Research

Research Phase: From Data to Understanding

The research phase is the foundation of all work. Here data is collected, jobs are formulated, segments are built, and the basis for strategic decisions is created. All artifacts in this phase are based on the Product DNA methodology.

Pain Map

Methodology

The Pain Map is based on the Push forces concept from the Product DNA Switch formula. "Pain" in the methodology is not an abstract inconvenience, but a specific dissatisfaction with the current solution that pushes the user to seek an alternative. Push = pain intensity × frequency of occurrence.

In Product DNA, pains are classified by the type of job they relate to:

Pain Type	Description	Example	Product Significance
Functional	The task is not performed or performed poorly	"Forget to start the timer"	Direct functional value of the product
Emotional	Negative emotional state during work	"Feel guilty about overestimating hours"	Emotional job — how they want to feel
Social	Problems with how others perceive them	"Client thinks I'm unprofessional"	Social job — how they want to appear
Tax pain	Arises from defects of the current solution	"Re-entering data because export broke"	Tax job — can be killed through automation

How Generation Works

You describe user problems in chat (or upload Telegram data, review files)
AI extracts "pain" type facts with weights from 10 to 100 (weight = intensity × reliability)
Pains are grouped by category (functional, emotional, social, tax)
The generator builds a prioritized map: critical → medium → low

Quality Criteria

Specificity: "Forget to start the timer" is better than "inconvenient to track time"
Context binding: each pain is tied to a specific situation (trigger)
Traceability: each pain references a specific chat message
Quantity: minimum 5 pains for a meaningful map, optimally 10-15
Diversity: all types should be represented (functional, emotional, social)

Example

For a freelancer time-tracking app:
Critical (Push > 0.7):
— "Forget to start the timer" (functional, 12 mentions, weight 95)
— "Don't know where my time went this week" (functional, 9 mentions, weight 85)
Medium (Push 0.4-0.7):
— "Embarrassed to invoice by estimate" (social, 6 mentions, weight 65)
— "Feel like I'm working inefficiently but can't prove it" (emotional, weight 60)
Low (Push < 0.4):
— "Want a nice report for the client" (functional, 3 mentions, weight 35)

Tips

Upload a Telegram chat export from your target audience — AI will extract pains from real conversations
Quote real user words (in quotes) — this increases the fact weight
For each pain, try to identify its type — this helps understand which Job is behind it

Job Statements

Methodology: Product DNA Job Statement

The Job Statement is the fundamental unit of analysis in Product DNA. The formulation describes not what the product does, but what the user is trying to achieve in a specific context.

Canonical formula: "When [situation/trigger], I want to [action], so that [desired outcome]"

Each Job Statement belongs to a specific level in the Job Hierarchy:

Level	Definition	Example
Deep Need	Fundamental need that drives a person's life	Financial independence and freedom
Big Job	Major life/work outcome	Build a stable freelance income
Core Job	Specific task for which the product is hired	Accurately track time per project
Small Job	Sub-tasks within the core job	Start timer, categorize task
Micro Job	Atomic UI interactions	Tap the "Start" button in one tap

Job Types

Each job is classified by type. The type determines how to work with it:

Type	Definition	Example	What to Do
Ordinary	Direct action toward a result	"Buy a jacket for a hike"	Solve better than competitors
Orientational	Understanding, choosing, evaluating	"Find a CRM for my team"	SEO traffic capture point
Tax	Arises from defects of the current solution	"Re-entering data because export broke"	Kill through automation
False	No behavioral evidence	"Someday I want to learn Chinese"	Filter out of the roadmap

Important

Orientational jobs = search queries. If your product solves an orientational job ("which CRM to choose?"), that's an organic capture point: content on blogs, YouTube, SEO. Tax jobs = bugs/friction in current solutions. A product that kills a competitor's tax job gets a powerful Push.

Job Properties

Each job has three operational properties that influence product strategy:

Property	Definition	Scale	Product Impact
Sequence	Position in the job chain	Beginning → Middle → End of chain	Early jobs = commoditize; late jobs = capture value
Frequency	How often it occurs	Daily / Weekly / Monthly / Quarterly / Once	High frequency = habit, retention
Virality	Is it performed with/for others?	Solo / Shared / Social	Shared/Social = organic growth, referrals

Killable Jobs

Jobs earlier in the chain that can be automated or eliminated are "killable jobs." A product that "kills" preceding jobs captures more value. Example: Yandex.Rent killed the jobs of inspection, deposit, and contract — everything before "move into the apartment."

