The Human Side of AI - part 6

Where Is Your Human Threshold?

By: Kara Maddox

AI can automate more of the customer journey than ever before. The question is where automation begins changing what the relationship means.

Over the course of this series, I've been trying to answer one question:

How should consumer-facing businesses use AI without automating away something their customers actually value?

The research didn't lead to:

Customers don't trust AI.

That's too simplistic.

It didn't lead to:

Always keep a human involved.

Also too simplistic.

Instead, five patterns kept appearing.

Context matters.

Degree of AI involvement matters.

Human effort can itself be a signal.

Human oversight only means something if the human contributes something meaningful.

And relational language carries implications about the person behind it.

Put those together, and I think there is a boundary businesses need to understand.

I call it:

THE HUMAN THRESHOLD.

What is the Human Threshold?

The Human Threshold is a working framework for identifying:

The point in a customer interaction where additional automation begins to change what the customer believes about the human relationship behind the communication.

It's not a fixed line.

And it's not the same for every company.

The threshold for:

Your order has shipped.

will probably be different from:

We reviewed what happened to your family and decided how we're going to make this right.

Same company.

Same technology.

Different social meaning.

Five things got us here

1. Context

Research on AI trust shows that willingness to rely on AI varies across tasks and stakes.

People don't simply decide whether they "trust AI."

They decide whether they trust it here.

2. Degree

Human-authored, AI-assisted, and fully AI-mediated communication do not always produce the same response.

Research comparing these forms suggests that the extent of AI involvement matters (Sahebi, Formosa, & Bankins, 2026; Sedefoglu-Ulucak, Ohly, & Schmelz, 2026).

3. Effort

AI disclosure can reduce perceived human effort—and perceived effort can affect engagement (Patil & Rice, 2026).

Efficiency changes not only how much work happens.

It can change what recipients believe happened behind the communication.

4. Meaningful oversight

"Human-reviewed" isn't a single behavior.

Verification is different from revision.

Revision is different from judgment.

Judgment is different from ownership.

If human involvement matters, businesses need to be able to say what the human actually contributed.

5. Relationship

AI can generate warm, empathetic, polished language.

But authorship and degree of AI mediation can affect perceived authenticity, sincerity, validation, warmth, and trust (Brandtzaeg, Skjuve, & Følstad, 2026; Glikson & Asscher, 2023).

The words aren't the whole relationship.

So how do we find the threshold?

I would start with three questions.

1. How much human attention does the customer reasonably expect?

At one extreme:

Your reservation is confirmed.

Very little.

At another:

You told us something serious happened. We're responding.

Quite a lot.

As expected human attention rises, full automation deserves more scrutiny.

2. How emotionally consequential is the interaction?

Not:

Does the message sound emotional?

Instead:

How much does this interaction matter to the customer?

Checking store hours?

Low.

Responding after a major service failure?

High.

Helping someone make a consequential financial, health, professional, or personal decision?

Potentially very high.

3. How much judgment and accountability does it require?

Some questions have straightforward answers.

Others require someone to decide:

What's fair?

Should we make an exception?

What should this customer do?

How do we repair this?

What are we willing to stand behind?

AI can support that decision.

But somebody still owns it.

Three AI modes

Those questions give businesses a useful starting model.

GREEN — AUTOMATE

Low expected human attention - Low emotional consequence - Low judgment/accountability

AI can potentially own most or all of the interaction.

Examples:

routine confirmations,

status updates,

scheduling,

basic FAQs,

straightforward information.

Goal: efficiency.

Don't insert human labor simply so the workflow looks responsible.

YELLOW — AUGMENT

Moderate human expectation - Moderate consequence - Meaningful judgment

AI does substantial work.

It can:

summarize,

analyze,

retrieve,

compare,

draft,

recommend.

A human remains meaningfully involved through:

verification,

revision,

contextualization,

judgment,

or approval.

Goal: capacity.

Let AI remove work without removing the part of the work that needs a person.

RED — HUMAN-LED

High expected human attention - High emotional consequence - High judgment/accountability

AI may still be incredibly valuable.

But it works behind the relationship.

It prepares.

Analyzes.

Summarizes.

Flags risks.

Surfaces history.

Suggests options.

The human owns:

the judgment

the decision

the responsibility

and

the relationship.

