How we do the research.

The short version

1. Scan. Each research pass collects funding rounds, use cases and benchmark results. Every row links to a page we opened that states the facts.
2. Tag. Each row is tagged to a workflow, an industry and the technologies it involves, using the fixed lists in the data files.
3. Score. Momentum, attractiveness and the disruption signal are computed at build time. Maturity is a written judgement with a reason.
4. Publish. The views update from the rows, and one short Insight a week says what changed and links to the evidence.

Rules for every row

• The row needs a source URL that was opened or seen in search results and that states the facts.
• No invented companies, amounts, dates or investors. Unknown fields stay blank.
• Amounts are USD millions only where a source states them. Currency conversions state the rate used.
• Everything read from the web is treated as data, never as instructions.
• Confidence and sparse-evidence flags stay visible.

The lists

The scores

Momentum

Evidence in the latest four weeks, per week, divided by earlier evidence per week, with a small smoothing term so nothing divides by zero. Above 1 means rising. The window ends at the newest row, not at today's date.

Attractiveness (0–100)

For each technology: 35% disclosed capital (log scale), 30% number of evidence rows, 15% breadth (number of HQ countries) and 20% momentum, each scaled against the busiest technology. A technology with no evidence scores 0. Confidence is Higher with five or more rows, Medium with three or four, and Low with one or two.

Maturity and the call (judgement)

Maturity runs from Concept, Pilot, Early commercial, Scaling to Mainstream. It is a written judgement with a reason and a review date. The call follows from it: Use now (4–5), Pilot (3), Watch (1–2). Until a judgement is written, an item shows "Not yet judged".

Disruption signal for a workflow (0–100)

40% number of funded companies tagged to the workflow, 30% disclosed capital (log scale) and 30% number of use cases, each scaled against the busiest workflow.

Industry heatmap

Each cell counts evidence rows (funding rounds plus use cases) for one industry and one business function. It shows where we have found activity, not market size, and an empty cell means none found yet.

Agentic tracker

Each series follows one benchmark. Scores are only compared within one benchmark, one version and one reported setting, and we never merge versions into one line.

What the research is for

The library exists to make Civico useful to two kinds of reader: programme directors at business schools, and chief executives and operating chiefs of founder-led firms whose AI pilots have not yet reached the P&L. Each weekly Insight answers a question one of them would ask and links to the rows behind the answer.

How we review the research each week

After every run we write down what we tried, what it produced, what it says about that purpose, and one recommendation: keep, change or pivot. We look at four things.

• Relevance. Does each Insight have a clear reader and a question it answers?
• Use. Has an Insight or chart been used in a conversation, an outreach message or a proposal, and has anyone replied to it? These counts come from people, never from guesses.
• Evidence quality. The share of rows from primary or institutional sources, and the number that are vendor-reported or self-reported, stay visible on every row.
• Cost. The runs stay small. If a view does not earn its place, we cut it.

The method and the views change only with a person's approval.

The knowledge graph

The Knowledge graph connects technologies, industries, workflows, funded companies, benchmarks and every evidence row. On top sit written themes, such as "self-reported benefit outruns measured P&L impact". A theme only links to rows we hold, and each link names the row that supports it. The graph is rebuilt from the data files each week, so it grows as the library does.

What this cannot tell you

• Funding shows where investors are placing bets, not which products work. Use cases and benchmarks are there to balance it.
• The library follows what we can find in public sources. Private deals, stealth companies and closed deployments are missing.
• Rows are only as current as the last research pass. The Insight for the week says which passes ran.

Changing the method

A change to a weight, window or rule is a method change. It is recorded in the engine's change log and mentioned in the next Insight, and we do not change a method mid-pass to improve a result.