Back to Blog

How to clean duplicate accounts out of your SaaS metrics

Duplicate accounts bend activation, CAC and churn in different directions. How to correct each number, and why the quick filters leave it wrong.

Sanket DattaFounder & CEO

On Friday you told your co-founder this would be the easiest investor update you'd ever written. This morning, the first Tuesday of October, the email from your product analytics tool says your invoice is $1,260.00. Last month it was $1,140. You pay $0.04 per tracked account, so the extra $120 is 3,000 new accounts. That's the number you were about to celebrate in Thursday's investor update.

You open the list to see who those 3,000 people are. Near the top, one Gmail mailbox appears four times: once with dots, once without, and twice with a plus sign and a word after the name. Gmail delivers all four to the same inbox. You stop writing the update.

This is a scenario, so check the arithmetic as you read. At 9:30 you pull 400 accounts at random from the quarter's 9,000 and read each one. A minute per account is 400 minutes, which takes you to 4:10 p.m. Of the 400, 140 belong to someone who already had an account. That's 35%, and 35% of 9,000 is 3,150 accounts you reported as new people. Searching for duplicate accounts SaaS metrics at 4:11 p.m. is not how you pictured this day.

Now read the update again, line by line.

  • Signups. It says 9,000. About 5,850 people signed up.
  • Activation. It says 30%, or 2,700 activated accounts. Half of those were repeat accounts back for more free credits, which leaves 1,350 of 5,850 people, or 23.1%.
  • CAC. You spent $36,000 and counted 180 new customers, so $200 each. You read all 180, and 30 are former customers who came back on a new account for the first-month price. That's 150 new customers at $240. The 3,150 repeat accounts also used an average of $2.40 in free credits, $7,560 that bought you nobody. ($36,000 + $7,560) ÷ 150 = $290.40.
  • Churn. It says 72 cancellations from 600 paying customers, or 12% for the quarter. Those same 30 people are in there, so you really lost 42, or 7%.
Before and after, illustrative (%)Activation wasActivation nowChurn wasChurn now05101520253035Percent
Before and after, illustrative (%)Activation wasActivation nowChurn wasChurn now020Percent

Each of those numbers has already made a decision for you. At $49 a month and 80% margin, a customer earns you $39.20 a month, so a $200 CAC pays back in 5.1 months. That 5.1 is why you hired a growth marketer in August at $9,000 a month. The real payback is 7.4 months ($290.40 ÷ $39.20). With $420,000 in the bank, that hire moved monthly burn from $35,000 to $44,000 and your runway from 12 months to 9.5.

The 12% cost you something else. Your one engineer spent six weeks on a new cancellation flow because of it, and the export feature your biggest customers keep asking for moved to next year. On Monday you told the team you'd test $39 instead of $49 in November. A $10 cut across 708 customers is $7,080 a month, or $84,960 a year, to fix a churn problem a little over half the size you thought.

And the invoice? If 35% holds across all 31,500 tracked accounts, $441 of it goes on 11,025 accounts that were never new people. That's $5,292 a year, the cheapest thing on this list.

It's 8:40 p.m. The update goes out Thursday. You can send a number you now know is wrong, or send a smaller one and explain why the last seven updates were probably wrong too.

Why duplicate accounts distort SaaS metrics

Your analytics tool counts accounts because an account ID is all it has. Give one person five IDs and you get five signups, five activation events and, when four of them go quiet, four lost users. Benefits are what drive it. If a trial or a pack of credits comes with each account, a new account is how you get another one.

The obvious fix is a haircut: 35% of accounts are duplicates, so take 35% off. That changes your signup count and nothing else, because (2,700 × 0.65) ÷ (9,000 × 0.65) is still 30%. Rates only move when duplicates behave differently from real people, and they do. In your sample, 60 of the 140 repeat accounts activated (42.9%), against 60 of the 260 new people (23.1%). Someone back for more free credits already knows where the button is.

Activation got worse while churn got better, so each metric needs its own correction.

How to correct your SaaS metrics for duplicate accounts

Save what you reported. Copy every number as it stands, with its date. You'll need the before column.

Write down what a duplicate is. The same person taking a benefit again. A colleague at the same company isn't one, and someone returning after a long gap may not be either, so a written one-account policy helps. Count anything you can't decide as a real person, so your correction stays conservative.

Label a random sample. Random means random, because the newest 400 accounts will mislead you. Have one person read each account and its support history, then record new person, repeat or unsure. At 35%, a sample of 400 lands within about 5 points either way, so your honest figure is 30% to 40%.

Read every paying account. New customers and cancellations are small groups, so skip the sample. Here that's 180 + 72 = 252 accounts, a little over four hours.

Apply one correction per metric, using the table below. To scale the sample up, multiply by 22.5 (9,000 ÷ 400), so 140 becomes 3,150 and 60 becomes 1,350.

Publish both columns. Keep the reported number, add the corrected one and say how you got it.

MetricWhat duplicates doCorrected asReportedCorrected
SignupsCount one person many timesAccounts minus repeats9,0005,850
ActivationRepeats activate more than new peopleActivated people ÷ people30%23.1%
CACHide free cost, count old buyers as newFormula below$200$290.40
ChurnCount a moved customer as lost(Lost minus returned) ÷ starting customers12%7%
Corrected CAC=Acquisition spend+Free cost of repeat accountsNew customers−Returning buyers\text{Corrected CAC} = \frac{\text{Acquisition spend} + \text{Free cost of repeat accounts}}{\text{New customers} - \text{Returning buyers}}

You may book free-plan cost somewhere other than CAC. That's fine, as long as you pick one place and know what a free user really costs you. One figure never moves: 600 + 180 minus 72 and 600 + 150 minus 42 both give 708 customers. Revenue and cash were right all along. The split between new, returned and lost customers wasn't.

Save reported numbersDefine a duplicateLabel 400random accountsRead everypaying accountCorrect signupsand activationCorrect CACand churnPublish before and after
Save reported numbersDefine a duplicateLabel 400random accountsRead everypaying accountCorrect signupsand activationCorrect CACand churnPublish before and after

Don't delete the accounts

Keep your labels in a separate table. Some of those accounts will turn out to be real, and if you do combine any, merge them with care.

Quick filters for duplicate accounts, and how they break

A day of reading accounts every quarter makes you want a filter instead. Each of these does something real, and none of them tells you whether two accounts are one person.

FilterWhat it removesWhat stays inWho it wrongly removesUpkeep
Email verificationTypos, dead addressesAny real second inboxPeople whose code went to spamDeliverability work
Unique emailsOne address used twiceEvery other addressAlmost nobodyVery little
Disposable-domain blocksDomains on your listNew domains, normal mailboxesPeople using privacy forwardingA list that's never finished
Gmail dots and plus addressingOne Gmail inbox written several waysA second real mailboxPeople who sort their mail with plus addressesUpkeep as mail services change
CAPTCHASimple scriptsPeople signing up by handScreen reader and privacy tool usersFees and lost signups
Phone or card requirementCasual second accountsAnyone with another number or cardStudents, people who want to try firstPer-text costs, fewer signups
One per IP addressRepeats on one connectionAny other networkOffices, campuses, householdsExceptions and tickets
VPN blocksKnown VPN rangesEverything off the listStaff on company VPNsAnother list
Rate limitsBurstsSlow, spread-out signupsReal users in a busy hourTuning
Usage capsOne account overrunningEach new account's fresh capAnyone at the capPicking the cutoff
Shorter trialsSome cost per accountMore accountsEvery honest userOld rates stop comparing

Filter disposable domains out of your dashboard and signups fall, the chart looks cleaner, and a repeat account on an ordinary mailbox stays in. You now have a number that looks corrected and isn't. Add a one-per-IP rule and your best customer's office can't add a second person, which shows up next month as an activation dip you'll spend a week explaining.

Stack CAPTCHA, a phone step and a card, and you've changed who reaches your funnel, so this quarter can't be compared with the last. Manual review is the one method here that holds up, and it doesn't scale. Labeling all 9,000 accounts at a minute each is 150 hours, for one quarter that's already over.

Keeping duplicate accounts out of next quarter's numbers

Rerun the review each quarter and give the abuse rate its own line on your metrics sheet. Then look at where the error starts. Every number above went wrong at the moment someone received a benefit they'd already had, and a dashboard filter can't reach that moment.

AbuseZero works at that moment. Your domain list knows the domains on it, while AbuseZero checks mailbox aliases, disposable domains and mail routing. Your IP rule knows one connection, while AbuseZero sees beyond an IP address with VPN, hosting, network reputation, Tor and proxy intelligence. It recognizes repeat devices across new accounts, even when email addresses keep changing, so creating a fresh account isn't enough to erase previous activity. It flags abuse that single checks miss. Its multi-layer AI risk analysis examines 140,000+ signals and detects linked claims. AbuseZero uses a proprietary algorithm, with AI that checks the claim again.

Your server makes one API call before the benefit goes out and gets a clear allow or deny, under rules you control. You don't have to build your own abuse prevention system. Observe mode shows you what would be denied before anyone is blocked. The quickstart protects your first benefit.

Frequently asked questions

Do duplicate accounts change MRR?

No. They change how you split it between new, returned and lost customers, and that split drives your CAC and churn.

How big should my sample be?

At around 35%, 100 accounts gets you within about 9 points and 400 within about 5. Small rates need larger samples.

Should I tell investors?

Send both numbers with your method, and say plainly that revenue and cash are unchanged.

Count people next quarter

Keep reading

Copying your prompt…

Copied the to your clipboard. Give it to your coding agent & it will handle the rest.