Build or buy: the decision criteria that matter most

Which criteria matter most in a build or buy decision?

Five criteria carry most build or buy decisions: how often the decision repeats, which skills your team holds, how fast the first score must arrive, how much control of the equations you need, and how sensitive the data is. Each one has a measurable form. Frequency is a yearly count, skills are named people, speed is a date, control is a yes or a partial, and sensitivity is a tier of confidentiality.

These five build on the three paths of building, buying and licensing, and add a weight to each. A criterion earns its place when it changes the answer for your team. Add a sixth only when a real constraint appears, such as a procurement rule or a deadline set by a board, and write the evidence beside it so that a reviewer can follow the score back to a fact.

How does decision frequency change a build or buy decision?

Frequency sets the scale of the effort. A build carries a fixed cost, because the model, the data flows and the validation exist whether you run 1 decision or 100. A purchase carries a variable cost, because each decision has its own fee. Count the decisions you expect in the next 12 months and write that number at the top of your sheet.

The contrast becomes concrete with a price. A Decision is priced at US$910 and delivered in 48-72 hours, so 10 decisions in a year mean 10 fees of that size (EQOS Asia pricing page, as of October 2026). Compare that line with the hours your modellers would spend in the same year. A low count favours buying, and a high, steady count favours a build or a licence.

Which skills does a decision engine build require from your team?

A build needs three roles that work together: a modeller who designs the equations, a data engineer who keeps the inputs current, and a domain expert who agrees on the definitions. Name a person for each role and the share of the week that person can give. A role listed with 0 hours counts as a gap, and the sheet should show it as one.

Buying moves the skill requirement to your side of the table: you supply facts, documents and a clear question. The EQOS Asia workflow accepts internal documents, interviews and existing data in any format, digital or physical (EQOS Asia method page, as of October 2026). Rate your team high on this criterion when it can name a fact owner and answer provider questions within a day.

How fast do you need the first score from a decision engine?

Speed is a date on a calendar. Write down when the decision must be taken and count back the days you need for briefing, collection, computation and debrief. A build must also add the time for design and validation, which makes speed the criterion where building scores lowest on a first decision. Rate each path by whether it meets your date, and not by its average pace.

EQOS Asia publishes its timings. A Reading is delivered in 24-48 hours, a Decision in 48-72 hours, and a Study, which activates the 3 pillars of audit, modelling and prediction, in 5-7 days (EQOS Asia pricing page, as of October 2026). The standard workflow has 4 steps, from NDA and scoping to the deliverable and debrief (EQOS Asia method page, as of October 2026).

How much control and verifiability should a decision engine give you?

Control means the right to see and change the equations. A build gives it fully, and your team can adjust a weight the day a market changes. Verifiability is a related criterion: can a reviewer replay a verdict and obtain the same result? Ask for that replay before you rate any path, because a score you can reproduce is a score you can defend to a board.

A purchase gives a different mix. In the EQOS Asia method, the Codex Universalis core is deterministic, and every verdict can be recomputed identically by a data scientist with access to the Codex under NDA, while the expanded forms and weighting coefficients stay proprietary (EQOS Asia method page, as of October 2026). Rate control as partial and verifiability as high for that path.

What does a weighted build or buy decision example look like?

Take 5 criteria with weights that sum to 100, all figures illustrative: frequency 30, skills 25, speed 20, control 15 and data sensitivity 10. Rate each path from 1 to 10, where a higher number always favours the path. Building scores 9 on frequency, 4 on skills, 3 on speed, 9 on control and 8 on sensitivity. Buying scores 4, 9, 9, 5 and 7.

The weighted total for building is 0.30 × 9 + 0.25 × 4 + 0.20 × 3 + 0.15 × 9 + 0.10 × 8, which equals 6.45. Buying gives 1.20 + 2.25 + 1.80 + 0.75 + 0.70, or 6.70. The gap is 0.25 points, small enough to call for a test on one real decision before any commitment, as the next section describes.

How do you keep a build or buy decision reversible?

Set 3 exit rules before the first order: the review date, the score that triggers a change of path, and the files you take with you. A first purchase at the lowest tier, or a build prototype on one past case, keeps the commitment small and gives you a real number for effort. Review the weights after that first result and rerun the sheet.

For a purchase, ask how your data comes back. Under the EQOS Asia NDA, the receiving party returns or destroys confidential information within 10 business days of a written request (EQOS Asia NDA page, as of October 2026). Start with the free 30-minute pre-diagnostic and the Λ simulator to view an output before ordering, and use the pricing page to see the next tiers.

Frequently asked questions

About the author

Stéphane Jambu

President, EQOS Forecast Asia

Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile