Foundations · updated 2026-04-11

Signal Flow

The Signal Pipeline

Every email triggers a cascade of signals that flow between three parties: the sender, the mailbox provider, and the recipient. Understanding this flow is essential because deliverability is not decided by any single check — it is the cumulative result of signals evaluated at every stage of the pipeline.

Stage 1: The Sender Emits Signals

Before a message even reaches a mailbox provider, the sender has already broadcast several signals:

Infrastructure signals:

The IP address of the sending server and its reputation history
The hostname and whether reverse DNS resolves correctly
Whether the connection uses TLS encryption
The SMTP HELO/EHLO banner and whether it matches expectations

Authentication signals:

SPF record alignment — does the sending IP match the domain’s authorized senders?
DKIM signature — is the message body signed with a valid cryptographic key?
DMARC policy — does the domain tell providers what to do with unauthenticated mail?
ARC (Authenticated Received Chain) headers — for forwarded messages, is the chain of custody intact?

Message-level signals:

The From, Reply-To, and Return-Path headers and whether they align
Subject line patterns (urgency words, ALL CAPS, excessive punctuation)
HTML structure, text-to-image ratio, and code quality
URL domains in the message body and their reputation
Attachment types and sizes
List-Unsubscribe header presence and format
MIME structure and encoding

Each of these signals is evaluated independently. A sender with a perfect IP reputation but failing DKIM will still face scrutiny. The system is additive — more positive signals improve the overall score.

Stage 2: The Provider Evaluates

When the message arrives at the mailbox provider’s infrastructure, a multi-layer evaluation begins:

Connection filtering (MTA level): The provider’s mail transfer agent (MTA) checks the sending IP against known blocklists (Spamhaus, Barracuda, etc.), verifies rDNS, evaluates connection behavior (rate, concurrency, retry patterns), and performs initial reputation lookup. Messages from IPs with severe reputation issues may be rejected outright with a 5xx bounce.

Authentication verification: SPF, DKIM, and DMARC are checked. Results are recorded in Authentication-Results headers. Failures don’t always cause rejection — they lower the trust score that feeds into the next stage.

Content filtering: The message body, headers, and structure are analyzed. Modern providers use machine learning models trained on billions of messages. These models evaluate patterns rather than specific keywords. A message that structurally resembles spam (even with benign content) will score poorly.

Reputation integration: The provider combines real-time signals with historical data:

Domain reputation: How has this domain performed over weeks and months?
IP reputation: What is the track record of this IP address?
URL reputation: Are the links in this message associated with spam or phishing?
Sender cluster reputation: For shared IPs, what is the collective reputation?

Engagement prediction: This is where modern filtering diverges from legacy spam filters. Providers like Gmail use machine learning to predict how likely a specific recipient is to engage with this message. The prediction is based on:

The recipient’s past interactions with this sender
The recipient’s general engagement patterns (do they read promotions?)
Similar recipients’ behavior with similar messages

This means the same message from the same sender can receive different placement decisions for different recipients.

Stage 3: The Recipient Reacts

After placement, the recipient’s behavior generates new signals that feed back into the system:

Positive engagement signals:

Opening the message (especially within the first few hours)
Clicking links inside the message
Replying to the message
Moving the message from spam to inbox
Adding the sender to contacts
Forwarding the message to others

Negative engagement signals:

Marking the message as spam (the strongest negative signal)
Deleting without reading
Consistently ignoring messages from this sender
Unsubscribing (mildly negative — indicates declining interest)

Absence signals:

No interaction at all — over time, this signals declining relevance
Inbox blindness — messages are technically visible but never engaged with

These signals close the feedback loop and directly influence how the provider scores the sender’s next message.

The Feedback Loop in Detail

The signal flow creates a self-reinforcing cycle:

Sender sends → Provider scores → Recipient reacts → Provider updates score → Next send is scored differently

This loop has several important properties:

Momentum

Good performance builds momentum. When recipients consistently engage, the provider raises the sender’s trust score, which improves placement for the next campaign, which makes engagement more likely (because more messages reach the inbox), which further raises the score.

The reverse is equally true. Poor engagement leads to worse placement, which reduces the audience that can engage, which further depresses engagement metrics, which lowers the score.

This is why deliverability problems accelerate. A sender who ignores early warning signs (declining opens, rising soft bounces) can enter a downward spiral that becomes increasingly difficult to reverse.

