Automate the Invisible

Problem

AI gets pointed at the relationship, not at the friction behind it

Ask leaders where they plan to use AI first, and the answer usually lands on the most visible work they have. Often that's the contact center or the customer chat window, but it can be the sales follow-up, the account review, or the client check-in. Visible work is easy to count - headcount, volume, handle time - so the business case writes itself. It demos well. And it's what most AI vendors are selling.

Visible work isn't automatically the wrong target. Routine, high-volume requests - a password reset, an outage status, a balance check - are often better served by automation, and customers are glad not to wait. The failure is narrower: automating the relationship moments. These are the high-emotion, high-judgment interactions where a customer decides whether they trust you - a dispute, a retention call, a complaint, a moment of loss. Automate those and you don't just save cost; you remove the thing that made you different from a competitor with a cheaper price.

Meanwhile, the work that should be automated first sits untouched: work that is low in visibility and low in regulatory risk. The handoffs between teams. The status chasing. The re-keying of data from one system to another. The same question asked twice because nobody passed the answer along. Customers never see this work, but they feel it - as delays, as transfers, as "let me check and get back to you." And it stays exactly where it was.

Klarna is the clearest public example. After moving aggressively to AI-first customer service, its CEO told Bloomberg in May 2025 that the chatbots were cheaper but delivered lower-quality service, and the company began hiring human agents again so customers could always reach a person. Klarna didn't fail because it automated customer service. It failed because it automated the moments that needed a person.

The result is the worst of both: the relationship is hollowed out, and the friction is fully intact. The organization has automated its differentiator and kept its dysfunction.

Value

More time for the work only people should do

People move to high-touch, high-impact, high-risk work. The time freed from coordination and data movement goes to the relationship moments, the judgment calls, and the decisions that carry real consequence.

Customers feel the difference. Fewer transfers, fewer repeated questions, fewer waits, because the friction behind the conversation is gone.

The relationship stays a differentiator. The moments where customers decide whether to trust you are still handled by people.

Green-light decisions you can defend. AI investments are selected by where friction is highest and risk is lowest, not by what is easiest to count or demo.

Context

When this pattern applies

This pattern applies when an organization is deciding which AI initiatives to fund and launch - especially the first ones. That includes portfolio and intake reviews, AI steering committee agendas, vendor pitches promising headcount savings in customer-facing roles, and any business case built mainly on reducing visible work.

It applies most when the obvious candidate is a relationship moment. If the proposal automates the conversations where customers decide whether to trust you, slow down and look behind it.

The relationship and the friction are connected. When customers complain, it's rarely about the person on the line. It's about being transferred, asked the same thing twice, or promised a callback that never comes - and every one of those is invisible work failing upstream. Fixing that work doesn't just free time; it makes the human conversation better. Automating the front of a broken process only gives customers a faster way to hear "I don't know where your case is."

It does not mean visible work is off limits. Routine, high-volume requests are often the right target. The pattern is about order and selection: friction first, relationship moments last, if ever.

Solution

Find the friction, count it, and green-light there first

Step 1

Map the value stream - and use AI to do it

People have normalized invisible work, so asking "where's the friction?" rarely finds it. Map one end-to-end process, from request to resolution, using value-stream mapping. At each step, ask three questions: who touches this next, how long does it wait, and what gets re-entered? If mapping a process would take a quarter, that is the first thing to automate: use AI to read the tickets, emails, system logs, and process documents and draft the map in a day, then have the people who do the work correct it.

Step 2

Sort what you find - from the organization's view and the customer's

Group	What it is	What to do
Relationship work	High emotion, high judgment, high consequence - where trust is won or lost	Keep people here
Invisible work	Low visibility, low regulatory risk - friction customers feel but never see	First AI candidates
Hidden but regulated work	Low visibility, high consequence - compliance, financial reconciliation, revenue recognition	Not a first target; needs its own controls

Then check the sort against the customer's view. Pull complaint and contact data and look for the moments customers say mattered. Something the organization calls routine can be a relationship moment for the customer - a balance check is routine until it's the customer whose paycheck didn't arrive. Treat those as relationship work.

Step 3

Baseline and count the friction

Before anything changes, measure what each candidate costs today: cycle time, days in queue, touches per transaction, rework hours, repeat contacts. The baseline is what lets you prove the value afterward - skip it and you never will. It also gives invisible work a business case that can compete with the headcount line.

Step 4

Green-light by friction, risk, and readiness - not by volume of voice

Rank proposals by how much friction they remove, how low their risk is, and how ready the team is to absorb the change - not by how easy they are to demo, or by which executive is loudest and best funded. The groups pushing hardest for the newest technology are sometimes the least prepared to succeed with it. Ask every proposal that targets visible work: is this routine work, or a relationship moment?

Name who makes this call. Ideally it is an AI governance board. Where there isn't one, it is whoever owns the intake decision today - often the EPMO or the portfolio review - and that owner applies the same test.

Step 5

Reinvest the time where it matters

Before launch, name where the freed time goes - the specific high-touch, high-impact, high-risk work it moves to - the same way you would name a budget line. If it isn't redirected on purpose, it gets absorbed and the value disappears. (See Pattern 11, The Productivity-to-Headcount Trap.)

Notes

Signs you're automating the relationship

Watch for these after launch. They are the early warning that AI has landed on a relationship moment:

Repeat contacts riseCustomers come back because the first answer didn't resolve it.
Escalations climbMore requests to "speak to a person."
The numbers splitSatisfaction falls while handle time improves.
Complaints shiftFrom the problem itself to the experience of getting help.

Related. Redesign the Stream, Not the Role (coming soon) - the invisible work is what value-stream mapping surfaces; this pattern is related, but is not necessarily that method's first step. Pattern 04, The AI Operations Manager, is one way to automate coordination work. AI Friction (coming soon) covers using AI where it adds no leverage.