Building Data Moats in Climate and Ag: A David Friedberg-Inspired Playbook for Startups

‍

David Friedberg gets one question more than almost any other in climate and ag tech: how do I build a defensible data moat that compounds over time.

I wrote this playbook to answer that question in plain language and give founders a practical path to building a durable advantage.

I will show you how to source unique climate data, turn it into proprietary signals, package it for enterprise sales, and protect it with smart contracts and product choices.

The goal is simple.

Build a moat that outlasts features, forces competitors to pay tolls, and compounds with every new customer and partner.

‍

1) Why David Friedberg’s thinking matters for climate and ag startups

I study David Friedberg because he operationalized something most founders only talk about.

He turned messy, commodity weather and agriculture data into decision-grade products that farmers and enterprises actually paid for.

His work at WeatherBill and The Climate Corporation was not just about better models.

It was about a better system for ingesting, cleaning, and compounding data into unique insights.

Here’s what I take from his approach.

   
• Start with outcomes, not algorithms.
   
• Use many small data advantages that add up to a big one.
   
• Make the feedback loop part of the product.
   
• Lock in recurring data streams with contracts and incentives.
 

If you adopt this mindset, every new field, sensor, and grower improves your product and widens the gap.

2) What is a data moat and why it outlasts product features

A data moat is a compounding advantage created by proprietary data and the models, processes, and contracts around it.

Features can be copied.

Data cannot be copied if it is uniquely sourced, context-rich, and contractually protected.

In climate and agriculture, a moat usually comes from four levers.

   
• Access: You collect data others do not have, or cannot legally use.
   
• Context: You label and structure the data more accurately than anyone else.
   
• Feedback: Your customers generate new truth data that improves your models.
   
• Time: Your historical archive compounds value and accuracy.
 

When you get these right, switching away from your product becomes risky and expensive for customers.

3) The four layers of a climate-ag data moat

I build moats in layers because it keeps the strategy simple and testable.

   
• Data acquisition: Satellites, weather stations, soil probes, machinery telematics, imagery, lab results, purchase orders, and field notes.
   
• Normalization and ontology: Standard schemas for plots, crops, growth stages, phenology, and events.
   
• Signal generation: Yield prediction, evapotranspiration, NDVI and EVI composites, soil moisture, disease risk, irrigation scheduling, and carbon MRV.
   
• Decision product: API, dashboard, integration, or insurance that delivers an outcome with SLAs.
 

Each layer strengthens the next, and your contracts and pricing tie them together.

For more on go-to-market mechanics, see our blog post: Pricing, Packaging, and PLG for B2B AI.

4) Sourcing hard-to-get climate and agriculture data ethically

Great moats start with ethically sourced, legally unambiguous data.

I avoid gray areas because they create future liabilities.

Here are sources that work.

   
• Public remote sensing: Sentinel, Landsat, SMAP, and ERA5 reanalyses.
   
• Partner networks: Co-ops, input retailers, agronomists, and irrigation dealers.
   
• Device-light telemetry: Smartphone surveys, photos, and Bluetooth scans from tractors and pumps.
   
• Enterprise systems: ERP, farm management software, and purchase records with user consent.
   
• Third-party datasets: Weather providers, soils maps, topography, and water rights.
 

I always secure explicit data rights, define permitted uses, and offer value back to the data originator.

If growers win, your data flywheel spins faster.

5) Turning raw data into proprietary signals that compound

Raw data does not build moats.

Signals do.

I focus on derived features that need your unique context to compute.

   
• Yield deltas normalized by hybrid, planting date, soil type, and precipitation bands.
   
• Field-specific evapotranspiration calibrated with on-farm truth data.
   
• Disease risk scores combining canopy moisture, temperature hours, and phenology.
   
• Irrigation timing recommendations from soil texture, weather, and pump telemetry.
   
• Carbon sequestration estimates with plot-level management histories for MRV.
 

Competitors can rent the inputs.

They cannot recreate your labels, ground truth, and historical corrections at scale.

6) Instrumentation strategies: hardware-light or hardware-led?

I choose my instrumentation strategy based on two constraints.

Customer friction and time to data density.

Hardware-light works when remote sensing and enterprise data give enough signal to start.

Hardware-led works when the key insight requires direct measurement.

Here’s how I decide.

   
• If a phone photo or simple weather station gets you 80% of the value, go light.
   
• If yield, soil moisture, or quality cannot be inferred reliably, deploy targeted sensors.
   
