Predicting Bounces Before You Hit Send
What if you could know your bounce rate before you actually sent the campaign? Not a gut feeling or a hopeful guess, but a data-driven prediction based on the composition of your list and historical verification-to-bounce correlations.
This is not hypothetical. If you are tracking the right data, you can build a bounce prediction model that is accurate enough to tell you whether a specific campaign is safe to send or likely to blow past your bounce threshold. More importantly, it tells you which segments to re-verify or exclude before sending.
The Core Concept: Verification Categories as Bounce Predictors
Email verification tools classify addresses into several categories: valid, invalid, catch-all, unknown, risky, and disposable. Each category has a different probability of producing a bounce when you actually send to it.
Valid addresses have the lowest bounce probability, typically under 0.5%. These are addresses where the verification tool confirmed the mailbox exists through a successful SMTP handshake. Even here, there is a small chance of bouncing because the mailbox could have been deleted between verification and sending time.
Invalid addresses have a near-100% bounce probability. These should never be in your send list at all. If they are showing up, your verification pipeline has a gap somewhere.
Catch-all addresses are where prediction gets interesting. When a standard verification tool labels an address as catch-all, you know the domain accepts all mail at the SMTP level. But some of those addresses are real mailboxes and some are not. Without specialized catch-all verification, the bounce probability for this segment ranges from 5% to 25% depending on the domain mix. With specialized catch-all verification from a tool like CatchallVerifier, the bounce probability for verified catch-all addresses drops to under 2%.
Unknown addresses are ones where the verification tool could not get a definitive answer. This happens when the receiving server times out, is temporarily down, or uses greylisting. Bounce probability for this category: 10-30% depending on the reason for the unknown status.
Risky addresses include role-based addresses (info@, sales@, support@), addresses with high complaint history, or addresses at domains with known deliverability issues. Bounce probability sits around 3-8%, but the spam complaint probability is what really makes these dangerous.
Disposable addresses from services like Mailinator or Guerrilla Mail have a time-dependent bounce probability. If the address was created very recently, it might still work. If it is more than a few hours old, bounce probability climbs toward 50-90%.
Building Your Prediction Model
The simplest useful prediction model takes the number of addresses in each verification category and multiplies by the expected bounce rate for that category, then adds them up.
Say you have a list of 10,000 addresses with this verification breakdown: 6,500 valid, 0 invalid (already removed), 2,500 catch-all (unresolved), 700 unknown, 300 risky.
Using industry-average bounce probabilities: Valid: 6,500 x 0.3% = 19.5 expected bounces. Catch-all (unresolved): 2,500 x 12% = 300 expected bounces. Unknown: 700 x 20% = 140 expected bounces. Risky: 300 x 5% = 15 expected bounces.
Total expected bounces: 474.5 out of 10,000. Expected bounce rate: 4.75%.
That is way above the 2% safe threshold. The prediction tells you this campaign will damage your reputation before you send a single email.
Now run the same calculation after adding specialized catch-all verification. The 2,500 catch-all addresses get resolved: 2,000 come back as verified deliverable, 500 come back as undeliverable and get removed. New list: 6,500 valid plus 2,000 verified catch-all plus 700 unknown plus 300 risky. Total: 9,500.
Valid: 6,500 x 0.3% = 19.5 bounces. Verified catch-all: 2,000 x 1.5% = 30 bounces. Unknown: 700 x 20% = 140 bounces. Risky: 300 x 5% = 15 bounces.
Total expected bounces: 204.5 out of 9,500. Expected bounce rate: 2.15%.
Better, but still above 2%. The prediction model now shows that the unknown segment is your biggest remaining risk factor. You could exclude the unknown segment entirely, bringing your list to 8,800 addresses with an expected bounce rate of 64.5 / 8,800 = 0.73%. That is solidly in the safe zone.
Calibrating With Your Own Data
The industry-average bounce probabilities above are reasonable starting points. But your actual bounce probabilities will differ based on your specific sending profile, data sources, and target audience.
