The genre is older than the web. “Email is dead” posts appeared in 2008 (RSS killed it), 2012 (Twitter killed it), 2015 (Slack killed it), 2019 (LinkedIn InMail killed it), 2023 (community-led growth killed it), and 2026 (AI agents summarising your inbox killed it). Every one of these claims was wrong. Email quietly remained the highest-ROI channel for most B2B and B2C organisations through the entire period.
What actually died, multiple times, was a particular generation of metrics. First the “open” became meaningless under MPP. Then the “click” became meaningless under Safe Links. Then the “reply” became noisier under LLM-drafted responses. Each time, instead of replacing the metric, the industry pretended the channel itself was in trouble.
Email is structurally the hardest marketing channel to disrupt: federated, owned-audience, no algorithm intermediary. The problem has never been the channel. It has always been the measurement layer, and the measurement layer is fixable.
Why email structurally persists
Three properties keep email alive through every platform shift:
- Federated. No single company owns email. Not Google, not Microsoft. A channel you cannot be deplatformed from is strategically valuable.
- Owned audience. An email list is a durable asset. Followers on any social platform are rented.
- No algorithm in between. Subject to spam filtering, email delivers to the recipient directly. No ranking algorithm decides who sees what.
These properties are load-bearing. Every time a new channel emerges and gets hyped, it sacrifices one of these properties in exchange for better instrumentation or distribution. Eventually the property loss bites, and email regains relative importance.
What actually died
The metrics.
- Open rate — killed by Apple MPP in 2021. Not fixable. Officially a diagnostic, not a KPI, in 2027.
- Unique click rate — killed by Safe Links and enterprise AV in 2022–2024. Still useful if you segment out datacenter IPs.
- “Delivered” rate — was never alive. “Delivered” means “accepted by the recipient MX”, which says nothing about placement.
- Industry benchmarks — killed by the above. When every underlying metric is corrupted, cross-company averages are noise amplified.
What replaces them
A shorter, harsher, more honest set of metrics:
- Inbox placement rate — measured against seed mailboxes, per provider, weighted by audience mix.
- Reply rate — impossible to fake, cheap to measure with correct threading.
- Pipeline / incremental revenue — the only business-level metric that matters.
- Engagement-seconds — a quality signal that distinguishes real readers from clickbait opens.
These are not new inventions. They are just harder to compute than opens, so for two decades the industry did not compute them. Now the old metrics are broken enough that computing the harder ones is the only option.
Most teams have never measured their inbox placement, ever. Not once. Inbox Check takes three minutes, is free, and produces a number that will quietly restructure your roadmap.
AI does not kill email either
The 2026 “agents summarising your inbox” argument is directionally interesting but factually wrong as a death sentence. If anything, LLM summarisers privilege well-written email over clickbait, because a summariser reads the content and passes a meaningful abstract to the human. Bad subject lines stop mattering as much; good paragraphs matter more.
AI changes what “a reader” is — sometimes it is the human, sometimes it is the human via an agent that briefed them. But the channel keeps working. The measurement has to adapt: engagement-seconds captures agent-summary behaviour badly, reply rate captures it fine.
What to do Monday
- Find the last campaign report you sent upstream. Remove every metric except placement, reply rate, and pipeline.
- If you do not have a placement number, run a free placement test and add it.
- If you do not have a reply rate number, set up correct threading on your inbound and backfill from IMAP.
- Present the one-page report in the next QBR. See who flinches.
The channel is not going anywhere. The metrics that described it are. The teams that update their measurement layer first will spend the next decade outperforming teams that keep reporting opens.
Stop mourning the metrics
The metrics were never particularly good. They were convenient. Opens measured pixel loads; clicks measured HTTP GETs. Both were always proxies for human behaviour. The proxies held up reasonably well for twenty years because the ecosystem around them was stable. It is no longer stable. Move on.