Market Sizing and Category Economics: Implementation Guide

A useful question identifies the action the model will inform. Examples include whether to investigate a workflow segment, how to sequence two buyer groups, or which unknown deserves a research budget. Avoid questions such as how big is AI, because no consistent buyer, offer, unit, or period can answer them. The brief should specify whether the output is current demand, a future scenario, an ecosystem value pool, or revenue available to a defined offer.

Assign a decision owner and model owner separately when appropriate. The model owner is accountable for sources, calculations, assumptions, and version history. The decision owner judges whether the evidence is sufficient for action. Also name reviewers for finance, product, legal, security, or market claims where their domains are material. This responsibility map prevents analytical polish from becoming implicit authority to approve spending or publish a conclusion.

Define the funded job, eligible buyer, included products and services, substitutes, geography, period, and currency. Add concrete inclusion and exclusion tests. If a product only drafts text, does it count? If consulting is bundled with software, where is that revenue recorded? If a buyer builds internally, is the cost a substitute signal or category vendor revenue? Resolving these questions early reduces overlap and later debate.

Ask two researchers to classify a short sample independently. Where they disagree, the rule is not operational enough. Refine the wording until the distinction can be applied consistently, while retaining ambiguous cases in an exception log. The goal is not to force every company into a category. It is to make the modeled boundary reproducible and to expose the cases where judgment materially influences the total.

Potential sources include official statistics, audited filings, regulatory records, procurement notices, primary buyer research, product documentation, pricing evidence, usage records, and clearly documented specialist studies. A source should be evaluated on authority, recency, method transparency, and correspondence to the target population. A rigorous source about all enterprises may still be a weak proxy for organizations with one particular workflow and readiness profile.

Store the original link or reference, publication date, measurement date, geography, currency, population, definition, sample method where disclosed, and exclusions. Mark whether the record is observed, derived, inferred, or forecast. Include a short fit note explaining how it enters the model. If only a secondary report is available, identify it as such and schedule retrieval or replacement of the primary evidence when the input is material.

Do not delete a source because it complicates the preferred thesis. Put comparable estimates in a reconciliation table and normalize their definitions where possible. A vendor survey about planned use, a procurement record about funded intent, and a renewal record about retained use describe different adoption stages. Their disagreement may reveal the distance between interest and durable demand rather than an error that should be averaged away.

Maintain a rejected-source log with the reason for exclusion, such as missing methodology, stale period, incompatible geography, category overlap, promotional conflict, or inability to recover the original evidence. This prevents a discarded number from reappearing in a later deck without context. It also gives reviewers a fair view of the research process and makes future refreshes more efficient when a source is updated.

Start with a population that matches the geography and period. Apply eligibility filters for buyer characteristics and the funded job. Apply readiness filters for data, process, authority, integration, procurement, and risk. Then model purchase timing. Keep each rate separate rather than using a single adoption factor, because each barrier suggests a different product, research, or go-to-market response. Show the population removed at every stage.

Connect the resulting buyer or workload count to the commercial unit. A contract may include a platform component, usage, implementation, and support. Avoid combining recurring and one-time revenue without labeling the period. Record whether price evidence comes from actual approved contracts, public offers, comparable products, buyer research, value logic, or a hypothetical assumption. Price should be a range when evidence does not support a point estimate.

Model direct delivery at the same unit used for revenue. Inputs may include model and tool usage, data, storage, monitoring, retries, human review, implementation, support, and other supplier-backed costs. Mark actual cost, allocated cost, estimated cost, and missing cost separately. A usage-heavy product priced only by organization can hide exposure if workload or exceptions grow faster than contract value.

Add channel reach, sales cycle, onboarding throughput, implementation capacity, service quality, retention, and capital constraints to the obtainable view. Market availability is not company capacity. If the provider could sign more customers than it can safely onboard, delivery becomes the ceiling. If demand is uncertain but delivery is repeatable, distribution may be the constraint. The model should identify the limiting system rather than calling every shortfall market risk.

