GTM Insight Generation: Experimenting with Deep Research +(Ir)Replaceability of Human Insight

I spent two hours leveraging deep research to generate GTM intel for a hardtech startup I have no connection to, and I'm giving away the prompts and process.

I spent two hours leveraging deep research to generate GTM intel for a hardtech startup I have no connection to and I'm giving away the prompts and process.

Why?

• Experiment and evaluate with these novel deep research tools (I'm evaluating different options)

• I needed to document the process for standardization/operationalization

• permissionless GTM is fun and I want to see what feedback I can generate

Some Learnings Around Deep Research:

• Each deep research tool I leveraged was a huge force multiplier and liability in their own special way. (I am planning on going more into detail on these nuances as I get more hands on time)

  • OpenAI Deep Research: cites too few sources to produce as many tokens are is does for my liking - This seems to be addressable via prompting but low visibility into the CoT can be a blocker

  • Perplexity R1: will have extremely dense source citation but will attribute key figures and claims or infer without disclaimers, etc. - addressable by having it verify the claims it made step by step

  • Open source versions: generally produced reports that were more reliable but primarily because they were shorter/surface level - need more hands on time but this seems like a fairly straightforward fix by increasing total num of iterations and iteration depth (tokens)

• Now that these tools are consistently citing sources research rigor is more important than ever - and I mean the research rigor of the person driving the research tool.

Some Learnings Around the Replaceability & Irreplaceably of Human Insight:

• The value emerges from engaging with the process. The process is the point, and there is no substitute for the process. No AI or other human being can engage in your unique process - it is axiomatically impossible.

• You can use these AI tools to help accelerate the process, but you must know what parts are augmentable and automatable, and where you need to make those connections yourself

• I wrote more about this here a while ago:

  How to Think About When to Use Gen AI: Where is the value? The Process Itself, or the Outcome?

Full Two Hours:

Permissionless GTM Intel Gen: Gecko Robotics - Watch Video

I took my entire transcript and and fed into Claude to map the operationalized process I developed to timestamps in the loom.

The Process - GTM Intelligence Generation: Operational Workflow

Phase 1: Initial Surface Area Creation

[AUGMENTABLE/AUTOMATABLE]

• Multi-tool deep research deployment

• Initial data gathering

• Source verification and tracking

• Basic market mapping

• Value chain initial analysis

Phase 2: Iteration Space ("ma" [look it up])

[NON-AUTOMATABLE]

• Deep engagement with material

• Pattern recognition

• Initial insight formation

• Problem space understanding

• Identification of potential directions

Phase 3: Volume Game Selection & Execution

[HYBRID]

Selection [NON-AUTOMATABLE]:

• Choose appropriate volume game based on insights

• Define parameters and constraints

• Set quality requirements

Execution [AUGMENTABLE]:

• Generate multiple hypotheses

• Map potential data sources

• Identify problem spaces

• Create initial datasets

Phase 4: Convergent Action

[AUGMENTABLE/AUTOMATABLE]

• Feed selected hypotheses back to deep researchers

• Systematic evaluation against criteria

• Data point verification

• Pattern validation

• Initial synthesis

Phase 5: Final Evaluation Space

[NON-AUTOMATABLE]

• Deep evaluation of generated output

• Strategic insight formation

• Value chain implications

• Go-to-market strategy refinement

• Final synthesis and recommendations

Phase 7: Insight Operationalization

[AUGMENTABLE/AUTOMATABLE]

• Transform strategic insights into actionable plans

• Define data collection and transformation workflows

• Map implementation steps

• Create execution roadmaps

Examples from transcript:

• NDT Specialist Dataset Creation:

  • Source identification

  • Data collection methodology

  • Transformation pipeline

  • Actionable TAM dataset structure

• Nuclear Power Plant Intelligence:

  • Value chain mapping

  • Stakeholder identification

  • Data source cataloging

  • Implementation sequencing

Implementation Notes

For Augmentable/Automatable Components:

• Maintain clear attribution chains

• Set explicit quality requirements

• Define confidence thresholds

• Establish verification protocols

For Non-Automatable Spaces:

• Protect time for deep engagement

• Maintain flexibility for insight emergence

• Allow for pattern recognition

• Enable strategic thinking

For Hybrid Components:

• Clear delineation between human and machine tasks

• Defined handoff points

• Quality control mechanisms

• Feedback integration processes

---

Phase 1: Initial Surface Area Creation [0:00-5:53]

• Setting up multiple research tools: "I have these four ones I'm going to be playing with today"

• Initial prompting for broad research

• Establishing source quality requirements: "constraining the model and, like, making sure it doesn't hallucinate or cite for sources"

Phase 2: Iteration Space [5:53-25:00]

• First engagement with returned research

• Questioning source validity: "Where does source 25? This is actually, like, so exciting"

• Initial pattern recognition about their ICPs and inspection capabilities

Phase 3: Volume Game Selection & Execution [25:00-57:00]

• Decision to generate hypotheses: "you're asking it to touch the specks and what it works"

• Using AI to generate "hundred, or as many hypotheses as you can"

• Pattern matching across sources

Phase 4: Convergent Action [57:00-1:26:00]

• Focus on NDT specialist staffing challenge

• Deep dive into security clearances and requirements

• Initial exploration of maritime opportunities

Phase 5: Final Evaluation Space [1:26:00-1:35:00]

Key breakthrough moment at 1:26:32: "Wow, that's a crazy supply chain. That's why. It's not just maintenance... It's generating, uhm, structural data on existing platforms... to use in producing better ships in the future."

