Demand Validation

Workflow record

Field	Value
Bottleneck	Demand validation research takes hours, if not days. Automate this task. Human just needs to go/no-go decision.
Inputs	Reddit, Hacker News, niche forums, review sites, competitor pages, Keyword Planner data, landing-page results, operator notes, and existing product artifacts.
Agent role	Normalize source material, generate demand queries, collect search metrics, cluster demand pockets, surface competitor gaps, draft hypotheses, and package evidence for review.
Human role	Pick the segment, judge evidence quality, approve the hypothesis, set the budget, and decide whether the next test is worth running.
Known limits	The agent can measure and organize signals, but it cannot decide strategy, interpret bot-mediated interactions as buyer intent, or spend money without approval.

First useful pilot

A weekly demand signal brief for one offer or segment that turns scattered market evidence into a clear go/no-go packet.

The pilot should answer:

• Is there measurable demand for this problem language?
• Which segment or demand pocket looks most reachable?
• Which incumbent gap or customer complaint is worth testing?
• What is the next validation instrument: landing-page smoke test, problem interviews, meetup/webinar probe, pricing test, or concierge pilot?

Workflow shape

1. Intake a product brief, target segment, user role, repository or product artifact, and any source notes the operator already trusts.
2. Generate a scoping brief and seed keywords that describe the problem in customer language.
3. Expand those seeds into compact Keyword Planner phrases, retry with broader language when the first pass shows no volume, and record zero-demand searches instead of hiding them.
4. Cluster validated keywords into demand pockets with monthly search volume, competition index, representative keywords, opportunity signal, and capability-fit rationale.
5. Scan adjacent demand so the operator can see nearby problems the original frame missed.
6. Generate product candidates for each demand pocket, including conventional extensions and less obvious pivots.
7. Research competitors and substitutes for the chosen candidate, looking for pricing gaps, turned-away customers, stale product lines, review failures, and forum complaints.
8. Draft falsifiable hypotheses and recommended tests tied to specific keywords, segments, competitor gaps, and success criteria.
9. Prepare the selected experiment artifact: landing-page copy, ad keywords, campaign assets, event topic, interview brief, or another test-family packet.

Evidence packet

The review packet should be compact enough for a founder to read before a decision call:

• Scoping brief and target segment.
• Source snapshot with links, notes, and confidence level.
• Demand pockets ranked by search volume, competition, buying intent, and founder capability fit.
• Representative keywords with monthly search estimates and competition index.
• Adjacent demand signals that may suggest a better segment.
• Competitor gap summary: pricing gaps, underserved customers, neglected lines, and complaint themes.
• Landing-page signal quality: likely human visits, agentic bot traffic, form submissions, reply quality, and any follow-up proof that a real buyer exists.
• Proposed hypothesis with primary assumption, test family, expected signal, minimum sample, budget cap, and stop condition.

Decision gates

The agent should not treat "interesting" as "validated." It should move work through explicit gates:

Gate	Pass signal	Human decision
Demand exists	Search, forum, review, or interview evidence repeats the same problem language.	Continue, reframe, or stop.
Segment is reachable	The buyer can be targeted through search, communities, events, lists, or direct outreach.	Choose the first segment to test.
Gap is credible	Incumbents leave a pricing, workflow, service, or trust opening the offer can actually exploit.	Approve the positioning angle.
Test is bounded	The next test has a clear KPI, budget or time cap, and rejection threshold.	Run the test or ask for more evidence.

Post-agent landing-page caveat

Landing-page experiments do not mean what they meant five years ago. A page can now receive enough traffic, clicks, form fills, chat interactions, or waitlist signups from agentic browsing, automated research tools, and bot-mediated workflows to look like a go signal.

That does not make the signal useless. It means the test has to distinguish market attention from buyer intent.

For landing-page experiments, the packet should separate:

• Human-looking sessions from known automated traffic.
• Raw submissions from qualified replies.
• Bot-assisted interactions from buyer-owned follow-up.
• Curiosity clicks from budget, urgency, authority, or workflow pain.
• Form fills from conversations that survive one human review step.

The better go signal is not "the page converted." It is "a reachable buyer took a costly-enough next step after the page." That may mean replying to a follow-up, booking a call, answering a problem-specific question, joining a live event, accepting concierge onboarding, or describing the workflow in enough detail to prove the pain is real.

Agent boundaries

Good autonomy for this workflow:

• Draft and refresh research packets.
• Pull structured search-demand data.
• Generate and broaden keyword sets.
• Cluster signals into demand pockets.
• Draft hypotheses and experiment briefs.
• Flag missing evidence and contradictory signals.

Human approval required:

• Selecting the market segment.
• Claiming an idea is validated.
• Publishing landing pages.
• Launching paid campaigns.
• Changing budgets.
• Contacting prospects.
• Choosing the product direction.

First build

Start with one narrow offer and one weekly cadence. The useful first version does not need to run every validation method. It needs to reliably produce a founder-ready demand brief that includes:

• 40-60 generated search phrases.
• Keyword Planner results with retries and zero-demand reporting.
• 3-8 demand pockets or opportunity areas.
• Adjacent demand signals.
• A short competitor-gap scan.
• Landing-page instrumentation that labels automated traffic separately from human-qualified responses.
• One recommended next test with success and stop criteria based on buyer intent, not raw conversions alone.

The first build is successful when the human reviewer can make a go/no-go/reframe decision from the packet without spending another afternoon reconstructing the research trail.

Ready to test this workflow?

Send the current bottleneck, source material, and first human-owned decision.

I need this agent