Goal: trust.

This is not a traffic-light policy

Reality is messier.

A low-emotion interaction may require substantial judgment.

A highly emotional interaction may have a simple factual answer.

Different customers will have different attitudes toward AI.

Different industries have different norms.

Existing trust will matter.

Disclosure will matter.

And the type of human involvement may matter.

So green/yellow/red isn't supposed to be a regulatory system.

It's a diagnostic.

A way to stop asking:

Can we automate this?

and start asking:

What happens to the relationship if we do?

The next question is empirical

The Human Threshold is not yet a validated scientific model.

That's important.

The literature gives us the pieces.

What we need next is the experiment.

Take the same customer communication.

Change only what the customer is told about how it was created.

Human

This message was written by a member of the organization.

AI

This message was generated using artificial intelligence.

AI + meaningful human review

This message was generated using artificial intelligence and then reviewed and revised by a member of the organization.

No disclosure

Nothing.

Then measure:

perceived human involvement

perceived effort

warmth

competence

authenticity

trust

behavioral intention.

Then change the interaction.

Confirmation.

Recommendation.

Complaint.

Apology.

Sensitive communication.

Now we can start looking for the threshold.

Not philosophically.

Empirically.

And then it needs to leave the laboratory

A controlled experiment can tell us whether the mechanism exists.

Businesses ultimately need to know whether it changes behavior.

Do customers:

reply differently?

click differently?

convert differently?

escalate complaints differently?

report different satisfaction?

remain customers?

The next stage would be a field partner:

a consumer-facing company already using AI in customer communication and willing to test where meaningful human involvement changes customer response.

That's where this becomes more than an interesting framework.

It becomes useful.

The goal is not less AI

After six posts about preserving the human, this is worth saying clearly.

The Human Threshold is not anti-AI.

Quite the opposite.

Automate the work nobody benefits from humans doing manually.

Let AI make teams:

faster,

better informed,

better prepared,

more capable.

Then preserve human capacity for the moments where humans contribute something the machine is not merely reproducing in language:

judgment

responsibility

presence

relationship

and perhaps:

care.

The future isn't human or AI.

It's deciding how the two should work together.

And for consumer-facing businesses, I think that starts with a better question:

Where is your Human Threshold?

---

Research note

This article concludes KJMdigital's six-part The Human Side of AI series.

The Human Threshold framework—its name, three proposed dimensions, and Automate/Augment/Human-Led model—is a KJMdigital working framework developed from patterns across the literature reviewed in this series.

It is not yet empirically validated.

That's intentional.

The next step is not to market the framework as settled science.

It's to test it.

---

References

Brandtzaeg, P. B., Skjuve, M., & Følstad, A. (2026). AI aversion? Effects of author disclosure on young people's perceptions of mental health advice. Cyberpsychology: Journal of Psychosocial Research on Cyberspace, 20(2), Article 1.https://doi.org/10.5817/CP2026-2-1

Glikson, E., & Asscher, O. (2023). AI-mediated apology in a multilingual work context: Implications for perceived authenticity and willingness to forgive. Computers in Human Behavior.https://doi.org/10.1016/j.chb.2023.107922

Patil, R. K., & Rice, D. H. (2026). Is that your best effort? The impact of AI-use disclosure on consumer engagement. Journal of Business Research, 215, 116354.https://doi.org/10.1016/j.jbusres.2026.116354

Sahebi, S., Formosa, P., & Bankins, S. (2026). The AI penalty and disclosure paradox: Trust, authenticity and knowledge uptake in AI-mediated communication. Computers in Human Behavior: Artificial Humans, 8, 100304.https://doi.org/10.1016/j.chbah.2026.100304

Sedefoglu-Ulucak, D., Ohly, S., & Schmelz, J. (2026). Empathy in leadership communication: Experimental evidence from two vignette studies on AI's role in message improvement and observers' perceptions. Computers in Human Behavior, 185, 109096.https://doi.org/10.1016/j.chb.2026.109096

Zhang, Z., Geng, J., & Qi, C. (2026). Observer perceptions of GenAI use and interpersonal trust: The roles of warmth, competence, and AI literacy. Behavioral Sciences, 16, 1221.https://doi.org/10.3390/bs16071221