Email is not the villain

[Chris Chambers1]

Earlier this year I worked with a large European IT services and digital transformation provider on its bid to take over the customer service operations of a global professional services firm. With the key sales meetings and presentations now done, they remain a prospect and we are yet to hear back on our success (or failure) in the bid. Their operation is huge: nine service desks supporting the firm's own corporate clients across more than 1,500 business applications, and roughly 23,000 cases each month, with some 14,000 of them arriving by email. Together with the various in-app and portal web forms, those channels account for around 98% of the volume.

By the Best and Final Offer stage I was in front of the prospect's leadership, setting out where ServiceNow could carry the transformation. Our headline use case was AI Email-to-Resolution: let the platform read, validate, check and draft — resolving a meaningful share of inbound email untouched by a human, and handing the rest, already triaged, to a human agent, where AI was unable to take it further.

The Chief Customer Officer pushed back and challenged us on it: inbound email was seen as burdensome and a scourge on their resources, consuming much of their team’s time with triage and routing. They viewed email as the channel that should be avoided altogether, and they already had plans to close it down over time.

I disagreed and said so. Not because email is elegant — it isn't — but because the instinct, which I hear right across the industry, is aimed at the wrong target. Email is not the enemy; how we manage it after it lands, is.

Where the bad reputation comes from

I have been a ServiceNow consultant long enough to have heard that refrain (email is the devil, let’s get rid of it/we aren’t going to offer it in the new platform) from many of my customers. I started paying attention to what actually happens after an implementation goes live. Every service organisation ends up catering for inbound email in an implementation — because “today they have to” — and those same organisations never close it down. OK, you could argue the reason, is that resources (people, time, energy, funding, sponsorship) wane after an implementation, and nobody really ever tracks this initial goal through to completion anyway (let alone tries to measure how they did against it), but I think there’s a different reason that depletes the impetus of that desire: email is where the customer already is.

The real issue lies in structure. A well-operating desk runs on structured data to drive the process: data points such as category, priority, entitlement, assignment group, affected service, and so on. Those data points drive the process, the SLA and the routing. A self-service catalogue item form supplies all of them — or could — at submission. An email supplies only semi-structured data with a sender, a subject line and a body. Everything else — what the customer wants, how urgent it is, which of several hundred queues it belongs to, whether they are entitled to the service at all — must be inferred by a person reading prose.

That inference is the cost. On the busiest of the nine desks, first-level resolution sat at around 30%, and roughly 45% of volume was an unstructured request for assistance: someone describing a problem in their own words. The complaint was never really about email. It was about the reading and routing that must happen before any help begins.

What the numbers say

Email is not a legacy channel awaiting decommission. The Radicati Group¹ counted over 4.4 billion email users worldwide in 2024, forecast to pass 4.9 billion by 2028. Salesforce's sixth State of Service report² found 90% of service organisations offer email support — near-universal provision of a channel those same organisations say they want to shrink.

Customers have not left it either. Asked by YouGov³ in March 2025 how they would prefer to contact a business, Americans put email second at 23%, behind phone at 35% but well ahead of live chat at 10%, web forms at 5% and chatbots at 1%. Read that last figure, again: the channel the industry wants to retire is preferred to the chatbot the industry keeps building by more than twenty to one.

Then the finding I found quite alarming: SuperOffice's⁴ benchmark sent a straightforward service email to 1,000 companies. 62% never replied. Where a reply came, it took 12 hours and 10 minutes on average. 90% never acknowledged receipt. Only 20% answered the question in full, first time. None of that tells me email is the problem. It actually points to what organisations do with it.

Customers want their own record

There is a second reason to stop treating email as the problem, and it has nothing to do with operations. I am a customer too. When I need help from my mobile provider, my local council or a utility, I want the exchange in my own inbox — not in someone else's portal, behind someone else's login, on a system I lose the day I switch supplier. I want a copy I own and can still search in three years, when it matters.

That is a quiet and near-universal requirement, assumed by customers as utilitarian. Removing the channel does not remove email from the process, either. It only makes people resent dealing with you. Furthermore, email is often the only channel that genuinely works for screen-reader users, for anyone with high cognitive load (I can read it in my own time, when I am ready), and for anyone on a poor connection — a portal form is frequently the least accessible option on offer. Plus, if you offer a portal, you need to make sure it works across device types and browsers, thus loading more upfront cost on that channel. When it’s email, the “interface” is the customer’s cost to bear (as negligible as it might be).

The self-service trap

Consider what you are really asking when you push customers to self-serve. You want a form filled in. Reasonable enough — provided the experience is as easy as writing an email, and preferably easier. With a well-structured catalogue, that form gives you identity, intent, category, priority and routing at the point of submission, and the expensive triage never happens. But watch what follows next: the form is submitted, then you email the customer a confirmation; they reply to it (which ideally lands back on the original case), and an agent replies back. The case runs to resolution over email regardless. Organisations spent years telling customers to stop emailing, only to then conduct the whole customer interaction by email — meaning, they still accept inbound mail on open tickets, because refusing would be absurd. So the channel was never going away — only the pretence was.

