Cold outbound deliverability now sits somewhere between 55 and 75 percent depending on the benchmark you read, and aggregate reply rates have fallen to between 1 and 3 percent. One 2026 analysis covering roughly a hundred thousand outbound emails found volume up more than sixfold against 2022 while raw reply rates fell 38 percent over the same period.
Your CRM knows none of this. It records that a sequence step executed, attributes the activity to a rep or an agent, and reports it as work completed. When a quarter or more of that work never reached a human being, the activity number on your dashboard is not a small overstatement. It is a different quantity from the one everyone believes they are reading.
Sent and delivered stopped being the same number
Every CRM in common use models an email as an event with a timestamp and an owner. That model was designed when deliverability was in the high nineties, so the distinction between sent and arrived was genuinely immaterial and nobody built a field for it.
At 60 percent delivery the distinction is the single largest source of error in your outbound reporting. Two reps sending identical volumes can have delivery rates thirty points apart because one is on a domain that has been overworked, and the CRM shows them as equally productive.
The downstream effect is worse than an inaccurate dashboard. Reply rate is calculated against sent, so a rep whose mail is being filtered shows as having a message problem. They get coaching on their copy, they rewrite it, nothing improves, and everybody concludes the market is difficult. The actual fault is in infrastructure nobody is looking at, because the system has no field in which that fault could appear.
Volume stopped being a lever some time ago
The reason this matters now rather than three years ago is that the volume went up and the returns did not.
Sending more used to work. If a sequence returned two meetings per thousand, three thousand returned six, and the constraint was how many emails a person could physically send. Removing that constraint looked like the obvious win, which is exactly what automated sending and then AI assisted prospecting delivered.
What happened instead is that everybody removed the same constraint simultaneously. Inbox providers responded with the strictest filtering they have ever applied, Microsoft 365 most aggressively, and the marginal email now has a materially lower chance of arriving than the one before it. Published figures put domain reputation collapse from over sending as capping close to half of attempted AI assisted outbound deployments within their first ninety days.
This is worth stating in the uncomfortable form. Sending more can now reduce the number of people who hear from you, because the additional volume degrades the channel that carries all of it. That is a genuinely counterintuitive relationship and no standard sales dashboard is capable of showing it.
Read More Your Category Is Searched Less Every Year. Where Does Demand Come From Now?
Domain reputation is a pipeline asset
The consequence for anyone building or configuring a CRM is that sending identity has become something the system must model, and almost none of them do it natively.
Which domain sent this, which mailbox, what that domain's recent delivery rate looks like, and how much volume it has carried in the last thirty days. Without those, you cannot tell a copy problem from an infrastructure problem, and those two diagnoses lead to completely different work.
Treat a warmed sending domain the way you would treat any other finite asset. It has a capacity, it degrades under load, it takes weeks to recover, and it is shared across everyone using it. Most organisations discover it is a shared resource at the point where one team's campaign has spent it.
We have added exactly these fields to client systems more than once, and the first report they produce is usually the awkward one: a large share of the quarter's outbound activity concentrated on one or two domains, with a delivery rate nobody had ever measured.
What the record actually needs to carry
The changes here are small and they are not defaults, so somebody has to decide to make them.
Delivery outcome on the activity record, not just dispatch. Bounced, filtered, delivered, and unknown are four different states and collapsing them into sent destroys the information you need. Unknown is a legitimate value and pretending otherwise is how these systems end up lying with precision.
Sending identity on the activity record, so that delivery can be aggregated by domain and mailbox rather than only by rep. This is the field that turns an anecdote about a bad week into a diagnosis.
A reply that is distinguished from an auto response. Out of office replies, delivery notifications and automated acknowledgements inflate reply rate in most integrations, and they inflate it most exactly when deliverability is worst, which is the opposite of useful.
None of this is difficult. It is unglamorous schema work of the kind that gets deferred indefinitely because no one asks for it, which is the same dynamic behind the adoption failures we wrote about in why CRM adoption fails.
Change what counts as activity
If activity metrics reward volume, you will get volume, and volume is now the thing damaging the channel. The metric is not a neutral observation; it is an instruction.
Replace emails sent with something that cannot be gamed by sending more. Conversations started is the honest version. Accounts genuinely engaged is better if you can define it. Both are smaller numbers and both are real, which makes the transition politically difficult and worth doing anyway.
Expect the reported numbers to fall considerably when you make this change, and prepare people for that before it happens rather than after. A team whose activity number drops by two thirds overnight will assume something is broken unless somebody has explained that the old number was measuring a different thing.
This is the same problem as cleaning a pipeline that shrinks when the stale deals come out, which we covered in why dirty pipeline data breaks automation. In both cases the accurate number is the smaller one and somebody has to be willing to publish it.
The channels that still work are the ones you do not track
Published guidance for 2026 converges on cutting sequence volume substantially and moving a large share of sourcing effort to warm channels: introductions, referrals, existing relationships and communities.
Those channels are almost always the worst instrumented part of a CRM. There is a rich object model for outbound sequences and frequently nothing at all for who introduced whom. So the organisation measures the channel that is declining in great detail and the channel that is working not at all, and then wonders why the reporting does not explain the results.
Adding a referral source, a relationship owner and an introduction record is a day of configuration in most systems. It is the highest return change available in this whole area and it gets skipped because it does not feel like sales technology.
We have watched that particular day of work get cut from scope repeatedly, on the reasonable grounds that it is administration rather than capability, and then get requested twelve months later when somebody senior asks where the good customers actually came from and the system cannot answer.
Our CRM development team treats delivery outcome and referral attribution as part of the core schema now rather than as reporting extras, because a system that cannot see either is describing a version of your sales process that stopped being accurate a couple of years ago.
Frequently Asked Questions
How do we find out our actual delivery rate?
Your sending platform knows it even if your CRM does not. Export delivery outcomes by domain for the last quarter and compare the total against what the CRM reports as sent. The gap between those two numbers is the size of the problem, and it is usually larger than anyone expects.
Is this an argument against AI assisted outbound?
No, it is an argument against using it to increase volume. The same tooling used to send fewer, better targeted messages from a healthier sending infrastructure works well. The failure mode is treating a capacity increase as permission to use all of it.
Should we buy more sending domains?
It postpones the problem and does not solve it, and it is what most teams do first. More domains spread the same volume across more reputations that all degrade, and it makes attribution harder. Reduce volume before adding capacity.
What is a realistic reply rate now?
Aggregate benchmarks put cold outbound at 1 to 3 percent, with well targeted campaigns to a genuine fit audience doing better. If you are comparing against a number you established in 2022, you are comparing against a different channel and the gap is not a performance failure.
Will our reps object to the metric change?
Some will, because their reported numbers fall. The way through it is to change the target at the same time as the metric, so nobody is measured against an old baseline with a new instrument. Doing one without the other is what makes it feel like a punishment.
How do we track referrals without adding admin?
One required field on the opportunity asking how this came to us, with a short fixed list of options rather than free text. It takes a second to complete and it is the only reliable source for this, because analytics attribution cannot see a conversation that happened at a conference.
Does any of this apply to inbound?
The delivery tracking does, for anything transactional or nurture related. A password reset or an order confirmation that is filtered produces a support ticket rather than a missed meeting, and most teams have no more visibility into that than they do into outbound.
Talk to our team if your outbound activity is rising, your meetings are not, and nobody can say how much of it is arriving.