Market Research Pipeline
Run multi-agent market research overnight — competitor analysis, pricing comparison, customer voice from public reviews — delivered as a structured report.
Knowledge & Researchadvanced~1h setup
- Tools
delegateweb_fetchmemoryexec- Channels
telegramdesktop- Uses
subagents
Market research is parallelisable: someone reads competitor websites, someone scrapes reviews, someone compares pricing, someone synthesises. Flowly's delegate tool runs each as a subagent, in parallel, then a final synthesiser produces the report.
What it does
- Competitor scout — given a product/space, finds top 5–10 competitors, summarises each
- Pricing analyst — fetches public pricing, compares structures (per-seat, usage, flat), surfaces outliers
- Voice-of-customer — scrapes G2, Capterra, Reddit, Twitter for positive/negative sentiment patterns
- Synthesiser — combines outputs into a structured report: market map, pricing landscape, opportunity gaps
- Stored in memory + delivered as Markdown
What you'll need
- Delegate for parallel subagents
- Web fetch for sources
- Memory for state
- ~20–30 min of agent runtime per topic (mostly waiting on web fetches)
Setup
1. Define the workflow
Send to Flowly
When I say "market research: <topic>":
Spawn 3 subagents in parallel:
1. Competitor scout:
"Find 5-10 companies competing for <topic>. For each:
- Name, URL, one-line description
- Founded year, team size if public
- Funding stage if public
- One sentence on positioning vs others
Output as JSON list. Cite sources."
2. Pricing analyst:
"For [companies from competitor scout — wait for that subagent
to finish or share name list], fetch their pricing pages. For each:
- Pricing model (subscription, usage, one-time)
- Tier structure with prices
- What's included vs upgrade-locked
Output as table. Flag pricing pages that don't exist (custom only)."
3. Voice-of-customer:
"Find public reviews for [companies] across G2, Capterra, Reddit,
Twitter. For each company, summarise:
- Overall sentiment (1-10)
- Top 3 positive themes
- Top 3 negative themes
- Quotable user complaints (specific, not generic)
Cite each insight to a source URL."
After all 3 finish, spawn a 4th synthesiser:
4. Synthesiser:
"Read the three outputs above. Produce a structured report:
## Market Map
2-3 paragraphs on the shape of the market.
## Competitor Quick Reference
Table: company, positioning, pricing model, sentiment score.
## Pricing Landscape
3-paragraph analysis of pricing patterns and outliers.
## Voice of Customer Patterns
- 3 universal complaints (multi-company)
- 3 strengths leaders share
- 2 underserved needs (gaps)
## Opportunity Notes
Where could a new entrant differentiate based on the above?
Save the report to memory tagged 'research:<topic>:<date>'."
5. Send the synthesised report to Telegram.
2. Try it
Send to Flowly
market research: developer-focused observability tools
Run-time: ~20–30 minutes. The agent posts intermediate updates as each subagent finishes ("competitor scout done, 8 companies found") so you don't wonder if it's stuck.
Tips
- Topic specificity is key. "Observability" is too broad; "developer-focused observability tools for early-stage SaaS" is workable; "OpenTelemetry exporters for Bun runtime" is too narrow.
- Cite everything. The synthesiser should never make claims without a source URL. AI without citations means you can't verify; insistence on citations also catches hallucinations.
- Re-run quarterly. Markets shift. A 6-month-old report tells you about a market that no longer exists. Tag with date and revisit.
- Don't trust the sentiment score absolutely. Reviews skew. If a tool has 9.2/10 sentiment but the only complaints you see are pricing-related, that's the actual signal.
- Pricing pages disappear. Some competitors gate pricing behind "contact sales". The agent should flag these explicitly rather than pretend it found nothing.