about the two groups is the same: same list, same period, same email and ad exposure. After the campaign, compare the purchase rate of the mailed group with the purchase rate of the holdout. The difference is the lift.

An example. You mail 9,000 customers and hold out 1,000. Over eight weeks, 540 of the mailed group buy, which is 6.0 percent. In the holdout, 40 buy, which is 4.0 percent. The lift is 2.0 points. Applied to the mailed group, that is about 180 incremental purchases. If your matchback had counted all 540, you would have overstated the effect by three times. If your promo code count had captured 150, you would have understated it by a little. The holdout gives you the real number.

Make it random. Do not choose the holdout by hand. A hand picked group will differ from the mailed group in some way that distorts the result. Use a random draw, and confirm that the two groups look alike on past purchase value and recency before the campaign starts.

Size matters. A holdout that is too small gives noisy results. If you expect small differences, you need more people in each group. As a rough rule, if the purchase rate is a few percent and you hope to detect a lift of one or two points, you need thousands in each arm to be confident, not hundreds. If your file is small, run the test across several drops and combine the results.

The cost of holding out. Skipping a few hundred customers has a real price, since you forgo some sales. Frame it as the cost of knowing. A campaign that runs for years on a wrong assumption costs far more than a holdout.

Method six: geographic and market tests

When you cannot hold out individuals, hold out places. Pick two sets of similar markets, mail one set and not the other, and compare sales over time. This works for retailers and local services, and for saturation campaigns where every address on a route gets a piece. It is cruder than a person level holdout, because markets differ in ways that are hard to match, but it measures something nothing else can: the total effect on a place, including word of mouth, walk in traffic, and searches by name.

To keep the test fair, match markets on population, past sales, season, and competitor activity. Use a before and after comparison: how did sales change in the mailed markets relative to the control markets, compared with the same period last year? Run it long enough to cover the response window. One week of data is nothing.

Be careful about spillover. If your mailing area and control area share media, a local event, or a border, some of the effect leaks. Choose markets that are separated by something, such as distance or a distinct media market.

Because mail makes people go online and search, the clearest digital footprint of a mailing is a bump in branded search and direct visits. Track weekly branded search volume and direct sessions, and mark every in home date on the chart. A good drop produces a visible rise that begins two to three days after delivery and fades over a few weeks.

This does not prove causation on its own, but it adds weight to the other evidence, and it explains why digital reports often show a mysterious jump in organic traffic after a mailing. Share the chart with whoever manages your paid search and organic reports so they stop attributing the surge to something else. It also guards against a subtle problem: paid search ads on your own brand name can steal credit for a sale that mail created. If your branded search conversions rise after a drop, give mail some of the credit.

A practical attribution stack for a small team

You do not need everything at once. Here is a stack, in the order we would build it, that a small team can set up in a week and extend over time.

Layer one, always on. Every piece carries a campaign specific code or offer, a QR code to a campaign page, and a tracking phone number if you take calls. Your team asks every customer how they heard about you and records the answer in a set list of choices that includes "mailer."

Layer two, after every campaign. Run a matchback within sixty to ninety days of the in home date. Save the mailed file and results in one place so that you can compare campaigns over time.

Layer three, for campaigns above a certain budget. Hold out a random five to ten percent. This is the one that tells you whether the channel is worth the money. Do it on at least one campaign each quarter, and use it to calibrate the cheaper methods. If your codes typically capture a third of the lift measured by holdouts, you can apply that factor to later campaigns where you did not run one.

Layer four, quarterly. Review branded search, direct traffic, and in store or phone volume against the calendar of drops. Look for patterns you did not expect.

Together these give you a range, not a single number. Your promo code count is a floor. Your matchback is a ceiling. Your holdout lift sits between them and is the figure to use for budgeting. A range is more honest than a false precise number, and a finance team that sees the method behind it will trust it more.

Mistakes that make mail look worse than it is

Measuring too early. If you check results after two weeks, you will miss the late responders. Wait at least four to eight weeks, and plot responses by week.

Using last click for everything. A last click report gives mail zero credit for any sale that finished online. It also gives all the credit to whichever paid or organic link was last. Use a model that gives weight to earlier touches, or, better, use lift tests.

Mailing and promoting at the same time. If you launch a big email push, a sale, and a mailing in the same week, you cannot untangle them. Stagger them if you want to learn something, or use a holdout to isolate each.

Forgetting the control. Without a control group, every number is a guess dressed up as a result.

Dropping the record. Save the mailed file, the version assignments, the in home dates, and the codes. Teams that lose this data cannot answer a question six months later, and they re-learn the same lessons every year.

Privacy, consent, and doing this honestly

Matching mailing lists to sales records uses personal information, and privacy rules vary by state and country. Keep the use of customer data consistent with what you told people when you collected it. Provide a simple way to opt out of mail, and apply it everywhere. If you use personalized QR codes or links, say on the landing page what is recorded and why. When in doubt, ask a qualified advisor. Good measurement and respect for the recipient do not conflict. The campaigns that last are the ones that people do not mind receiving.

How to report the results

When you present a campaign result, show the range and explain it. Report mailed quantity, delivered estimate, all in cost, direct response count by method, matched response count, and measured lift if you ran a holdout. Then show cost per incremental customer and the profit based return. Include what you would change.

If you are reporting to a client or a boss, tell them in advance what you will measure and how long it will take, so they do not judge the campaign in week two. Agree on the method before the money is spent. If you resell print and mail for clients, build the tracking mechanisms into the job from the start, with unique codes and phone numbers on each version, and keep records for them. The Trade Print program is built for resellers who want production they can rely on, so that the only variable left in the report is the marketing.

For the cost side of the equation, our direct mail ROI math guide shows how to set the break even rate before you mail, and the guide to triggered mail covers how to build holdouts into an automated program. If you are choosing a format, start from the postcard and direct mail pages.

Frequently asked

Why do UTMs not work for direct mail?

UTM parameters only record a click on a tagged link. A postcard has no link, and people respond by searching the brand name, scanning a code, calling, or walking in. Many of those sales appear in analytics as organic search or direct traffic.

Are QR codes reliable for tracking?

They are useful but they give a floor. Many people who intend to buy will search your name instead of scanning. Use a unique code or landing page per campaign or segment, and treat the scan count as the minimum measure of response.

What is a matchback analysis?

You compare the list of people you mailed with your sales records after the in home date. Anyone who appears in both within the response window counts as a match, whether or not they used a code. It is more complete than codes, but it includes sales that would have happened anyway.

How large should a holdout group be?

Commonly five to ten percent of the eligible list, chosen at random. It also has to be large enough to detect the lift you hope to see. With purchase rates of a few percent, that usually means thousands of people in each group, so small lists should combine several drops.

How long should I wait before judging a mailing?

At least four to eight weeks. Responses to direct mail arrive over time, and many come well after the in home date. Measure at thirty, sixty, and ninety days and note when the response curve flattens.

Ready to put this into production?

Apply for the Trade Print program to get published pricing, proofing, and blind shipping on your direct mail and print jobs.

Apply to the Trade Print program