Skip to main content
Automation•8 min read

Self Service: What Customers Will Actually Do

November 2, 2026•By Ajan Kanagalingam

Every business that builds a customer portal expects contacts to fall. Plenty of them do not, and the usual explanation offered internally is that customers prefer to call. They do not. They prefer whatever is faster, and for most portals calling is faster.

Where the line sits

TaskWill they self-serve?Why
Order or job statusYes, eagerlyFaster than waiting on hold
Rebooking or reschedulingYes, if availability is visibleNo negotiation needed
Invoices and documentsYesThey want it now, not tomorrow
A refund or a creditRarelyThey want it confirmed by a person
An exception to a ruleNoA portal cannot say yes to them

The pattern is consistent. Customers self-serve information and routine changes, and they want a person for anything where the answer could be no, where money moves against them, or where their circumstances are unusual enough that a rule would not cover it.

Pushing refunds and exceptions into a portal is the most common mistake, and it costs more than doing nothing. The customer tries, fails, and arrives at your phone line already annoyed, which makes the contact longer and the outcome worse.

Stop measuring deflection

Deflection counts contacts that did not happen, and it cannot distinguish a customer who got what they needed from one who gave up. Those are opposite outcomes recorded identically, which is why a deflection number can improve while satisfaction falls.

Three measures tell you something. Task completion rate on the self-service path, meaning people who started and finished. Repeat contact rate on the same issue. And how many people who used the portal contacted you anyway, which is the direct read on whether it worked.

That last number is the one nobody tracks and the one that changes decisions. Keeping it visible monthly is what a simple dashboard is for, and the measurement distortions that come with automating support are covered in customer service training when AI answers first.

The login is usually the problem

Asking a customer to identify themselves before answering a simple question is where most self-service traffic is lost. Checking an order status should not require remembering a password set eighteen months ago.

Where the information is not sensitive, a reference number and a postcode is enough. Reserve the login for things that genuinely need it, which is a shorter list than most portals assume, and let the low-risk lookups happen without a gate.

Now assume software is doing it

An assistant acting for a customer changes which constraints matter. It has unlimited patience for clicking and none of a person's ability to infer. Instructions that sit in a paragraph beside a field are invisible to it. A requirement implied by a heading does not exist.

Which means labels, validation messages and error text have to carry the meaning on their own, rather than relying on a human reading the page around them. That is also, usefully, what makes a form work for someone using a screen reader, as covered in accessible AI and serving every customer.

Two more things become decisive. Availability and pricing that are readable rather than gated, because a comparison is happening whether you participate or not. And a working one-step unsubscribe and cancellation, since software will complete those at volume. The wider shift is in websites that talk to AI agents.

Keep the human route visible

Hiding the phone number to force portal use raises cost rather than lowering it. The customer finds you eventually through a worse channel, having already spent effort failing, and that contact is longer and more likely to escalate.

Make the self-service path genuinely the fastest and most people will take it without being pushed. Where after-hours coverage is the real constraint, the options are compared in answering calls after hours, and the build-versus-buy question for support agents is in build versus buy.

Frequently Asked Questions

Which tasks will customers actually self-serve?

The ones that are faster than asking and where they trust the result. Checking an order status, rebooking an appointment, downloading an invoice, updating an address and resetting a password all clear that bar comfortably. Anything involving money moving in their disfavour, an exception to a rule, or a judgment about their specific circumstances does not, and pushing those into a portal produces an annoyed customer who calls anyway.

Why do self-service projects fail?

Usually because the self-service path is slower than the human one. If finding the answer means three clicks, a login nobody remembers and a search that returns nothing useful, calling is rational. The test is not whether the information exists somewhere; it is whether a customer with a specific question reaches a specific answer in under a minute without being asked to identify themselves first.

Is ticket deflection a good measure?

It is the wrong one, because it counts contacts avoided rather than problems solved. A customer who gave up is deflected and a customer who got what they needed is deflected, and those are opposite outcomes. Better measures are task completion rate on the self-service path, repeat contact rate on the same issue, and how many people who used the portal contacted you anyway.

What changes when an AI agent self-serves for the customer?

The constraints shift from patience to legibility. Software does not get tired of clicking and it cannot read instructions that live in prose beside a field, or infer a requirement implied by a heading. Labels, validation messages and error text have to carry the meaning on their own, and anything gated behind a form or a phone call is effectively invisible to a comparison the customer is making anyway.

Should the self-service option be the only option?

Keeping a human route available is the cheaper position, because the cases that reach a person after a failed self-service attempt are the ones most likely to escalate. Hiding the contact route increases cost rather than reducing it, since the customer eventually finds you through a worse channel and arrives already irritated. Make self service the fastest path and let people choose it.

Make the fast path the obvious one

We work out which tasks your customers will self-serve, remove the gates that lose them, and keep a person on the cases that need one.

Related Articles

Automation

AI Forms: Stop Retyping What Customers Send You

August 28, 2026Read more →
Automation

Staff Scheduling With AI: What It Gets Right

October 26, 2026Read more →
Automation

Purchase Orders: Worth the Admin for a Small Business?

October 18, 2026Read more →
AK
Ajan Kanagalingam
Founder & ChatGPT Consultant, ChatGPT.ca

Ajan leads the ChatGPT.ca team: 200+ custom GPT builds and automation projects for 50+ businesses across 20+ industries. Based in Markham, Ontario. PIPEDA-compliant solutions.

Stay ahead of AI in Canada

Weekly case studies, new tools, and ROI playbooks for Canadian SMEs. One email, zero spam.