Full Job Statement Example

Core Job (ordinary, daily, solo):
"When I finish my workday, I want to know exactly how many hours I worked on each project, so I can invoice my client fairly and not feel guilty about overcharging."

Emotional Job: Feel in control of my time and income.
Social Job: Look professional to clients with detailed reports.
Trigger: End of workday, when hours need to be logged.
Current Solution: Google Sheets + phone timer.
Execution Criteria: Fast (< 3 min), accurate (± 5 min), no manual input.

Job Map (Job Graph)

Methodology: Job Graph

The Job Map is a visual representation of the job hierarchy and connections between jobs. This is not an ordinary mind map — it's a graph with typed edges and levels. At the center is the Core Job, with connections up (Big Job, Deep Need), down (Small Jobs, Micro Jobs), and sideways (Related Jobs, Alternative Solutions).

Edge types in the graph:

Edge Type	Meaning	Example
parent-child	Hierarchical job decomposition	Big Job → Core Job → Small Job
switch-trigger	Push force driving the switch	"Excel formulas break" → Core Job
desired-pull	Pull force attracting to the solution	"Automatic tracking" → Core Job
blocking-inertia	Inertia preventing the transition	"3 years on Excel, used to it"
blocking-anxiety	Fear preventing the transition	"What if I lose my data?"
competitor-satisfies	Competitor solves this job	Toggl → "Time tracking"

The key Product DNA strategic insight: moving one level up in the job graph is the most powerful competitive strategy. A competitor can copy a feature (Small Job), but cannot easily copy the position "we handle the entire outcome" (Big Job).

How to Use the Graph

Start with the Core Job — this is your current position
Look up: what Big Job is behind the Core Job? If you can solve the entire Big Job — that's more powerful
Look down: which Small/Micro Jobs can be "killed" (automated)?
Look sideways: which Related Jobs arise before/after? If you solve them too — that's lock-in

Graph Example

Deep Need: Financial freedom
└── Big Job: Build a stable freelance income
    └── Core Job: Accurate time tracking per project
        ├── Small Job: Start/stop timer
        ├── Small Job: Categorize by project
        └── Small Job: Generate report
    └── Related Job: Invoice the client
    └── Related Job: Project management
Competitors: Toggl (Core Job), Notion (Related), Excel (Non-consumption)

User Segments (ABCDX)

Methodology: Job-Based Segmentation + ABCDX

In Product DNA, a segment = people with the same jobs + same execution criteria in the same context. This is fundamentally different from demographic segmentation. A 25-year-old freelancer and a 55-year-old CEO can be in the same segment if they share Core Jobs and success criteria.

Segment Schema

Each segment is described using a standard schema:

Segment Description Format

Segment: [Descriptive name, not demographic]

Core Jobs (1-4):
— When [context], I want to [action], so that [outcome]

Big Job: I want [major outcome], so that [deep motivation]

Execution Criteria: [how they judge success: fast, cheap, reliable, prestigious, simple]

Who these people are:
— Role: [position / life role]
— Situation: [life/work context, stage, trigger]
— Decision type: [individual / committee / influencer]

Contextual Differentiators:
— Business type / life situation
— Organizational changes / external triggers
— Stakeholder dynamics (B2B: who else is involved)

Execution Criteria — Key Differentiator

Different segments may have the same Core Jobs but different execution criteria. A freelancer wants "fast and free." An agency wants "reliable and integrable." Same Core Job (time tracking), but different product solutions.

ABCDX Classification

After defining segments, each is evaluated using the ABCDX model — by economic attractiveness:

Grade	Definition	Action	Revenue Share
A	Ideal fit: jobs match, criteria match, willing to pay, fast decision	All-in: product, marketing, sales — for them	40-50%
B	Good fit: most jobs match, some friction, still profitable	Serve well, optimize conversion	30-40%
C	Marginal: some jobs match, low WTP, long sales cycle	Deprioritize — they consume 80% of support	5-10%
D	Poor fit: jobs don't match, negative ROI from serving them	Actively decline or redirect	0%
X	Unknown: insufficient data for classification	Research priority — run RAT tests	Variable

Critical Rule

80% of revenue should come from A+B. If C+D consume > 20% of team time — actively cut them. Common mistake: C-segment complains loudly, and the team spends resources on them instead of the A-segment.

