Opportunity Discovery: A Practical Guide for Builders Who Don't Know What to Build Next
The Problem in Concrete Terms
You have the skills to build things. You have evenings and weekends available. You might even have a small budget for tools or ads. But every time you sit down to start a new project, you hit the same wall: you genuinely don't know what's worth building.
This isn't "analysis paralysis" in the abstract sense. It looks like:
- Opening a blank Notion page titled "Ideas" and writing down three things before realizing two are features, not products, and the third is "Uber for X" but you can't articulate X
- Scrolling Product Hunt or Indie Hackers and thinking "that's already done" or "I'd never use that" or "too crowded" until you've burned 90 minutes with zero output
- Having a vague sense that "B2B SaaS is where the money is" but no specific problem you've personally witnessed a business struggle with
- Keeping a running list of "problems I notice" that stays stubbornly empty because you've trained yourself to ignore friction as background noise
- Feeling like everyone else found their opportunity through some magical insight you missed, while you're stuck waiting for lightning to strike
The result: months pass. You build nothing. Or you build something nobody wants because you picked the first idea that didn't immediately feel stupid.
Why the Usual Fixes Fail
1. "Scratch your own itch"
Solid advice if you have itches that map to paying markets. But most developers' personal itches are: better dotfiles, a vim plugin, a CLI tool for a niche workflow. These make great portfolio pieces. They rarely make rent.
2. "Talk to customers"
You can't talk to customers until you know who they are and what context they operate in. Cold outreach to "small business owners" gets you ghosted. You need a hypothesis first.
3. "Look at trending topics / AI wrappers / no-code tools"
Chasing trends means competing on speed with people who have more resources and less scruples. By the time you see a trend on Twitter, the easy money is gone.
4. "Just build something and iterate"
Iteration requires signal. If you build a solution to a problem nobody has, you'll iterate forever on the wrong thing. Motion ≠ progress.
5. Idea generation frameworks (SCAMPER, Jobs-to-be-Done, etc.)
Frameworks are useful after you have raw material. They don't generate the raw material. You can't SCAMPER a blank page.
Step-by-Step Solution
Phase 1: Build Your Observation Muscle (Weeks 1-2)
1.1 Keep a friction log, not an idea log
Carry a physical notebook or use a notes app with a single rule: record friction, not solutions. Every time you think "this is annoying," "why does this take so long," "I wish I didn't have to," write it down. No filtering. No "could this be a business?" judgment.
Target: 50 entries in 14 days. If you hit 10, you're not noticing—you're filtering. Lower the bar. "The microwave beeps three times and I can't disable it" counts. "My team's standup runs long" counts. "I can't find the right Notion template" counts.
1.2 Shadow a non-technical workflow for 2 hours
Find someone whose job involves repetitive computer work: an executive assistant, a recruiter, a property manager, a freelance designer. Ask to sit next to them (or screen-share) while they do their least favorite recurring task. Say nothing. Take timestamped notes on:
- Every context switch between tools
- Every copy-paste between systems
- Every "I have to remember to..." moment
- Every spreadsheet that functions as a database
- Every time they say "then I have to..."
You're not looking for ideas yet. You're collecting evidence of manual work that software could absorb.
1.3 Audit your own recurring expenses
Pull your last 12 months of credit card statements. Highlight every SaaS subscription, contractor payment, and tool purchase over $20/month. For each, ask: "What problem does this solve? What would I do if this disappeared tomorrow? What part of this workflow happens outside this tool?"
The gaps between tools are where opportunities live.
Phase 2: Filter for Signal (Week 3)
2.1 Cluster your friction log
Group your 50+ entries by:
- Context: work vs. personal, industry, role
- Frequency: daily, weekly, monthly, one-off
- Cost of failure: "annoying" vs. "I lose money/clients/reputation"
- Workaround existence: spreadsheet, manual process, "I just remember," "I pay someone"
Discard clusters that are: purely personal preference, one-off, or already solved by a tool you'd actually pay for.
2.2 Apply the "willing to pay" filter
For each remaining cluster, answer honestly:
- Who specifically feels this pain? (Name a title, not "people")
- What happens if they don't solve it? (Quantify: lost hours, lost revenue, compliance risk, churn)
- What's their current workaround costing them? (Time × hourly rate, or tool cost + gap cost)
- Do they have budget authority or influence? (A junior dev feels CI pain but can't buy; an engineering manager can)
Keep only clusters where: the pain is recurring, the cost of inaction is measurable, and the buyer is identifiable.
2.3 Run the "manual concierge" test
Pick your top 2-3 clusters. For each, design a manual service you could deliver this week to 3 real people:
- No code. No landing page. No automation.
- You do the work by hand (spreadsheets, emails, Loom videos, phone calls)
- Charge something—even $20—to test willingness to pay
Example: If your cluster is "agencies waste hours resizing creatives for each platform," your concierge test: "Send me your master creative and target platforms; I'll return resized files in 4 hours for $15/set." Post in 2-3 relevant communities. DM 10 people who've complained about this.
If nobody pays, the problem isn't real enough. If they pay, you've validated demand and learned the exact workflow.
Phase 3: Define the First Build (Week 4)
3.1 Map the happy path
From your concierge test, document the exact steps you performed manually. Identify:
- The trigger (what starts the process)
- The inputs (files, data, decisions)
- The transformation (what you did to the inputs)
- The output (what the customer received)
- The exception cases (what broke or required judgment)
3.2 Slice the thinnest valuable vertical
Don't build the full workflow. Build one trigger → one output that delivers the core value. Everything else stays manual.
If the concierge was "resize creatives for 6 platforms," v1 might be: "Upload one image → get Instagram + LinkedIn sizes automatically." Skip TikTok, Twitter, Facebook, Pinterest. Skip batch upload. Skip folders. Skip user accounts—use magic links.
3.3 Set a 2-week build constraint
Calendar it. Two weeks from today, you ship to your 3 concierge customers. No extensions. If scope doesn't fit, cut scope—not timeline.
3.4 Define "done" for v1
Write a one-sentence acceptance criterion: "A user can drag a 1920x1080 PNG onto the page and receive a zip with 1080x1080 and 1200x628 versions within 60 seconds, with no login required."
That's your spec. Build to that.
Phase 4: Validate or Kill (Weeks 5-6)
4.1 Deliver to concierge customers
Send them the link. Watch them use it (Loom or live). Note where they hesitate, what they expect that's missing, what they don't use.
4.2 Ask the retention question
After 3 days: "Would you be disappointed if this disappeared tomorrow?" (Sean Ellis test). If <40% say "very disappointed," you have a vitamin, not a painkiller. Decide: iterate on core value, or return to Phase 2 with new clusters.
4.3 Price it
Ask: "What would you pay for this monthly?" Don't suggest numbers. If they say "I don't know," ask