Tier 1 to 5 confidence scoring on email opens
Someone searching email open confidence scoring is typically an AE or sales-ops lead who has noticed that their current tracker's open rate doesn't match reality and is looking for a tracker that filters Apple MPP and scanner noise automatically.
Raw email open rate in 2026 is unreliable. Apple Mail Privacy Protection pre-fetches every tracking pixel on Apple servers before the recipient opens the email. Gmail's image proxy can register the same pixel as multiple opens. Corporate email scanners (Mimecast, Proofpoint, Microsoft Defender for Office 365) pre-fetch every link and image. A reported 70 percent open rate typically reflects a 25 to 35 percent real human-read rate on a typical B2B list.
Outsolvi's confidence scoring is the structural fix. Every open is graded from Tier 1 (high-confidence human, 80 to 100 percent confidence) to Tier 5 (bot or scanner, 0 to 20 percent confidence) based on IP block, User-Agent string, and timing signature. Anything below 25 percent confidence is excluded from the open count entirely. The remaining count actually correlates with buyer behaviour.
What it does
Confidence scoring grades every email open event with a confidence value from 0 to 100 percent based on whether the open is likely a human read or a machine pre-fetch. The grading runs in real-time on each open event and is exposed to the rep in the dashboard as a tier label (Tier 1 through Tier 5) plus the underlying confidence percentage.
Why it matters in 2026
- Apple Mail accounts for roughly 58 percent of global email-client market share per Litmus in early 2026. With MPP enabled (default for most users), every Apple Mail recipient generates a pre-fetched open that's not a real human read.
- Corporate scanners pre-fetch every link and image on every email for malware scanning. Microsoft Defender for Office 365, Mimecast, Proofpoint, and similar gateways are dominant on enterprise mail. A typical B2B list has 15-25 percent of recipients behind a corporate scanner.
- Without confidence scoring, the open rate is contaminated by 30-50 percent on most B2B lists. The rep cannot tell which opens are humans and follow-up routing runs on bad data.
- Reply rate is the cleaner secondary signal, but it lags — by the time reply rate drops, the campaign or messaging issue has already cost a quarter of pipeline conversion.
- Confidence scoring catches the problem at the open level, where the cost of bad data is highest (because opens are the most-watched and earliest-fired metric).
How it works
Each open event arrives at the Outsolvi tracking endpoint with metadata: the request IP, the User-Agent string, the request timing relative to email send, and headers like Accept-Language and Referer. The confidence model analyses these signals against patterns known to distinguish human reads from machine pre-fetches. Apple MPP relays have specific IP ranges and timing signatures (sub-second pre-fetch from Apple-attributable IPs with Apple User-Agent). Corporate scanners typically fire multiple link and image requests within a 3-5 second window from a single IP. Gmail proxy requests come from googleusercontent.com IP ranges with the Google bot User-Agent. Each combination produces a confidence score from 0-100 percent. The 25 percent floor excludes any open below that threshold from the count. Tier 1 (80-100 percent confidence) opens are surfaced to the rep with the highest priority. The model accuracy on B2B traffic is roughly 95-98 percent agreement with human-rated classifications on a held-out test set.
Outsolvi is the only tracker under $30 per user per month yearly that exposes Tier 1 to 5 confidence scoring to the rep in 2026. Mailtrack, Mixmax, Streak, Right Inbox, GMass, Saleshandy, Vocus, and most basic Gmail trackers count every pixel load as an open with no filtering exposed. Yesware Premium and HubSpot Sales Hub have stronger underlying filtering but do not expose the per-open confidence value to the rep — the dashboard shows a filtered open count without breakdown, so the rep cannot tell which opens are humans and which are filtered out. Cirrus Insight and Boomerang have basic User-Agent filtering only. The exposed-to-the-rep tier scoring is structurally rare in the category.
Use cases for Confidence Scoring
- AE sales prospecting where follow-up timing depends on knowing which opens are real human re-reads versus Apple MPP pre-fetches
- Proposal tracking where the canonical buying-window signal is multi-open at Tier 1 confidence within 48 hours of send
- Investor outreach during fundraising where the founder needs to know which VCs actually opened the deck versus which had MPP pre-fetch the pixel
- Customer Success engagement-velocity tracking where the renewal-risk signal depends on accurate open data on existing-customer communications
- Document tracking where confidence-scored opens plus click events on document links give a much more reliable engagement signal than either alone
Frequently asked questions
How accurate is the confidence scoring?+
Roughly 95-98 percent agreement with human-rated classifications on a held-out test set of B2B email traffic. False-positive rate (Tier 1 graded when the open was actually a machine pre-fetch) is under 2 percent. False-negative rate (Tier 4 or 5 graded when the open was actually a human read) is under 5 percent.
What is the 25 percent confidence floor?+
Any open event graded below 25 percent confidence is excluded from the open count entirely. Tier 4 (MPP pre-fetches, typically 15-20 percent confidence) and Tier 5 (scanner pre-fetches, 0-10 percent confidence) both fall under the floor. The floor is configurable in advanced settings but most teams keep the default.
Why don't other trackers expose confidence scoring?+
Two reasons typically. First, building and tuning a confidence model is non-trivial — it requires labeled training data, ongoing pattern analysis as Apple/Google/scanners change, and a willingness to show users that their previous-tool's open rates were inflated. Second, exposing confidence damages the vendor's reported open numbers — a tracker that reports a 30% open rate looks worse than one that reports 70%, even if the 30% is more accurate.
Does confidence scoring work on Outlook and Gmail equally?+
Yes. The scoring runs on the open event metadata regardless of which mail client the recipient is using. Apple MPP affects both Outlook-hosted and Gmail-hosted recipients (depends on the recipient's mail app, not the mailbox provider). Corporate scanners affect Microsoft 365 recipients heavily and some Gmail-hosted enterprise accounts. The model handles both ecosystems.
Will my reported open rate go down if I switch to Outsolvi?+
Yes, typically by 20-40 percent on Apple-heavy or scanner-heavy B2B lists. The drop is noise leaving the data, not real engagement disappearing. Reply rate is the cleaner side-by-side signal during a migration — if reply rate held while open rate dropped, the previous tool was reporting inflated numbers.
Can I see the confidence value for individual opens?+
Yes. Each open event in the Outsolvi dashboard shows the tier (Tier 1 through Tier 5) plus the underlying confidence percentage. Clicking the open expands to show the analysis (IP block, User-Agent signature, timing) that produced the score.
Free tools, articles, and glossary entries that share a topic with this feature.
Try Confidence Scoring for free
14-day free trial, no credit card. Confidence Scoring plus the full Outsolvi feature set including all the other features in this category.
Start 14-Day Free TrialWhere this feature is used
Use cases, workflow patterns, and personas that depend on Confidence Scoring.
Nate built Outsolvi after watching every email-tracking tool he had ever used lie to him about opens. Outsolvi runs Tier 1 to 5 confidence scoring on every open, native in Outlook and Gmail, so the number on the dashboard is one a rep can actually act on.
We update these pages when the underlying mechanics change — new mailbox-provider rules, new tracker behavior, new measurement gaps. The dates above are real revisions, not auto-touches.