“The flywheel is turning” sounds appealing, but it can make an unproven business look as if it has already succeeded. A cycle in which customers grow, data accumulates, and enterprise value rises does not happen automatically. Contracts may not turn into cash, large volumes of sensor data may not become evidence of reductions, and technical demonstrations do not guarantee repeatable revenue.

In this article, AI Safety Korea’s Cash, Evidence, and Data do not claim any current level of revenue, number of customers, reduction results, or investment performance. They are a management hypothesis to consider when designing a livestock-methane monitoring and climate-data business. Each flywheel needs starting conditions, metrics, stopping criteria, and verification stages. Only when real data confirms all three loops can they be said to contribute to enterprise value.

A misconception to correct first: does more data make the flywheel turn?

A flywheel is not the repetition of activity volume; it is a structure in which value returns as an input to the next stage. Even if installing more sensors produces more data rows, the Data flywheel is not turning unless customer decisions or verification quality improve. A demonstration report is not an Evidence flywheel if the result cannot be reproduced under the same conditions or reviewed externally. Even after a contract is signed, it is difficult to call it a Cash flywheel unless the cash remaining after installation and maintenance costs is reinvested in the next product improvement.

The three flywheels also move at different speeds. Cash can be managed monthly, while climate effects in livestock may require observations that cover seasons and production cycles. Data may arrive every second, but it does not mature into evidence without a prepared baseline and comparison group. The loops therefore must not be compressed into one growth rate. Each loop should be measured separately, and the transitions between them must be shown to occur in practice.

Enterprise value is not simply the sum of revenue, data records, and reports. It reflects many judgments, including the potential for future cash flows, the reproducibility of technology and operations, regulatory and market risks, data rights, and the team’s execution capability. This framework is not a formula that guarantees enterprise value. It is a tool for translating the conditions of a business that investors and partners can trust into operating practice.

Cash Flywheel: look for “repeatable cash” before revenue

The Cash flywheel hypothesis is that cash earned by delivering customer value is reinvested in product reliability, installation efficiency, and customer support, and that these improvements lead back to healthier contracts and collections. The first step is to break down what is being sold. Equipment sales, installation, recurring subscriptions, data analysis, maintenance, demonstration support, and assistance with verification documents have different cost structures and degrees of repeatability.

Looking only at revenue can hide advance hardware purchases, site visits, calibration gas, communications fees, and equipment replacement costs. For each contract, track the order value, recognized revenue, actual cash received, direct costs, support hours after installation, and payback period. Separate government grants and contingent carbon revenue from confirmed customer cash. Credits not yet issued and estimated reductions must not be treated as cash.

At an early stage, metrics that confirm repeatability matter more than one large contract. Check whether paid demonstrations convert into full contracts, installation time decreases, maintenance hours per site stabilize, customers renew, and contract terms secure the access rights needed for data collection and verification. If growing revenue causes cash to burn faster with each installation or sacrifices essential data rights, the flywheel is turning in the wrong direction.

The reinvestment rules for the Cash flywheel must also be explicit. Decide whether cash collected from customers should first go to sensor quality, the data pipeline, cybersecurity, field training, or methodology review. Repeatedly performing excessive custom development for short-term revenue can fragment the product and reduce the comparability of data. Distinguish between customer requests that can become standard features and one-off work for which an adequate price must be charged.

Evidence Flywheel: make verification stages an asset, not the claim itself

The Evidence flywheel hypothesis is that a clear hypothesis and measurement plan produce trustworthy results, which then open access to better partners, sites, and contracts and improve the quality of the next verification. Evidence here means traceability between a claim and its data, not the thickness of a report. It must be possible to reconstruct what was compared at which farm, for which group of animals and period, and under which feed, ventilation, and sensor conditions.

First distinguish the purpose of a demonstration. Technical verification of whether a sensor works in the field, efficacy verification of the causal effect of low-methane feed, operational verification of whether farm operations are repeatable, and methodology review of whether carbon-program requirements can be met are different questions. Claiming that one small demonstration proved all four overstates the evidence.

Manage evidence through acceptance criteria for each stage. Separate hypothesis formulation, laboratory or controlled-condition testing, a limited field demonstration, repetition across broader seasonal and farm conditions, independent review, and verifiability under a particular program. At every stage, define the required sample, baseline, comparison group, measurement period, data completeness, uncertainty, and approver. Even when the expected result is not achieved, classifying the cause without hiding the data and improving the next design creates a learning asset.

As in the GHG Protocol’s project-accounting perspective, a reduction claim must clearly define the baseline scenario and project scenario, boundary, period, and monitoring. Do not confirm an emissions reduction from concentration changes alone; also consider ventilation rate, herd size, production, feed intake, and seasonal changes. Label preliminary reduction estimates, completed internal review, completed external verification, and credit issuance as separate states.

The Evidence flywheel also places guardrails on sales language. Connect every claim on the website, in proposals, and in investor materials to its evidence level and permitted scope. Do not apply an observation from one farm to all farms, present a development target as a commercial outcome, or make an agreement look like completed technical verification. More precise claims may be less impressive in the short term, but they reduce long-term review costs and losses of trust.

Data Flywheel: turn raw sensor readings into reusable learning assets

The Data flywheel hypothesis is that field data whose quality has been confirmed improves analysis and products, and that improved products generate more stable data. Raw data, however, does not automatically become an asset. Sensor values need metadata describing time, location, equipment, calibration, units, measurement range, ventilation, and operating conditions.