• Use installation partners to keep your CAC sane.
   
• Offer free or subsidized sensors tied to multi-year data rights.
 

In one irrigation project, we shipped low-cost flow meters.

We tied them to savings guarantees and got precise pump usage data for years.

7) API strategy: when to expose, when to hold back

I expose APIs to increase distribution and contribution while protecting my secret sauce.

I keep the rawest data and learned features private unless a deal compensates me for the risk.

My rules are simple.

   
• Expose decision endpoints: risk scores, recommendations, and alerts with SLAs.
   
• Throttle and log usage to detect abuse and model extraction attempts.
   
• Offer paid tiers for batch history, postbacks, and premium latency.
   
• Keep labeling tools and feature stores internal unless a partner contributes high-value data.
 

APIs are not just distribution.

They are collection mechanisms for feedback and truth data.

For more on platform thinking, see our blog post: API-First Go-To-Market for Data Companies.

8) Data rights, licenses, and contracts that protect your moat

Good contracts make strong moats inevitable.

I include specific clauses every time.

   
• License scope: Commercial, research, or internal use only.
   
• Derived works: You own the derivatives but cannot reverse engineer inputs.
   
• Attribution and data lineage: Clear provenance for audits and ESG reporting.
   
• Data contribution credits: Discounts or revenue share for partners who send high-quality data.
   
• Non-compete on model extraction: No training of competitive models without consent.
 

I also define data retention, deletion SLAs, and anonymization policies upfront.

It reduces sales friction later.

9) From pilots to enterprise sales: packaging data into outcomes

I do not sell data.

I sell outcomes with proof.

In agriculture, that usually means yield lift, input reduction, risk reduction, or time saved.

Here is my pilot design.

   
• Pick a crisp KPI: bushels per acre, water savings, or disease losses avoided.
   
• Set a counterfactual with A/B fields or historical baselines.
   
• Lock in a clear measurement plan and shared visibility.
   
• Offer a simple commitment: three months, defined fields, flexible expansion.
 

I package the enterprise deal as a fixed platform fee plus per-acre or per-facility usage.

I include a multi-year discount for data-sharing commitments.

For more on selling to the enterprise, see our blog post: Enterprise Sales Sprints for Technical Founders.

10) Pricing models that reward contribution and lock in value

My pricing rewards behaviors that compound my moat.

I price the decision, not the raw data.

   
• Per-acre or per-asset decision pricing with volume discounts.
   
• Data contribution credits for telemetry, labels, and field truth.
   
• Premium tiers for historical archives, lower latency, and support SLAs.
   
• Outcome-based fees for insurance-like products and guarantees.
 

Price transitions are pre-written.

When customers contribute data, their unit costs drop and retention rises.

11) Building trust with growers and enterprises

Growers are rightfully skeptical.

I earn trust with clarity and reciprocity.

   
• Plain-language data policies and opt-in consent.
   
• Immediate, tangible value: alerts, savings, or recommendations they can test.
   
• Transparency on model limits and uncertainty bands.
   
• Local agronomy partnerships to backstop recommendations.
   
• Simple off-ramps with data export for comfort.
 

When people feel respected, they share more data and stay longer.

12) Human-in-the-loop operations that strengthen your models

Human-in-the-loop is not a cost center for me.

It is a training engine.

I use agronomists, annotators, and QA teams to close gaps models cannot yet handle.

   
• Field scouting confirms disease detections and growth stages.
   
• Labeling teams validate satellite segmentation and cloud masks.
   
• Customer success escalations create new edge-case datasets.
 

I track annotation throughput, agreement rates, and model lift per labeled sample.

When the lift decays, I automate and redeploy people to the next bottleneck.

13) Evaluating model performance under climate volatility

Climate volatility breaks brittle models.

I design for shift, not for stability.

   
• Out-of-distribution tests across years, regions, and extreme events.
   
• Backtests with synthetic drought and flood scenarios.
   
• Error decomposition by crop, soil, topology, and management practice.
   
• Uncertainty-aware outputs so users can hedge decisions.
 

I share performance dashboards with customers.

Trust grows when you expose reality and update fast.

14) GTM partnerships that accelerate data flywheels

Partnerships compress data collection timelines.

I look for partners who sit near rich data exhaust and frequent decisions.

   
• Input providers who see seed, fertilizer, and chemical usage.
   
• Machinery OEMs with telemetry and CAN bus integration.
   
• Insurance carriers and reinsurers who crave better loss ratios.
   
