chaewon lim
back to work

Fair Tipping Model · End-to-End Redesign · Agentic AI

Rethinking tipping for the age of AI

No more guessing. Tip for what actually happened.

Pre-service tipping in delivery apps is a high-pressure decision users have to make before the service is even complete. This concept proposal rethinks that flow end to end — starting from why delivery tipping specifically is the most broken tipping context (versus dining or coffee shops), through to a working AI-mediated solution and its ecosystem effects on couriers and platforms.

Problem & secondary research

Tipping is essentially a guess.

Apps prompt for a tip before delivery happens, at inconsistent points in the order flow, with no context to judge quality against. The result is guesswork — and regret either way.

But why is tipping so ambiguous?

Pre-service

Timing

Tips are requested before service is complete, forcing a decision with incomplete information.

77%

Unclear value

of tipping is driven by service quality — but quality is impossible to judge before the delivery has happened.

20 / 25 / 30%

Social pressure

Preset percentage screens make tipping feel obligatory rather than a genuine choice.

2/3

Missing context

of U.S. adults feel uncertain about when and how much to tip, with no standard reference point (Pew, 2023).

Why does this problem exist now?

When payment gets smarter, gratitude gets lost.

Payment evolution

Cash (spontaneous, tied to service) → card (post-service) → contactless (pre-selected) → automation. Payment moved forward 20+ years; tipping UX never did.

Dining diversification

Dine-in (face-to-face) → takeout (optional-feeling) → delivery. Technology stretched dining across contexts with no in-person cue for what to tip.

Generational shift

Boomers/Gen X treat tipping as customary choice; Millennials/Gen Z tie it to wage gaps, social pressure, and cost anxiety — same gesture, different meaning.

How might we redesign tipping to be clearer for customers and fairer for couriers?

As is

User

Guesses tip amount

Service quality

No data to inform the tip

Courier

Unclear outcome

Earnings

Unfair, despite effort

To be

User

Feels confident

Service quality

Clear, quality-based tip

Courier

Fair, quality-based pay

Earnings

More predictable income

Solution

Seamless agentic AI for an end-to-end tipping workflow

1

Pre-tip, post-reward mode

Users set tip priorities (speed, careful handling, communication) at checkout. The final tip settles after delivery, based on what actually happened — not a guess made before it did.

2

Dynamic tip adjustment

An AI agent tracks delivery signals in real time and adjusts the tip against the user's stated priorities, with plain-language reasoning: "Delivered 4 min faster → +$1.20 added."

3

Auto-claim resolution agent

When something goes wrong, the agent detects the issue from the user's report, opens a tip-adjustment claim, and resolves it against platform policy — no back-and-forth required.

Design iterations

How much control did users actually want?

Iteration 1

AI decides silently, user is notified after

  • Zero effort for the user
  • Felt like a black box — no way to see why a tip changed
  • Users didn't trust an amount they couldn't trace back to anything

Iteration 2

User manually reviews every adjustment

  • Full transparency and control
  • Turned a 10-second checkout into a multi-step review flow
  • Feedback: "post-delivery feedback was too high-effort"

Iteration 3 — shipped

Priorities set once, reasoning shown on demand

  • Default flow stays one tap; a "view reason" button surfaces the AI's logic only when wanted
  • Preserves trust without adding steps to the common case

Ecosystem impact

A self-reinforcing loop, not a customer-only fix

Customers

Only pay for quality received, with less pre-tip anxiety and low-friction issue resolution.

Couriers

Fairer, more predictable pay tied to actual service, plus a reputation record that compounds over time.

Restaurants

Fewer customer/courier disputes and steadier long-term order volume.

Platform

Higher user and courier loyalty, lower support costs, better delivery-performance data.

Takeaways

Trust needs a reason, not just an outcome

Users accepted AI-adjusted tips only once they could trace the adjustment to something concrete. Silent automation, even when correct, read as untrustworthy.

Fixing tipping means fixing timing, not just UI

The core dysfunction wasn't the tip screen's design — it was that tipping happens before the thing being tipped for. Any fix had to move the decision point, not just restyle it.