Build vs Buy AI Agent Scorecard: Decision Factors and TCO Assumptions

This software archive preserves version 1.0.0 of a build, buy and hybrid AI agent sourcing scorecard. It contains the runnable decision model, a registry of 12 weighted factors, editable three-year total cost of ownership (TCO) assumptions and eight reproducible synthetic scenarios. Pharos Production AI engineering developed the tool to make missing evidence, operating ownership and mandatory constraints visible during an AI agent sourcing review.

Use this release to reproduce a decision record, inspect the rule that produced an outcome and replace the illustrative cost inputs with your own documented assumptions. It is intended for engineering leaders, product owners and procurement teams comparing an internally built agent, a bought platform or a hybrid architecture.

Decision factors and evidence gates

The factors cover strategic differentiation, workflow uniqueness, data control, integration depth, time to value, team capability, lifecycle ownership, budget shape, scale economics, portability, evaluation control and the compliance path. Their total weight is 31. Known answers take values of -1, 0 or +1. Unknown answers retain a null value and do not enter the scoring denominator.

Calculate the index as 100 times signed known points divided by known weight, rounded half away from zero. BUY applies at -30 or lower, BUILD at +30 or higher and HYBRID between those thresholds. A decision requires at least 75% known-weight coverage and known answers for data control, team capability, lifecycle ownership and compliance. Missing evidence returns UNRESOLVED. A constraint that blocks the numeric outcome produces CONSTRAINT_REVIEW_REQUIRED and a provisional HYBRID boundary. A result within seven points of either threshold carries a boundary warning.

TCO assumptions and expected states

Three-year TCO uses the formula one-time total + (annual total × 3) + exit reserve. Eight cost categories separate discovery, implementation and data preparation from recurring platform, internal operations, evaluation and support costs, then add switching costs at the horizon.

Archived illustrative USD inputs produce totals of USD 2,100,000 for build, USD 1,990,000 for buy and USD 1,940,000 for hybrid. These amounts are synthetic planning assumptions. They are not quotes, market averages or forecasts. The arithmetic uses a constant annual amount without discounting, inflation, taxes or a benefits model. The lowest modeled cost does not override an unresolved or blocked sourcing path.

Four archived presets are supplemented by fixtures for a missing gate answer, insufficient coverage, a blocked buy result and a near-build threshold. Each scenario includes its inputs and expected JSON, CSV and Markdown exports. The reproduction script checks the expected decision status, index, known weight and TCO totals against the archived implementation.

Files, provenance and reuse

The ZIP contains all 50 tracked source files from commit c8ad4067654a6777cea3b652a80f62dc6aeb21fb, unchanged, plus a reproduction README, fixture generator, expected exports, citation metadata, a license manifest and SHA-256 checksums. Run npm run verify inside source/ and node reproduce.mjs from the package root with Node.js 22 or newer. The source verification suite passes 14 tests. The fixture generator checks all eight scenarios.

Code is licensed under MIT. Original registries, their mirrored copies and the new fixture data are licensed under CC BY 4.0. Consult the license manifest for scope. Linked third-party works retain their existing rights. This is an archive of an authored decision aid, not a new empirical study. Its weights and thresholds have not been statistically calibrated. The original source registry retains its August 25, 2026 verification dates.

This deposit does not claim a fresh review of every cited external work.

Companion resources

Read the build vs buy AI agent decision guide for the wider sourcing discussion. The guide uses a separate, simplified six-factor scale. Reproduce this release with its archived 12-factor registry. Try the interactive AI agent sourcing scorecard or inspect the versioned scorecard source repository.

If your comparison still lacks a production integration boundary, evaluation plan or operating owner, review the scope of AI agent development services from Pharos Production before converting the score into a delivery commitment.