To calibrate your model, track the actual bounce rate for each verification category across your last 10-20 campaigns. Build a simple table with columns for verification category, predicted bounce probability, actual bounce count, actual bounce rate, and the difference between predicted and actual.
After a few campaigns, you will see where the generic probabilities need adjustment for your specific situation. Maybe your catch-all addresses bounce at 8% instead of 12% because your prospect list skews toward well-maintained enterprise domains. Maybe your unknown addresses bounce at 30% instead of 20% because you are targeting smaller companies with less reliable email infrastructure.
Update your model with your real numbers. Over time, your predictions become increasingly accurate because they reflect your actual sending environment rather than averages across all senders.
Prediction by Domain Provider
Adding a domain provider dimension makes your prediction even more accurate. Bounce probabilities differ meaningfully by receiving provider.
Gmail domains (@gmail.com and Google Workspace custom domains): Gmail is relatively predictable. Valid addresses rarely bounce. Catch-all is less common on Google Workspace, but when it occurs, bounce rates tend to be moderate at 5-10% for unresolved catch-all.
Microsoft domains (@outlook.com, @hotmail.com, custom Microsoft 365 domains): Microsoft is trickier. The 26.77% B2B inbox placement rate means many emails that do not technically bounce still fail to reach the inbox. Microsoft is also more likely to accept an email and then silently filter it, which means your measured bounce rate can underestimate your actual deliverability problems on Microsoft domains.
Corporate custom domains: This is where catch-all dominates. Corporate domains have the highest catch-all rates (40% or more for enterprise) and the most variable bounce behavior. Your prediction model should give extra weight to the catch-all segment when the list is heavy on corporate domains.
Decision Framework Based on Predictions
Once you have a predicted bounce rate for a campaign, use this decision framework.
Predicted bounce rate under 1%: Send with confidence. Your list composition is solid and the expected bounce rate is well within safe territory. Monitor actual results to keep your model calibrated.
Predicted bounce rate 1-2%: Proceed with caution. Consider whether you can reduce the risk by excluding your highest-risk segments like unknown and unresolved catch-all addresses. If the campaign is important enough to justify the risk, send but monitor bounce rates in real time and pause if actuals exceed predictions by more than 50%.
Predicted bounce rate 2-3%: Take action before sending. Run additional verification on your catch-all segment. Exclude your unknown segment. Re-verify any addresses that have not been checked in the last 30 days. Re-run the prediction after these steps.
Predicted bounce rate above 3%: Do not send. This list needs significant work before it is safe to mail. Run complete verification including catch-all resolution, remove all unknown and risky addresses, and consider whether your data sources need to be replaced entirely.
Time-Based Decay Adjustment
Your prediction model should also account for time since verification. Email data decays at roughly 2-3% per month according to multiple industry sources. An address verified 6 months ago has a materially higher bounce probability than one verified yesterday.
Add a decay multiplier to your base prediction: addresses verified within the last 30 days use your base bounce probabilities. Addresses verified 30-90 days ago get a 1.5x multiplier. Addresses verified 90-180 days ago get a 2x multiplier. Addresses verified more than 180 days ago get a 3x multiplier.
This decay adjustment often reveals that a list is riskier than it appears. A list that was perfectly clean six months ago might now have an effective predicted bounce rate of 3-4% after applying decay multipliers. This is why regular re-verification matters, and your prediction model makes the case with concrete numbers rather than abstract best practice recommendations.
Automating Predictions as a Pre-Send Check
Once your model is calibrated, automate it. Before any campaign goes out, run the prediction automatically. If the predicted bounce rate exceeds your threshold, block the send and flag it for review.
This can be as simple as a spreadsheet that pulls your list composition data and applies the prediction formula. Or it can be integrated into your sending workflow via API: before sending, query your verification data, calculate the prediction, and proceed or block based on the result.
The goal is to make bounce prediction a standard pre-flight check, not an occasional exercise. Every pilot runs a checklist before takeoff. Every email sender should run a bounce prediction before hitting send. The data is already sitting in your verification results. You just need to put it to work.