• Retailers and commodity traders with quality and logistics data.
   
• Irrigation and water utilities with flow and allocation data.
 

I propose co-selling, data-sharing, and revenue sharing with clear metrics and attribution.

For more on partnerships, see our blog post: Designing High-Trust GTM Partnerships.

15) M&A and data-sharing deals: structure for defensibility

Not all data-sharing is equal.

I prefer structures that improve my moat without creating new competitors.

   
• Exclusive vertical rights in a crop or region for a period tied to contribution volume.
   
• Reciprocal licensing where only derived signals are shared, not raw data.
   
• Change-of-control clauses to protect your rights if the partner is acquired.
   
• Audit rights and kill switches if terms are breached.
 

When the cost to exit your ecosystem rises over time, your deal is working.

16) Regulatory tailwinds and compliance as a moat

Compliance can be a weapon, not a tax.

I lean into standards and reporting requirements that favor high-quality data and transparency.

   
• MRV for carbon markets and Scope 3 emissions by commodity and region.
   
• Water reporting and irrigation allocations in drought-prone basins.
   
• Food traceability, FSMA rulemaking, and chain-of-custody in supply chains.
   
• Privacy protections, consent management, and data residency.
 

When you build once for the hardest regime, you can sell everywhere.

17) Metrics for board reporting on your data moat

Boards do not want a lecture on model architectures.

They want a glide path to defensibility and revenue.

   
• Unique data coverage: acres, assets, or facilities with exclusive rights.
   
• Contribution rate: percent of customers sending back labels or telemetry.
   
• Signal quality: AUC, MAE, or lift versus baselines by cohort.
   
• Archive growth: months of historical depth per asset.
   
• Retention and expansion: net revenue retention linked to data value.
 

I include a simple red-yellow-green dashboard.

Green means the moat deepens as revenue grows.

18) Fundraising narratives investors believe

Investors hear “we have a data moat” every day.

I make it real with an evidence-based story.

   
• Show the unique sources and exact rights you own.
   
• Quantify compounding: each new customer adds X labels and reduces error Y%.
   
• Prove switching costs: how outcomes degrade without your history and ontology.
   
• Map milestones: from pilots to scaled enterprise contracts and partnerships.
 

I tie it to the big market tailwinds in climate risk, resilience, and sustainable agriculture.

For more on fundraising, see our blog post: The AI-Ready Data Room.

19) Common pitfalls and how to avoid them

I have made these mistakes and learned the hard way.

   
• Collecting data without a tight ontology, which makes it hard to use later.
   
• Overpromising accuracy in volatile seasons, which erodes trust.
   
• Giving away raw data in early partnerships, which dilutes your moat.
   
• Underinvesting in field truth, which limits model lift.
   
• Confusing a dashboard with a productized decision.
 

Avoid these and you will save a year.

20) The 24-month roadmap to a durable data moat

I work in two-week sprints, but I plan in 24-month arcs.

Here is a roadmap I have used with climate and ag founders.

   
• Months 0–3: Define ontology, secure first data rights, and ship a single decision API.
   
• Months 3–6: Launch pilots with A/B fields, instrument feedback, and measure outcomes.
   
• Months 6–9: Close first enterprise with contribution credits and SLAs.
   
• Months 9–12: Add human-in-the-loop loops and publish performance dashboards.
   
• Months 12–15: Strike two GTM partnerships with co-selling and reciprocal licensing.
   
• Months 15–18: Expand archive depth, introduce premium history tiers, and lower latency.
   
• Months 18–24: Lock multi-year deals tied to contributions and launch second decision product.
 

At 24 months, you should have a defensible, compounding asset that keeps competitors on the back foot.

Case study: a water-smart irrigation startup

Let me ground this in a simple story.

A founder I worked with started with remote sensing and public weather data.

They offered a basic irrigation scheduler with confidence intervals and a simple text alert.

They then subsidized a few flow meters and soil probes in Year 1.

They signed growers to multi-year contracts that exchanged data contribution for lower pricing.

By the end of Year 2, they had the largest private dataset of irrigation events and yields in their region.

They used the archive to build a premium drought-risk API that insurers paid for.

The insurer integration returned claim data, which further improved the risk model.

The moat deepened as they grew.

Generative Engine Optimization: making your content and API discoverable

Search is shifting to generative engines.

I now optimize product and content for retrieval and answer quality.

   
• Use clear problem-solution headers and FAQ structures.
   
• Publish API examples in simple language with copy-paste snippets.
   