B2B Extension: Decision-Maker Personal Jobs

In B2B, purchases are driven by the decision-maker's personal jobs, not just business objectives:

Business Objective	Decision-Maker's Personal Job (Real Driver)
"Reduce support costs by 30%"	"Close my KPI to keep my position"
"Increase team productivity"	"Get promoted to VP by year end"
"Migrate to the cloud"	"Don't be the one blamed if something breaks"

Rule

In B2B segments, ALWAYS include at least one decision-maker personal job as a Core Job. Google (2009): decision-makers buy expensive products 8× more often when personal benefit is obvious.

TAM/SAM/SOM per Segment

Each segment should include a market size estimate:

Metric	Definition	Calculation
TAM	Total Addressable Market — everyone performing these jobs	Number of people/companies × annual spend on this job category
SAM	Serviceable Available Market — reachable with current model	TAM × % in target geography × % matching delivery model
SOM	Serviceable Obtainable Market — realistically capturable in 1-2 years	SAM × realistic market share (typically 1-5% for a startup)

Segment Attractiveness Score

AI calculates a composite attractiveness score:

Formula

Attractiveness = (Value × 0.30) + (Profitability × 0.25) + (Scalability × 0.25) + (SOM × 0.20)

Value (1-10): How much better do we solve their jobs vs current solutions?
Profitability (1-10): Frequency × ARPU × retention potential
Scalability (1-10): Channel accessibility × purchasing habit × segment growth
SOM (1-10): Realistic size of capturable market

Persona Cards

Methodology: JTBD Personas

Unlike traditional personas (age, hobbies, "likes coffee in the morning"), JTBD personas focus on jobs, context, and criteria. A persona is a typical segment representative described through their tasks, not demographics.

Each card includes:

Core Job + Emotional/Social Jobs
Trigger — situation when the need arises
Current Solution — what they use now
Execution Criteria — how they judge success
Hiring Criteria — by what criteria they choose a product
Firing Criteria — when they abandon a product
Budget/WTP — willingness to pay
Switch Forces — Push, Pull, Inertia, Anxiety for this persona

Example

Persona: Anna, freelance UX designer
Core Job: Accurate time tracking per project
Emotional Job: Feel in control and fairness
Social Job: Look professional to clients
Trigger: End of workday, need to log hours
Current Solution: Google Sheets + phone timer
Execution Criteria: Fast (< 3 min), accurate, effortless
Hiring Criteria: "Timer launch should take less than 3 seconds"
Firing Criteria: "If I forget about the app for 3 days straight"
WTP: $10-20/month
Push: Sheets formulas break, manual entry is forgotten
Pull: Saw a colleague's auto-tracking — convenient
Inertia: 3 years of Sheets, used to it
Anxiety: "What if I lose data during migration?"

Interview Script

Methodology: Product DNA Switch Interview

AI generates a script for in-depth interviews using the Product DNA methodology. The key difference from standard custdev: questions target past behavior (what they did, how much they paid, when they decided to switch), not future intentions ("would you buy?").

Product DNA Golden Rule

Jobs are studied based on past spending — NEVER based on future intentions. "Would you buy?" = useless data. "How much did you pay for a solution last time?" = real signal.

Script Structure

The script is built around the 4 Switch formula forces and 19 data points:

Warm-up (2-3 min) — establishing rapport, consent to record
Context and Trigger (5-7 min) — DP01-DP02: "When did you first think about a solution? What happened?"
Current Solution + Inertia (5-7 min) — DP03: "How do you solve this now? How much time/money do you spend?"
Push Forces (10-15 min) — DP04-DP05: "What was unsatisfactory? When did it become unbearable?"
Pull Forces + Consideration Set (5-7 min) — DP06-DP08: "What did you consider? How did you learn about the solution?"
Anxiety + Decision (5-7 min) — DP07-DP08: "What were you afraid of? What convinced you?"
First Use + AHA Moment (3-5 min) — DP09-DP10: "First experience? When did you realize it works?"
Outcomes + What Told Others (3-5 min) — DP11-DP12, DP19: "How would you describe it to a friend?"
Wrap-up + sales attempt (2-3 min) — real offer or CTA