Hypothetical model: a source register contains an illustrative population of 8,000 organizations. The team assumes 45 percent have the required operating profile, 40 percent have a qualifying workflow, 50 percent meet data and authority readiness, and 15 percent may purchase in the selected year. The resulting modeled buyer count is 108: 8,000 multiplied by 0.45, 0.40, 0.50, and 0.15. All four rates are assumptions awaiting evidence.

Assume a hypothetical contract composed of 30 recurring units, 10 usage units at expected workload, and 8 one-time implementation units. The first-year modeled revenue is 5,184 units for 108 buyers, while recurring-equivalent annual revenue is 4,320 units before retention or expansion. Keeping the components separate prevents one-time services from making the recurring market appear larger and allows workload changes to affect only the usage line.

Suppose the company can reach 90 of the modeled buyers, expects an illustrative 20 percent purchase rate among qualified evaluations, and can onboard 14 during the year. Reach and conversion imply 18 customers, but onboarding limits obtainable first-year customers to 14. At the same hypothetical 48 first-year units per customer, obtainable revenue is 672 units. These numbers demonstrate constraint order; they are not forecasts.

Assume direct supplier and runtime cost of 7 units, implementation labor of 9, support and review of 5, and an uncertainty reserve of 3 per customer. The illustrative contribution before broader operating costs is 24 units per first-year customer. The team should then vary usage, exceptions, onboarding effort, price, and retention independently. The model is useful when it reveals what must be learned, not when it presents the base case as destiny.

Have a reviewer reproduce key calculations, trace material outputs to sources, test denominator consistency, inspect currency and period treatment, and search for overlap across category layers. A separate claims review should examine public wording, especially statements about category leadership, growth, customer outcomes, adoption, pricing, security, or regulatory posture. Unsupported certainty should be removed or recast as a labeled scenario, question, or internal hypothesis.

Red-team the thesis with alternative explanations. Observed interest may reflect novelty rather than budget. Vendor activity may reflect fundraising rather than retained demand. Job postings may indicate internal build preference. A lower supplier cost may invite more usage rather than improve contribution. Record which observations would support or weaken each explanation. The objective is not pessimism; it is a decision that remains defensible when favorable assumptions are challenged.

Assign refresh cadence by input volatility. Population data may change slowly, while provider costs, product boundaries, buyer controls, and purchase evidence may change quickly. Use a change log that records the old value, new value, evidence, owner, and impact on the decision. Preserve prior model versions so a later reviewer can distinguish genuine learning from silent revision.

Predefine triggers for expansion, another test, or pause. Examples include qualified paid demand, retained workload, acceptable exception burden, credible buyer value, manageable supplier variance, and delivery within authority boundaries. Stop or narrow if evidence cannot establish the buyer, price, outcome, cost, or serviceability required by the thesis. A market model is a learning instrument, not a reason to continue an offer that fails its declared conditions.

A strategy or market-intelligence owner should maintain the research and model, while an accountable executive owns the decision. Product, finance, go-to-market, legal, security, and claims reviewers participate when their boundaries are material. Deliverables include the brief, category rule, source and rejection registers, model, scenario comparison, sensitivity ranking, economics view, review record, limitations, failure conditions, and refresh schedule.

The guide does not establish future growth, attainable share, a guaranteed price, customer value, product availability, or profitability. Each conclusion remains bounded by its sources and assumptions. Qualified professionals should review material investment, accounting, competition, legal, and public disclosures. Where observed evidence is absent, label the model input as hypothetical or inferred and restrict the decision to the level of uncertainty the organization can responsibly carry.

Within current verified capabilities, OmegaOS can help route a bounded market question through Hermes research, source evidence, approved task context, accountable review, Forge work, economic observation, and Mnemosyne learning. The useful bridge is continuity from assumption to decision and later comparison. The platform does not supply missing primary research, determine the correct market boundary, or approve a consequential commercial claim on behalf of responsible people.

Start with a narrow model and a named next action. Attach the evidence packet, record the model version, assign approval and stop conditions, and observe what the experiment actually reveals. Verify package access, connector authorization, source rights, and data handling for the intended configuration. When evidence changes, preserve both versions and the reason. That operating discipline keeps market sizing connected to learning without turning OmegaOS into the subject of an unsupported category claim.