Phase 6: Final Non-Automatable Evaluation [1:35:00-1:42:00]

• Reassessment of initial hypotheses based on Navy data feedback loop insight

• Strategic implications for their focus on military contracts

• Evaluation of commercial maritime opportunities in light of new understanding

Phase 7: Insight Operationalization [1:42:00-end]

• Using AI to map out data sources for NDT specialist dataset

• Planning nuclear power plant intelligence gathering: "Let's think more about that, um, step by step"

• Designing implementation workflows: "the messy details of inspections and all stuff like that will depend on data availability"

---

End Product - Gecko Robotics: Market & Opportunity Analysis

Core Problem Statement

Critical infrastructure operators face significant challenges in conducting thorough, safe, and efficient inspections of their assets. Traditional "Joe on a rope" manual inspection methods are dangerous, time-consuming, and provide limited data coverage, leading to missed defects, unnecessary downtime, and potential catastrophic failures.

Value Proposition

Gecko Robotics provides an integrated solution combining wall-climbing robots and enterprise software (Cantilever®) that:

• Captures 1000x more data than traditional methods

• Completes inspections up to 10x faster

• Eliminates human exposure to dangerous environments

• Enables predictive maintenance through AI-powered analytics

• Reduces unplanned downtime (worth $500k+/hour in some facilities)

Ideal Customer Profiles (ICPs)

1. Defense Assets

   • U.S. Navy & shipyards

   • Focus: Ship hulls, submarines, critical infrastructure

   • Validation: 400% increase in robot usage (2024)

   • Key Metric: 90.9% reduction in inspection time

2. Power Generation

   • Utilities operating large power plants

   • Both conventional (coal, gas) and nuclear

   • Average plant age: 45+ years

   • 12,500+ utility-scale plants in U.S.

3. Oil & Gas/Petrochemicals

   • Refineries and processing facilities

   • ~700 refineries worldwide (130 U.S., 80 Europe)

   • Unplanned downtime cost: $149M per site

   • Strong regulatory/safety drivers

Total Addressable Market (TAM)

• Global NDT & Inspection Market: $10-16B (2024)

• Inspection Robots Segment: $1.25B (2022) → $7.1B (2029)

• North America: 37% of NDT market

• U.S. NDT Market: $5.2B (2023)

Current Likely Critical Challenges & Opportunities

1. Defense Contractor Staffing

• Rapid growth (400% increase) straining qualified technician supply

• Security clearance requirements limiting talent pool

• Specialized expertise needed (NDT + robotics)

• Current evidence: Active job postings for NDT specialists

2. Nuclear Power Generation Market Expansion

• Gecko partners with the Navy not just to provide inspection services but to generate data for use in future shipbuilding

• The recent boon and demand for nuclear power presents a similar opportunity: highly regulated industry with need for both NDT + data for future asset development (future nuclear power plant development could greatly benefit from historical inspection data)

Near-Term

1. Cleared Talent Pipeline Development

   • Build database of qualified NDT technicians with clearances

   • Create strategic sourcing partnerships

   • Develop military transition program relationships

2. Nuclear Power Plant Inspection + Data TAM Intelligence

   • Operating plants vs. decommissioned/restart candidates

   • Age of infrastructure

   • Proximity to major data center developments

   • Current inspection providers and contracts

   • Maintenance schedules and outage history

3. Commercial Maritime GTM Strategy (explored but not pursued after insight into value of data generated from inspections, this inspired pivot to similar play for Nuclear Power)

   • TAM expansion analysis

   • Pilot program design

   • Value proposition adaptation

   • Partnership strategy development

Strategic Benefits

• Accelerated market penetration

• Reduced hiring bottlenecks

• Diversified customer base

• More efficient resource utilization

• Scalable growth model

Market data sourced from published reports and company announcements. Specific figures and growth rates verified through multiple sources.

---

Cleared Talent TAM Painpoint:

Staffing for Defense Contracts

• Hypothesis: The rapid growth in defense contracts (with a documented 400% increase in Navy usage) is likely to strain the supply of technicians with the required security clearances.

• Supporting Data: Our research noted that defense assets (e.g., naval ship inspections) are growing quickly, and such high‑security roles require specialized clearance and training.

