Bajaj DCX · Research & Testing

Research Agents

A persona-driven research and testing engine — AI users grounded in real customer research that react to your designs before you ship them.

The engine

Four parts, built up in layers.

01 Personas Who the user is

Real customers, replicated — they think, decide and talk like the segment.

02 Perception How they use a screen

Shown the actual screen, they scroll and tap like a person — never the code.

03 Behaviour How real users act

Reactions weighed against how people actually behave on a phone.

04 Facilitator Who runs it

A moderator sets the task, probes without leading, and writes the report.

01 · Personas

Eight users, built from real customers.

India study 32 interviews · 3 income segments — how they handle money, decide, what frustrates them.
Recorded voices Hours of real customer audio — so each persona talks in Hinglish, short turns, real complaints.

Main 75–80% save kar leta hoon — parents ke saath rehta hoon, toh kharcha hai hi nahi.

Himanshu · India-3 · the 75% saver on a modest income
1
India 1 · High income

Earns the most, decides the least — agency sits with family.

3
India 2 · Mid income

Collaborative deciders, caregivers, the debt-resigned.

4
India 3 · Lower income

Extreme savers, sharp investors, disengaged spouses.

Outliers kept, not smoothed. Income ≠ agency; savings ≠ income.

01 · Personas

What’s inside one persona.

Himanshu India-3 · the 75% saver

Lives with his parents on purpose, banks most of a modest BPO salary, and lets compounding do the rest.

01
Who he is

A disciplined saver who defends living at home — the choice that makes the saving possible.

02
His money life

RDs, FDs, digital gold — he tracks what’s invested, not what he might spend.

03
How he decides

Compounding is the whole game; he resists one-click EMIs and takes the first option that looks right.

04
How he talks

Short, dry Hinglish — trails off, drops a “matlab”, never oversells.

05
How he uses a screen

Skims dense English, slides past promos, fumbles small buttons, rarely scrolls far.

The research behind the personas

Grounded in the field, not in hunches.

01Persona-LDecision transparency
02PersonaCraftHeadline + versioning
03PersonaFlowMany narrow personas
04UXAgentUsers skim, not read
05EQ-NegotiatorEmotional range
06Promise with a CatchGrounding imperative
07Low-Resource MisalignmentStay grim where data is grim
08PersonaLLM · Big FiveTrait anchors
09The Personality IllusionClaim ≠ behaviour
10The Empty ChairPreserve outliers
11Multi-Persona ThinkingBias mitigation
12DiverseDialogueReal-voice features

12 papers and 5 recorded interviews — every persona decision traces back to a source.

02 · Perception

It sees the screen, not the code.

What it sees

The actual screen, as an image. It scrolls, taps and types like a person — on a phone, by default.

What it never sees

The code, the page structure, the element names. No shortcut to the answer — only what a real user would look at.

Honest by design A missed tap is recorded as a missed tap — a finding, never hidden.
03 · Behaviour

Weighed against how real users behave.

Attention stays up top

Most never scroll past the first screenful.

People skim, don't read

Dense copy gets glanced at, then skipped.

First good-enough wins

They grab the first plausible option, not the best.

Promo-shaped gets skipped

Anything that looks like an ad is ignored, even if useful.

Small targets mis-tap

Crowded, tiny buttons cause fat-finger errors.

Long forms lose people

Every extra field sheds users; finance forms worst.

Dialled per persona — Mohammed reads the fine print others skim; Danish is pulled by the promo everyone else ignores.

A realism lens “Reached the button only by scrolling deep — most users wouldn’t, so flag it.”
04 · Facilitator

A moderated session, start to finish.

01 Set the task

A moderator briefs a real goal on a real flow.

02 See the screen

The persona is shown the actual screen, nothing else.

03 Think aloud

It scrolls, taps and reacts, in character throughout.

04 Probe

The moderator asks neutral, non-leading follow-ups.

05 Debrief

Steps out of character and explains its own choices.

06 Report

Findings written up, weighed against real-user behaviour.

In character, then out It stays in character the whole run — then steps out to explain exactly why it did what it did.
04 · Facilitator

Run it four ways — get a report back.

Four ways to run it
UsabilityCan they finish the task?
First-clickWhere do they go first?
ConceptDoes the idea land?
InterviewWhat do they actually think?
What you get back

A written report on every run — findings weighed against how real users behave.

The persona’s own debrief — out of character, why it did what it did. Run one persona, or all eight — a report each.
What it gives you

Test with your customers before you build for them.

On demand

Segment-true reactions to any flow — before you recruit a single participant.

In minutes

Ask how India-2 reacts to a change and have an answer the same hour.

Grounded

Every reaction traces to real customer data — and the voices your flows usually miss.