How it works

Three ideas make it
work: adaptive personas,
strategic search, and
conversation intelligence.

High-stakes conversations — sales calls, salary negotiations, vendor deals — rarely go the way you expect. Dialectiq gives you a realistic sparring partner, a strategy engine that has evaluated thousands of possible moves, and a feedback loop that shows you exactly where you went wrong and what would have worked better.

One · Adaptive personas

Every conversation is different because every person is.

A direct, data-driven approach might work with one counterpart but backfire with another who values relationship-building first. Generic roleplay misses this entirely. Dialectiq builds virtual individuals from three composable layers.

01LAYER

Personality traits

A cognitive profile — how they process information, make decisions, handle conflict, respond to pressure. Communication style, emotional triggers, risk tolerance, decision patterns. Not surface-level labels.

02LAYER

Context & background

A CFO negotiating a vendor contract behaves differently from a hiring manager discussing salary. Every persona is grounded in a specific role, industry, scenario, and set of constraints you define.

03LAYER

Generative layer

Controlled randomness — hidden motivations, varying resistance levels, undisclosed constraints. You can't memorize answers, only build real skill. Thousands of unique virtual individuals, each requiring genuinely adaptive approach.

Two · Strategy intelligence

The engine doesn’t pick a response.
It searches.

When you ask for a hint, Dialectiq uses Monte Carlo Tree Search combined with AI agent simulations to explore thousands of conversation paths. Most end in failure. Through repeated simulations, the engine finds the specific sequence of strategies with the highest probability of reaching your goal.

MCTS · SIMULATION #001 / 2,412
WIN FAIL EXPLORING
EXPLORING · ROOT
EXPLOITATION

Double down on strategies that already look promising. Don’t waste time on dead ends.

EXPLORATION

Test new strategies that haven’t been tried. Don’t get stuck on the first decent idea.

Deep neural networks guide and accelerate the search — prioritizing promising strategies earlier, reducing simulations needed. Trained on proprietary data and anonymized outcome data: which strategies worked against which personality profiles, with all conversation text and personal information removed.
Three · Conversation intelligence

The search space isn’t random.
It’s a curated library.

Twelve negotiation strategies, six tonal variations, 72 distinct moves — each tailored to different personality profiles. The engine tracks momentum turn-by-turn and evaluates each exchange against your counterpart’s profile. The hint you see is the move that scored highest across thousands of continuations of your conversation, against your counterpart, at this point.

Help me win

After the session,
the counterfactual replay.

Takes your actual transcript and explores alternate paths — what would have happened if you’d chosen differently at each turn. The engine identifies your biggest missed opportunities and shows you exactly what you could have said and why it would have worked.

Practice. Analyze. Improve.

SAMPLE REPLAY · TURN 3 OF 7
T1 · Opening anchorHELD
You held + reframed. +14
T2 · He pushed on scopeHELD
Calibrated question. +9
T3 · You offered 8% discountMISSED
Logroll would have scored +22 vs -14
T4 · Confirmed termsHELD
Summary + next steps.
Voice + difficulty modes

Real conversations happen out loud, under pressure.

Speak naturally with the microphone button — the AI persona responds with a voice matched to their personality. Three difficulty modes scale the pressure. The timer starts after the persona finishes speaking; if it runs out, whatever you’ve said gets sent. Just like a real conversation where silence is its own signal.

EasyLEVEL
Text or voice · no timer
Learn new approaches. Build confidence.
MediumLEVEL
Voice only · 30s response timer
Build fluency under moderate pressure.
HardLEVEL
Voice only · 15s response timer
Simulates real pressure. No time to think.
More than a chatbot

“Can’t I just ask ChatGPT?”

You can — but asking a chatbot to help you negotiate is like asking a chess engine to play both sides. Dialectiq separates the counterpart from the coach, so the counterpart fights back realistically while the coach searches thousands of moves to find your best response.

GENERIC AI
Chatbot
PURPOSE-BUILT
Dialectiq
COUNTERPART
Same generic assistant every time
Adaptive personas with hidden motivations, unique pressure points, distinct voices
STRATEGY
One-shot response, no search
MCTS evaluates thousands of possible paths
COACHING
Generic tips ("be confident")
Specific moves backed by simulated evidence — what to say and why
PRACTICE
Text only, no memory of your scenario
Text + voice, difficulty modes, persistent sessions
ADAPTIVITY
Agrees too easily, no real pushback
Pushes back realistically based on personality profile
AFTER SESSION
Nothing — conversation is gone
"Help Me Win" shows what you missed and what would have worked
Built for preparation

A preparation tool — for individuals
practicing a tough salary conversation,
for sales teams sharpening their skills,
for anyone who wants to walk in having
already rehearsed the hard parts.

The strategies the engine discovers are grounded in negotiation theory, tested across thousands of simulated conversations, and tailored to the personality you’ll face.

Try the live demoGet in touch