Key Public Data Sources (All Accessible):

1. Primary High-Value Sources:

• ASNT's public directory of certified NDT professionals

• USAJobs.gov historical postings for NDT positions (shows clearance requirements)

• Department of Labor O*NET data on NDT technician employment

• Military exit data by MOS/Rate codes (Navy inspectors, etc.)

• LinkedIn data for professionals with NDT certifications

2. Enrichment Data:

• Bureau of Labor Statistics wage data by region

• State occupational employment statistics

• Defense contractor facility locations (public records)

• Navy shipyard employment statistics

• Professional association membership data

The 80-20 Approach:

1. Step 1: Core Dataset Build

• Pull ASNT directory data (gives us base of certified professionals)

• Map against geographic locations of Gecko's key facilities/customers

• Add salary bands from BLS data

• Layer on security clearance likelihood based on work location (i.e., near defense facilities)

2. Step 2: Basic Enrichment

• Add LinkedIn profile matches for validation

• Cross-reference with defense contractor locations

• Add wage data by region

• Map against facility/customer locations

Expected Output:

• Total number of NDT certified professionals by region

• Subset likely to have/be eligible for clearances based on work location

• Salary expectations by region

• Heat map of talent density vs. Gecko's needs

---

Nuclear Power Plant Inspection + Data TAM Intelligence

Hypothesis: The convergence of aging nuclear infrastructure (45+ year average age) and increasing power demands from AI/ML data centers creates an urgent need for advanced inspection capabilities, particularly as plants like Three Mile Island plan restarts and life extensions.

Supporting Data: U.S. has 12,500+ utility-scale power plants with critical inspection needs, costing $500k+/hour in downtime. Nuclear plants specifically require more frequent and thorough inspections due to safety regulations and aging infrastructure. The Department of Energy projects significant increases in power demand from data centers, with nuclear power being positioned as a key solution for reliable baseload power.

Additional Validation Points:

• Three Mile Island restart announcement by Constellation Energy

• Average nuclear plant age of 45+ years cited in EIA data

• Microsoft, Google, and other tech giants announcing major data center expansions

• Nuclear Regulatory Commission's increased focus on aging plant management

Process:

• Map all U.S. nuclear facilities and ownership structures

  • Operating plants vs. decommissioned/restart candidates

  • Age of infrastructure

  • Proximity to major data center developments

  • Current inspection providers and contracts

  • Maintenance schedules and outage history

• Identify Key Decision Makers & Influencers

  • Plant operators

  • Engineering firms handling inspections

  • Nuclear regulatory compliance officers

  • Safety oversight committees

  • Digital transformation leaders

• Build Nuclear-Specific Value Props

  • Radiation exposure reduction metrics

  • Digital twin creation capabilities

  • Predictive maintenance for aging infrastructure

  • Life extension data analysis

  • New plant design insights

• Partnership Opportunities

  • Nuclear engineering firms

  • Plant construction companies

  • Nuclear safety consultants

  • Digital transformation providers

  • AI/ML data center developers needing power

Nuclear Power Plant TAM Buildout Process

1. Base Dataset: Nuclear Plants

• Source: NEI.org plant listings

• Core Data Points:

  • Plant name & location

  • Operational status

  • Capacity/size

  • Age/commissioning date

  • Planned restarts/decommissions

2. Account Layer Mapping:

A. Owner/Operator Companies

• Utility companies

• Parent corporations

• Operating subsidiaries

• Data Sources:

  • Company websites

  • SEC filings (10-K, 10-Q)

  • State utility commission documents

  • Press releases

  • News articles

B. Organization Structure for Each Account:

• C-Suite

• Operations/Maintenance Leadership

• Plant Management

• Engineering Directors

• Safety/Compliance Heads

• Data Sources:

  • LinkedIn company pages

  • Corporate websites

  • Annual reports

  • Industry news

  • Conference speakers/participants

3. Data Structure:

Plant Level:
- Plant Name
- Location
- Status
- Capacity
- Age
|
Account Level:
- Parent Company
- Operating Company
- Subsidiaries
- Ownership %
|
Org Structure Level:
- Key Departments
- Leadership Roles
- Reporting Lines
- Decision Makers

4. Web Research Workflow:

• Start with plant list from NEI.org

• For each plant:

  • Search company websites for org charts

  • Pull relevant SEC filings

  • Find news about management changes

  • Map current leadership through LinkedIn

  • Cross-reference industry news

5. Transformation Steps:

• Clean plant data

• Match companies to plants

• Extract org structures

• Map key roles

• Validate relationships

• Add contact data where public

• Tag decision makers

• Note reporting structures

---

Research Chats with Citations:

• https://www.perplexity.ai/search/you-are-an-expert-gtm-strategi-zHQO2ej5RI.IamdoAMM92g

• https://chatgpt.com/share/67aaa64b-eb4c-8006-9b89-1fa8f80fd87e

TL;DR: Deep research tools are are huge accelerants for intel gathering but need to be used to augment and not replace the process - in this case the process of insight generation.