For barn methane data, concentration and emissions must be distinguished. ppm is the mixing ratio of methane in air, while kg CH₄ is the mass that crosses a system boundary over a given period. Converting concentration into emissions requires flow or ventilation rate, temperature and pressure, moisture, and a spatial boundary. A high concentration at one point does not mean high emissions from the entire farm, and greater ventilation can lower concentration even when the generation rate remains the same.

Measure data quality by usability rather than collection volume. Track sensor availability, missingness by channel, time-synchronization error, calibration validity, handling of anomalous values, lineage between original and corrected versions, and completeness of field metadata. If an algorithm removes anomalies or interpolates missing values, record the rule and version and preserve the original. Arbitrarily excluding data to produce a favorable result damages the Evidence flywheel as well.

Data rights and security are prerequisites for the flywheel. Contracts should define what data the farm, feed company, platform, and verification body own and to whom they may provide it for which purposes. Minimize personal and commercially sensitive information to what the purpose requires, and manage access rights and retention periods. Because sensor hacking or manipulation of time data can distort carbon performance, security controls and change records should also be operating metrics.

AI models can accelerate the Data flywheel, but they are not automatic arbiters of truth. If training data is concentrated in particular farms and seasons, performance may fall at other sites. Record the model version, input range, validation data, false alarms and missed detections, human review, and safe fallback procedures. The governance, measurement, and management perspectives of the NIST AI Risk Management Framework can serve as a reference for addressing responsibility and impacts in addition to model performance.

Where the three flywheels connect

The three loops are not a linear pipeline that each turns once in sequence. A paid demonstration contract can contribute to Cash while producing new field data. But without measurement conditions and data-use rights in the contract, it does not lead to Data. Even accumulated field data does not turn into Evidence without a baseline and quality controls. Even strong evidence does not return to the next Cash cycle if it fails to solve the customer’s cost, regulatory, or operational problem.

The links between loops therefore need “transition gates.” Moving from Cash to Data requires checking installation and operating costs, data-access rights, and collection of required metadata. Moving from Data to Evidence requires reviewing representativeness, comparability, missingness, uncertainty, and reproducibility. Moving from Evidence to Cash requires checking the permissible scope of claims, customer value, contract terms, and verification costs.

A good connection might work as follows. Analysis of field incident records standardizes the installation procedure, reducing both visit costs and missing data. More complete data lowers uncertainty in reduction estimates, and reviewable reports help the customer obtain internal approval. If the customer pays for that value and renews the contract, the collected cash is reinvested in equipment and quality management. Every arrow must be confirmed with actual metrics.

There are dangerous connections as well. Presenting a preliminary reduction estimate as externally verified to attract investment may improve the prospect of short-term Cash, but it damages trust in Evidence. Repeating customer-specific data conversions without records may generate revenue while destroying Data comparability. Rapidly expanding sensor deployment while cutting calibration and security budgets increases the quantity of data but reduces its long-term asset value.

Stage-specific KPIs and stopping criteria

Cash KPIs include monthly cash balance, net burn, runway to the next milestone, cash-basis revenue, gross profit by contract, installation and support costs, paid-demonstration conversion, and renewal rates. Evidence KPIs include the number of preregistered hypotheses, acceptance-criteria attainment, comparable observation periods, independent-review status, evidence level for each claim, and unresolved data-quality issues. Data KPIs include active sites, sensor availability, completeness of required channels, time-synchronization quality, calibration compliance, the share of data with preserved lineage, and security incidents.

Stopping criteria are as important as growth targets. Stop the analysis and the claim if a safety threshold is violated or data integrity is in doubt. Withhold publication of reductions if missingness and confounders exceed predefined limits. Reconsider deployment expansion if support costs per site consistently exceed the contract value or the customer does not permit essential data access. Stopping is governance that prevents a larger error, not an attempt to hide failure.

A monthly review should look beyond the three loops’ headline numbers and identify the weakest link. Ask whether cash rose while data quality fell, whether data grew without answering verification questions, or whether evidence became stronger without connecting to the customer’s reason to buy. Allocate the next month’s resources not only to increasing the largest number, but also to removing the bottleneck.

Operating checklist

  1. Have Cash, Evidence, and Data been documented as management hypotheses to test rather than current achievements?

  2. Have contract value, actual collections, contingent income, and expected carbon revenue been separated?

  3. Is the full cost of installation, calibration, communications, and support known for each site?

  4. Are the question and predefined acceptance criteria for each demonstration clear?

  5. Are preliminary reductions, internal review, external verification, and credit issuance separated?

  6. Are sensor readings linked to metadata on time, location, equipment, calibration, ventilation, and husbandry conditions?

  7. Do contracts address data ownership, purposes of use, access rights, and retention periods?

  8. Are versions and change histories preserved for models and data-processing rules?

  9. Are the transition gates and responsible parties between the three flywheels defined?

  10. Are there criteria for stopping or withholding claims based on safety, integrity, and economics?

  11. Does the monthly review allocate actual budget and staff to the weakest link?

Conclusion: enterprise value comes from verified connections, not three numbers

Cash, Evidence, and Data are not a formula that automatically explains AI Safety Korea’s enterprise value. They are a strategic hypothesis that a cash-generating business, trustworthy evidence, and reusable data can reinforce one another. Each loop must be measured independently, and transitions between loops must pass contractual, quality, and verification gates.

Integrity, not speed, is the core of this structure. Do not exaggerate evidence for revenue, sacrifice quality and rights for data volume, or halt customer-value validation while waiting for perfect research. The flywheel becomes a structure for enterprise value only when the connection in which cash funds better measurement, better data creates stronger evidence, and that evidence returns as real customer trust and willingness to pay is confirmed repeatedly.

Sources