Circa 2004, Bill Gates suggested that every email sent should have a small charge, like a digital postage stamp — perhaps X seconds of compute time — for every message. He was responding to the early-ish days of email SPAM. (For context, Blackberry had around a million subscribers in 2004. The main driver of growth for that innovative and solid little device was email access.) With the proliferation of PCs, and Internet connections shifting from slow dialup to fast fixed, skyrocketing message volume and plummeting quality were straining inboxes and tempers. No e-stamps, but plenty of growth:
- Email messages sent daily in 2004: 77 Billion (source)
- Email message sent daily in 2024: 362 Billion (source)
134 trillion messages or so will circle the planet this year.
Email at scale is not entirely free, of course. Where individual consumers have gotten accustomed to an unlimited inbox “for free” thanks to Gmail et al, businesses and organizations pay something for both sides of the email equation. They pay for their domains, inboxes, addresses and so on on the inbound side. They pay for email platforms, deliverability and sending on the outbound side. And they pay for people and expertise on both. But it’s worth noting that sending at scale, while not free, is not prohibitive. On the low side, Amazon SES runs about $1 per 10,000 — $100 to send 1 million messages. On the higher side, 1 million messages through MailChimp would cost $700 or so.
The big costs on the outbound side of email at scale are in those other things - email platforms, deliverability, content, design, data storage and management, infrastructure (authentication, IP addresses) and more. In general, these costs are more fixed than variable. Including another 1,000 people in an email campaign probably doesn’t make or break the campaign budget.
Now let’s add AI to the mix.
AI’s eventual capabilities - who knows! Circa 2025, LLMs are capable of producing subject lines and copy suitable for email messages. Debating whether AI copy is good enough for the email job is fun fodder for panel discussions, but beside the point. The written word is the core of email. If AI tools can do that job well enough (or better), and faster, and cheaper, the tools will be used. There’s more to email at scale than writing copy, but use the copy job as a proxy for a bunch of the current jobs-to-be-done in email - let’s call that ‘message-making’ – and assume that AI tools will gradually make those jobs easier, faster and cheaper.
If sending is nearly free already, and message-making becomes nearly free as well, what do you think is going to happen to message volume and message quality?
I think we’re going to see a “big split” over the next few years.
On one side of the split, we’ll see extraordinary personalization, where “an email message” has content that’s specific, timely and incredibly tailored to the recipient. (Personalization is the wrong term, since it’s fuzzier than a dropped lollipop, but it’ll have to do for now.)
On the other side of the split, we’ll see a deluge of digital crap - batch and blast by Terminator, at such relentless volume that everyone with an inbox will want to tear their hair out.
There won’t be a sticker saying which is which, and it’ll all be headed right to your inbox.
Ugh.
Maybe this is just a thought exercise, but look at “email marketing” really simplistically. Email was designed to be a person-to-people channel — one-to-one or one-to-many messages typed by a person and sent to people. A nanosecond or so before (or after) that glorious idea was formulated, the idea of faking it popped up. The guy named Fred at the receiving end really can’t tell if you typed “Dear Fred” or “Dear [firstname]” + mail merge. Pick your euphemism — campaign templates, mail merge, personalization, mass customization, customer intimacy — most of that stuff is a gigantic head-fake, stuffing data into templates for the appearance of person-to-people communication.
Now re-run that thought exercise with the most capable AI you can dream up doing the job. Couldn’t you — in theory — task Skippy the AI with perusing everything your company knows about a customer — website visits, orders, complaints — and everything the Interwebs know about that person to robo-draft a super-detailed, specific, useful, actually personalized not fake-personalized 1:1 message? Couldn’t you — in theory — have Skippy do that for each & every customer separately?
That’s a vision of the ‘extraordinary personalization’ side of the big split. In some ways, it’s an appalling thought; in other ways…not so bad. I’d rather have an AI labor away to send a hyper-relevant, useful message to me than get Yet Another mail-merge pretending to be personal. The individual employees at a given company don’t have the time, access and energy to do that job for every customer, but would it be an altogether bad thing if AI did? (Yes..and no. Hyper-personalized phishing and scam emails are already a thing, which is not good.)
The problem is, the other side of the split — where spammers live — is easier and cheaper. Tasking Skippy the AI to write an email template and a mail merge is probably less work and cost than generating that hypothetical 1:1 message — and eventually less work than having a human do it. So we’ll see more & more of that, and as noted, it’s all headed to the same inbox.
Which brings us back to Bill Gates and the fact that sending email is nearly free. What will constrain the rising flood of AI-plus-mail-merge drivel going forward? Email is a remarkably self-regulating industry, but can the orgs on the inbox side of the equation (Google, Microsoft, Yahoo, etc) make the investments required to fight AI crap with AI filtering forever? Is there enough business benefit for them to do that? Will some enterprising entrepreneur figure out how to make “hire my AI to manage your inbox” a business?
Prediction is a tough game, but I think the email channel is in for a very tough ride. The cost of message-making (and personalization-faking) was one of the constraints on unlimited volume. The poor-grammar-spammer was easy to spot, but perfect grammar is becoming free. How will we distinguish messages from the shiny side of the big split - truly personalized, truly useful - from spam, phish and time-wasting (but grammatically correct) drivel? I don’t think the answer is e-stamps, but I don’t AI detection of AI content will cut it either. Whether there’s enough of a problem, enough value, enough industry coherence and (for lack of a better word) collective willpower to make fundamental changes to the channel — we’ll see.
Illustration generated with the assistance of OpenAI’s DALL·E.