Latency

Signals don’t update instantly. Different providers have different update cycles:

Gmail updates domain reputation relatively quickly (hours to days) and is highly responsive to engagement changes
Microsoft tends to be slower to update reputation scores and relies more heavily on complaint data from their Smart Network Data Services (SNDS)
Yahoo uses a blend of engagement and complaint data with moderate update speed

This latency means that the impact of a bad campaign isn’t felt immediately. You might send a campaign with a 2% complaint rate on Monday and not see placement degradation until Wednesday or Thursday. By then, you’ve potentially sent additional campaigns that compound the damage.

Asymmetry

Building reputation is slow. Damaging it is fast. This asymmetry is intentional — providers are designed to protect users, so they react more aggressively to negative signals than positive ones.

Consider the math:

Earning back trust after a reputation drop might take 2–4 weeks of consistently positive signals
Causing a reputation drop can happen in a single campaign with a high complaint rate
Recovering from a blacklisting can take days to weeks, even after the root cause is fixed

This asymmetry means that prevention is always more efficient than recovery.

Signal Weighting: What Matters Most

Not all signals carry equal weight. While exact algorithms are proprietary, the industry has established a general hierarchy through testing and observation:

Tier 1: Highest Impact

Spam complaints — A complaint rate above 0.1% (1 per 1,000) is dangerous. Above 0.3% is critical.
Spam trap hits — Sending to known spam trap addresses is an immediate red flag.
Engagement patterns — Consistently low open rates signal irrelevance.

Tier 2: Strong Impact

Domain reputation — The aggregated trust score of your sending domain.
IP reputation — Especially critical for dedicated IP senders.
Authentication results — Passing SPF/DKIM/DMARC is expected; failing is penalized.
Bounce rates — High bounce rates indicate poor list hygiene.

Tier 3: Moderate Impact

Content patterns — URL reputation, structural similarity to spam.
Sending patterns — Volume consistency, time-of-day patterns.
List acquisition method — Double opt-in lists perform measurably better.

Tier 4: Minor Impact

Subject line content — Specific words matter far less than overall patterns.
HTML quality — Broken HTML or excessive code is a minor negative signal.
Image-to-text ratio — Image-heavy emails trend slightly toward spam.

The key insight is that sender behavior (Tier 1 and 2) vastly outweighs message content (Tier 3 and 4). Content-based filtering was the primary approach in the early 2000s. Modern filtering is reputation-based and behavior-based.

Signal Visibility: What You Can and Cannot See

One of the biggest challenges in deliverability is that most signals are invisible to senders:

Visible Signals

SMTP response codes (bounces, deferrals, rejections)
Feedback loops (FBL) — complaint reports from some providers
DMARC aggregate reports — authentication pass/fail data
Google Postmaster Tools — domain reputation, IP reputation, spam rate, authentication
Microsoft SNDS — IP reputation data for Outlook/Hotmail

Invisible Signals

Per-user engagement scoring
Content filter scoring details
Exact reputation thresholds
Silent drops (no bounce, no feedback)
Spam folder placement (unless you test with seed accounts)
Provider-specific algorithm weights

This visibility gap is why inbox placement testing is essential. Without it, you are flying blind on the most important metric: whether your messages actually reach the inbox.

Practical Implications

Understanding signal flow leads to several actionable principles:

Authenticate everything. SPF, DKIM, and DMARC are the minimum. They don’t guarantee inbox placement, but failing them almost guarantees spam.
Monitor engagement relentlessly. Opens, clicks, and complaints are the strongest signals you can influence. Track them per campaign, per segment, and per provider.
Respect the feedback loop. When engagement drops, reduce volume and re-engage your most active subscribers. Don’t blast harder — blast smarter.
Test inbox placement directly. Don’t rely on delivery rates. Use seed testing to verify where messages actually land across providers.
Watch for latency effects. A change you make today won’t show results until the signal propagates through the provider’s scoring system. Be patient with improvements and urgent with damage control.
Build reputation gradually. The system rewards consistency. Sudden changes in volume, content, or audience trigger defensive responses from providers.

Signal flow is the nervous system of deliverability. Every decision you make as a sender creates signals. Every signal influences the provider’s model. And the provider’s model determines whether your next message reaches the inbox.