ArkoliteArkolite
Our POV · POV 03 · Customer

Can AI build loyalty, or only transactions?

Eugene Yap, Managing Partner3 min read
The short answer

AI makes loyalty programmes cheaper to run and easier to personalise. Loyalty itself comes from the whole experience, and from trust. The programmes that last use AI to understand why customers behave as they do, design for the relationship as well as the next purchase, and give customers a say in how their data is used.

The number

Two views of the same relationship

Source: Twilio, State of Customer Engagement Report, 2025. 7,640 consumers and 637 business leaders in 18 countries.

Efficiency is easy to mistake for effectiveness. Most AI in loyalty is tuned for the next transaction: the email opened, the offer redeemed. Those signals are useful, but a customer who returns for the discount is loyal to the discount. Over time the model learns that offers work, and customers learn to wait for them.

In practice
Klarna

Cheaper service, then a step back.

In 2024 Klarna’s AI assistant was handling most customer chats, the work of about 700 agents. A year later its chief executive acknowledged the service had become lower quality, and promised customers there would always be a human if they wanted one.

Starbucks

Less discounting, more reasons to return.

Starbucks has long used AI to personalise offers for its 35 million active Rewards members, who drive nearly 60% of its US company-operated revenue. In 2026 it rebuilt the programme around tiers and engagement, aiming to bring members back more often with less reliance on heavy discounts.

Check your programme

Three questions for your loyalty AI.

1

Is it built for emotional loyalty, or only the next click?

Models rewarded on redemptions will find the customers most responsive to offers, and train them to wait for the next one.

2

Does it tell you why customers act, as well as what they will do?

Models trained on yesterday’s behaviour can drift when the world changes. Zillow’s home-buying arm lost hundreds of millions in 2021 when its pricing models misread a shifting market.

3

Do customers know how their data is used, and can they opt out?

Only 15% of consumers fully trust brands with their data, and 84% want control over personalisation.

Use AI the way a drunkard uses a lamppost: for support rather than illumination.
What to do next
  1. 1

    Measure the relationship.

    Track repeat visits, share of wallet and advocacy alongside clicks and redemptions.

  2. 2

    Pair the model with insight.

    Put customer research and frontline conversations next to the predictions, so the team understands why behaviour changes.

  3. 3

    Make data a fair exchange.

    Tell customers how AI uses their data, let them choose, and make the value they get back obvious.

Questions leaders ask
Can AI improve customer loyalty?

Yes, when it serves the whole experience. AI is strong at personalisation and efficiency. Loyalty grows from a valuable proposition, good service and trust.

Why do AI-driven loyalty programmes lean on discounts?

Most models optimise for the next transaction, so they learn that offers work. Over time customers learn to wait for the offer.

How should brands govern AI in loyalty?

Be open about how customer data feeds AI, give customers control including the choice to opt out, and keep people accountable for the decisions the models influence.

Sources
  1. Twilio (2025). State of Customer Engagement Report.
  2. Entrepreneur (2025). Klarna CEO on hiring customer service agents.
  3. Axios (January 2026). Starbucks overhauls loyalty program.
  4. Zillow Group (2021). Q3 2021 shareholder letter on winding down Zillow Offers.
Put it to work

Strategy for the AI-first leader.

Our Advisory partners help brands use AI to deepen customer relationships, from loyalty strategy to the data and governance behind it.

Next brief
POV 04AI strategy3 min

What does AI need around it to do real work?

The model is the part everyone can buy. Results depend on the harness that sets it up and the infrastructure it stands on.

Read the brief