Research index

Every citation. Every mechanism.

If we say a feature makes customers feel valued, we name the paper, the year, and the finding that grounds it. This is the index. Each entry links to its implementation in the platform.

Book a demo → CX science page

— 01

The peak-end rule.

Kahneman · Thinking, Fast and Slow (2011) · also Kahneman, Fredrickson, Schreiber & Redelmeier · 1993 (Psychological Science)

People judge an experience almost entirely by its emotional peak and its ending, not the average and not the duration.

How we use it. Recap cards: at the natural close of a conversation, we render a recap with reference numbers and a sign-off shaped by the detected emotional peak. Negative-peak conversations get "we'll make it right." Positive-peak ones get "glad I could help today — enjoy."

Closure cards

Peak-aware sign-off on conversation resolution

— 02

Empathic acknowledgment outperforms apology.

Roschk & Kaiser · 2013 meta-analysis · 41 studies (Journal of Service Research)

Empathic acknowledgment that names the specific impact of a problem produced higher post-recovery satisfaction than apology-only across every study reviewed.

How we use it. The CX master prompt directs the model to lead with the impact, not the apology: "A scratched table with in-laws arriving Friday — that's a real problem." Then process explanation. Then remedy.

Acknowledge → Impact → Act

Master CX prompt directive

— 03

Concrete language is the signature of attention.

Packard & Berger · 2021 (Journal of Marketing) · also Moore (2022) on personalization backlash

Analysis of 1,000+ customer service exchanges. Concrete language ("the navy blue sectional with the pull-out") produced 30% higher satisfaction and measurable repurchase lift versus abstract phrasing ("that item").

How we use it. The model is directed to swap pronouns and generics for specifics pulled from page context, order data, and conversation history. "Your order" becomes "your Halcyon dining table." "Soon" becomes "Thursday between 10 and 2."

Concrete language

Master CX prompt directive

— 04

Service recovery has three justice dimensions.

Tax, Brown & Chandrashekaran · 1998 (Journal of Marketing) · also Smith, Bolton & Wagner · 1999

A recovered service failure can produce higher loyalty than no failure at all — but only when interactional, procedural, and distributive justice are all present. Interactional was the strongest predictor.

How we use it. Failure responses are structured by design. Empathic acknowledgment first. Then process explanation. Then remedy. Never collapsed into a single sentence.

Service recovery paradox

Three-layer responses

— 05

Active listening scales to text.

Weger, Castle & Emmett · 2010 (Int. J. of Listening) · Bodie · 2011 (AELS scale) · Shapiro et al. · 2020 (chat-based therapy)

Paraphrasing produces higher rapport and perceived understanding than direct answers, even controlling for answer quality. The effect persists in asynchronous text.

How we use it. For complex or emotional questions, the first response sentence reflects the customer's situation back in their concrete terms before the answer.

Paraphrase-before-solve

Active listening directive

— 06

The personalization-privacy paradox.

Aguirre, Mahr et al. · 2015 (Journal of Retailing) · John, Kim & Barasz · "Ads That Don't Overstep" · HBR 2018

Personalization improves satisfaction only when the data source is perceived as reasonable — explicit data (what the customer told you) outperforms covert data (what you tracked) on both trust and satisfaction.

How we use it. Returning-visitor recognition only references things the customer themselves said in prior chats. "Last time you mentioned the basement" is allowed; "I see you've visited 4 times" is not.

Recognition without surveillance

Returning-visitor handling

— 07

Reciprocity via useful disclosure.

Cialdini · Influence (1984, rev. 2021) · Morales · 2005 (J. of Consumer Research)

Unrequested but genuinely useful micro-disclosures trigger reciprocal loyalty stronger than discounts of equivalent economic value.

How we use it. The CX directive instructs the model to share one non-obvious helpful fact per relevant conversation — never forced. "One thing worth knowing — this model's feet unscrew, so it'll clear a 29-inch doorway."

Reciprocity micro-disclosures

Master CX prompt directive

— 08

The labor illusion.

Buell & Norton · 2011 (Management Science) · Buell · 2019 ("Operational Transparency")

Customers rate identical operational outcomes higher when they can see the work being done — even when total time and result are exactly the same.

How we use it. The widget surfaces specific intermediate work — "Pulling specs for the dining table… searching the knowledge base… checking your delivery window." Never fake delays. Real narration of real steps.

Visible effort

Progress narration

— 09

Name use has an inverted-U curve.

Howard, Gengler & Jain · 1995 · Packard & Berger · 2024 (overuse research)

Name use increases compliance and warmth perception — but overuse (more than twice in a short interaction) trips the "salesperson heuristic" and reduces trust.

How we use it. The CX directive caps name use at twice per conversation: optionally near the start (only if we already know it without asking) and optionally at the close. Never mid-response.

Name use cap

Master CX prompt directive

— 10

Closure shapes memory.

Gentile, Spiller & Noci · 2007 (European Management Journal) · also Zeigarnik effect research

Explicit closure markers ("you're all set — your claim is #4471, I've logged what we discussed, done for today") produce higher retention of satisfaction than conversations that fade out.

How we use it. Recap cards: a closing card with reference numbers and a deliberate sign-off. Tells the customer the interaction had a shape — a visit, not a ticket drain.

Closure rituals

Recap card on resolution

Audit, tune, disable

Every principle is admin-toggleable.

If a principle doesn't fit your brand voice, turn it off. The CX master directive is one toggle. Recap cards another. Returning-visitor recognition another. Studied. Cited. Optional.

Book a demo → CX science page