19 Data Points

Each interview should aim to collect all 19 data points (see 19 Data Point Types). Data quality is determined by coverage:

Gold: DP01-DP19 all present, N ≥ 5 interviews
Silver: DP01-DP12 present, N ≥ 3 interviews
Bronze: DP01-DP07 present, N ≥ 1 interview
Experimental: < 7 DPs, N = 1

Interview Tips

Don't read the script word-for-word — use it as a guide for the conversation
Record interviews (with consent) and upload audio to chat for automatic analysis
After 3-5 interviews, regenerate the script — it will become more targeted
End the last 2-3 interviews with an actual sales attempt — this is the best validation

Respondent Recruitment

Methodology: Finding People with Real Experience

Critical recruitment rule: only people who have ALREADY PAID money for solving the relevant job or a competing solution. Stated intentions ("I would buy...") = useless data.

9 Recruitment Channels

#	Channel	Best For	Cost	Speed	Data Quality
1	Existing customers	Any product with users	Free	1-2 days	Highest — they already paid YOU
2	Personal network (team)	Rare expertise, B2B	Free	1-3 days	High — warm contact
3	Snowball referrals	After first 2-3 interviews	Free	Ongoing	High — pre-qualified
4	Telegram channels/chats	RU/CIS, B2C, SMB	$0-30	3-7 days	Medium
5	LinkedIn + email	B2B, international	$0-50	5-14 days	Medium-High
6	Conferences/webinars	Niche professionals	$0-150	7-14 days	Medium
7	Paid panels	Urgent, USA/Europe	$50-200/interview	1-5 days	Medium (incentive bias)
8	Recruiting agencies	B2B, C-level	$100-500/interview	7-21 days	Medium-High
9	B2C agencies	Local markets	$30-100/interview	5-10 days	Medium

Screening: 5 Mandatory Filters

Screening Questionnaire

1. Have you paid to solve this task? (mandatory — no exceptions)
— What exactly? How much? When?

2. In the target segment?
— Does the role/situation match? Does the context (trigger, stage) match?

3. Can they describe PAST behavior?
— "Tell me about the last time you..." If only "I would..." → decline

4. Recency: was the job performed in the last 6 months?
— Older = memory distortion (Kahneman: peak-end rule)

5. Decision autonomy: did this person make the purchase decision?
— B2B: were they the decision-maker or just a user?

Recruitment Anti-Patterns

Anti-Pattern	Why It's Wrong	Fix
Recruiting from your own circle	Homogeneous sample, confirmation bias	Use 3+ different channels
Accepting "I would buy..."	Stated intent ≠ behavior	Only past payment experience
Only interviewing fans	Survivorship bias — you don't see churn	Include churned users + competitor users
Skipping screening call	30-60 min wasted on unsuitable candidates	5-minute phone screening
Too high incentive	Attracts professional "survey takers"	$20-50 B2C, $0-100 B2B

Channels by Product Stage

Pre-MVP (0 users): Personal network → Telegram/LinkedIn → snowball
Early (1-50 users): Customer base (priority!) → snowball → Telegram
Growth (50+ users): Customer base (segmented by ABCDX) → targeted outreach for X-segment

Auto Research

Beyond manual data entry, AI CPO conducts automated niche research. Click the "Research" button in the chat or type "Research the [niche name] niche."

How It Works

Niche matching: NicheMatcher — keyword LIKE match → LLM fallback → find/create niche
Cache check: If the niche was already researched and data is fresh — returns instantly (zero network calls)
Query expansion: LLM generates search queries for DuckDuckGo and Telegram
Channel discovery: Finds Telegram channels (filter: subscribers ≥ 50, posts fresher than 90 days, bot filter)
Post collection: Parses 2-3 pages from each channel (~60 posts, pagination via before=)
Article search: Finds articles on vc.ru, Habr (JSON state + generic readability)
LLM filtering: AI keeps only pain-relevant insights
Iteration: If insights < 5 — research repeats (up to 3 rounds)

Caching

Research results are cached for all users (shared corpus). TTL: posts 24h, articles 7d, insights 90d. Repeat research of the same niche may be free if the cache is fresh.

Cost

Auto research costs 500 credits. Use it after initial context collection via chat — then AI will search for insights relevant to already collected data.

Previous Quick Start Next Strategy