Automate the triage, not the customer

So, the answer is not to close the door on email. It is to stop making a human read everything that comes through it.

That was the design we put forward. Intent and sentiment extracted the moment the case is created. Predictive intelligence proposing category, priority and assignment group. An agentic triage workflow that validates the contact against the account, checks entitlements against contracts, assets and sold products, pulls entities from the email body and attachments, and puts a grounded question to the knowledge base. Then a decision: resolve and reply, ask for the missing information, route to a queue, or hand to a human with the picture already assembled. On resolution, the close notes write themselves, and a knowledge article is drafted where the answer was new.

For our bid, we set what we believed to be a reasonable target of 20–30% of monthly email resolved end to end with no human handling — a figure to prove in a pilot, and not a promise (we had only high-level data figures to work with, and did not have their processes or overall operating model, so we based it on industry averages). If each of the 14,000 emails takes roughly five minutes to read-and-route as a measurement baseline, then 20–30% of that workload would be 230–350 hours of L1 desk analyst time saved each month.

Three caveats we called out: Out-of-the-box agentic workflows are templates, not turnkey resolvers; they need typical implementation activities, such as activation, cloning, security configuration, evaluation and tuning against real contact drivers and most importantly, data. They also need to start supervised, so that the service organisation does not lose control and damage its otherwise impressive CSAT and service-level attainment. And in this operation, two competing global brands could raise cases against the same application, so a misrouted case would not be a delay but a confidentiality breach. Nothing should auto-resolve until the customer context is confirmed, and autonomy is earned, one contact driver at a time, signed off and audited.

The mindset is the villain

Signpost your customers by telling them, in the acknowledgement they receive, that the portal is faster/better — but of course, make sure it really is faster, and a better experience than sending an email. Do not punish someone for using the channel they chose, and do not build a transformation programme on the assumption that they will stop. Email is where your customers are — serve them there!

When I think of my own experiences as a customer who needs something from my providers (oddly, I have had a few too many of those recently, across very different kinds of organisation: local government, telecoms, technology, retail) I know I want a confirmation email carrying the case number, a summary of what I asked, what happens next and when I will hear back… even if I used their online forms (which were painful to find and then to fill in — use ServiceNow, for goodness sake!).

The best of my recent examples was my local council: after trying to find some way to self-serve, I found the website listed an email address, so I laid out my issue in full — with attachments — only to get an email back a few days later, telling me to fill in a very specific form (I would never have found it; nothing signalled to me that it was the one I needed). What happened next? An email thanking me for my submission, giving me a ticket number, and promising that somebody would be in touch within ten days. The case is still ongoing, and the whole thing is being conducted over — you guessed it — email. The good thing, for me, is that I retain the complete record of events and who said what — my own case history, on my own system, not theirs. Who is to say I would not lose access to theirs at some point, perhaps when I need it most?

I hope my small example shows that email serves as the documented record customers want to keep, and is what they reach for when they cannot find your form. And the unstructured prose that made it expensive is precisely what large language models are now good at.

The villain was never the channel. It was the assumption that a human had to read it first.

Sources

Before writing this, I wanted to check that my own experience as a consultant and as a customer, was not at odds with what the industry’s own research says; those organisations have far greater reach than I do. Here are the sources I used in this essay, for your own reading:

1. The Radicati Group, Email Statistics Report, 2024–2028 — over 4.4 billion email users worldwide in 2024, forecast to exceed 4.9 billion by 2028. https://www.einpresswire.com/article/751597875/the-radicati-group-releases-email-statistics-report-2...
2. Salesforce, State of Service (6th edition, January 2025) — 90% of service organisations offer email support. https://www.salesforce.com/au/blog/digital-customer-service-channels/
3. YouGov, “How Americans prefer to contact businesses for customer service”, March 2025 — phone 35%, email 23%, live chat 10%, website forms 5%, chatbots 1%. https://yougov.com/en-us/articles/51802-how-americans-prefer-to-contact-businesses-for-customer-serv...
4. SuperOffice, Customer Service Benchmark Report (1,000 companies tested) — 62% do not respond to customer service emails; average response time 12 hours 10 minutes; 90% do not acknowledge receipt; only 20% answer in full on the first reply. https://www.superoffice.com/blog/customer-service-benchmark-report/
5. Gartner press release, 5 March 2025 — agentic AI predicted to autonomously resolve 80% of common customer service issues by 2029, reducing operational costs by 30%. https://www.gartner.com/en/newsroom/press-releases/2025-03-05-gartner-predicts-agentic-ai-will-auton...