• Add structured data, entity names, and precise definitions.
   
• Answer the top ten buyer questions on one page.
 

When an LLM crawls your docs, it should find authoritative, direct answers.

Privacy-preserving learning in agriculture

Grower privacy is non-negotiable.

I use privacy-preserving techniques to learn without leaking.

   
• Federated learning to train models at the edge and aggregate gradients.
   
• Differential privacy for aggregate stats and benchmarks.
   
• Strict role-based access and audit logs across all environments.
 

These choices open doors with cautious enterprises and regulators.

From signals to insurance and guarantees

When your signals are strong, you can underwrite outcomes.

I like parametric products tied to clean triggers and fast payouts.

   
• Heat stress thresholds for livestock and poultry operations.
   
• Rainfall and evapotranspiration gaps for irrigated crops.
   
• Cold-chain breaks for produce quality during transport.
 

Insurance-grade signals command higher margins and stickier contracts.

Designing the data contribution UX

Contribution should feel effortless.

I design low-friction flows that trade small actions for instant value.

   
• One-tap photo uploads with auto-labeling suggestions.
   
• SMS replies to confirm events like planting or spraying.
   
• Bluetooth detection for machinery presence and operation.
   
• Automated field-boundary detection and correction prompts.
 

Every contribution updates the model and shows the user the lift they just created.

Defining your proprietary ontology

Your ontology is your language for the world you model.

It must reflect how growers actually work, not just how databases like tidy tables.

   
• Represent operations as sequences: prepare, plant, irrigate, scout, treat, harvest.
   
• Bind actions to time, weather, and growth stages.
   
• Use standardized crop codes, hybrids, and product SKUs where possible.
   
• Track provenance and confidence for every label.
 

A great ontology makes every new dataset interoperable on day one.

Playbook summary: the five commitments

Here is the playbook I would sign in blood.

   
• Commitment to outcomes: sell decisions, not dashboards.
   
• Commitment to contribution: make feedback the default.
   
• Commitment to contracts: protect rights and reward data sharing.
   
• Commitment to trust: privacy, clarity, and real-world agronomy.
   
• Commitment to compounding: archives, ontologies, and model lift over time.
 

If you do these five things, you will build a moat competitors cannot easily cross.

FAQs

Here are the most common questions I hear from founders and operators.

   
•      What makes a “David Friedberg-style” data moat unique?
       It blends diverse data sources, strong ontologies, ground truth, and tight contracts into decision products that improve with scale.    
   
•      Do I need hardware to build a moat?
       Not always.      Start hardware-light if remote sensing and enterprise data give a strong signal.      Add targeted sensors only where inference breaks.    
   
•      How do I keep customers from churning to a cheaper copycat?
       Tie value to your historical archive, feedback loops, and multi-year contribution discounts.      Show how accuracy drops without your data history.    
   
•      Should I open my API early?
       Yes for decision endpoints with throttling and SLAs.      No for raw data and feature stores unless compensated by exclusive rights or revenue share.    
   
•      What KPIs should I report to investors?
       Unique coverage, contribution rate, signal quality, archive depth, and NRR tied to data value.    
   
•      How do I get growers to share data?
       Give immediate value, clear privacy, and price reductions tied to contribution.      Keep contributions one tap or one text away.    
   
•      How do I price my product?
       Price the decision per acre or asset with premium tiers for history, latency, and support.      Add outcome-based options where feasible.    
   
•      What about climate volatility breaking my models?
       Test out-of-distribution, use uncertainty-aware outputs, and keep human-in-the-loop to catch edge cases.    
   
•      Can compliance really be a moat?
       Yes.      Build for the strictest MRV and traceability standards and sell that competency.    
   
•      When should I consider insurance products?
       When you have stable, backtested signals and partners for capital and claims.      Start parametric with clean triggers.    
 

Conclusion

The climate and agriculture markets reward teams who turn messy, shared data into proprietary, compounding decision systems.

If you adopt a David Friedberg-inspired playbook, you will design for outcomes, build in contribution, and protect your advantage with smart contracts and pricing.

Do this for 24 months and you will have a data moat competitors cannot cross, a product customers trust, and a business Capitaly.vc would be proud to back.

If you found this useful, share it with a founder and subscribe for more.

Subscribe to Capitaly.vc Substack (https://capitaly.substack.com/) to raise capital at the speed of AI.

And remember, the fastest way to a durable competitive advantage in this space is to build a data moat the David Friedberg way.