Enterprise AI search work fails for different reasons than mid-market work does. The analysis is usually competent. The recommendations are usually right. And two quarters later almost nothing has shipped.
The constraint is not knowledge. It is the distance between a correct recommendation and a deployed change.
Map the approval path before the audit
Most enterprise programmes begin with a technical audit and discover the governance problem at implementation. Reverse it.
Before any analysis, document who approves a content change, who approves a template change, who approves structured data, and what evidence each of them requires. Legal, brand, security and platform engineering usually all have a veto, and they rarely block at the same stage.
A recommendation nobody can approve is worth less than a smaller one that ships this month.
One baseline per market, not one for the company
Assistants return materially different answers by language and region. A prompt set built for the UK market does not transfer to Germany, and a global visibility figure averaged across both tells you nothing actionable.
Run a separate baseline per market you actually sell into. It is more work at the start and it is the only way the numbers mean anything later.
The same applies to brands. If a group operates several brands in one category, they compete with each other inside generated answers, and reporting them as one entity conceals that entirely.
Corroboration is harder and more valuable at scale
Enterprises have advantages here that most never use deliberately.
- Press coverage that already exists but is never structured or linked
- Executives with genuine industry visibility whose content is unconnected to the brand entity
- Analyst mentions, filings and regulatory records that are highly citable third-party sources
- Customer reviews across multiple platforms with no owner and no process
Leave an audit trail
Every change dated, owned and measured. This sounds like bureaucracy and is actually the mechanism that makes the next approval faster than the last one.
The pattern is consistent: the first change takes six weeks of stakeholder management. The tenth takes a day, because a record exists showing that the previous nine shipped without incident and moved a number.
Programmes that skip documentation re-litigate the same approval every quarter and mistake it for organisational resistance.
What to do in the first quarter
Pick one brand or one market and take it end to end. Baseline, technical remediation, schema, answer-first restructuring of the top templates, re-measure.
A completed pilot with a recorded before-and-after is worth more internally than a comprehensive audit of the entire estate, because it converts the argument from theoretical to demonstrated. It also surfaces every governance obstacle at small scale, where they are cheap to solve.
The honest caveat
None of this is deterministic. Nobody controls model training data or retrieval ranking, and enterprise procurement should be sceptical of any supplier who implies otherwise.
What is controllable is whether your information is accurate, structured, corroborated and retrievable — across every brand, market and template you operate. At enterprise scale that is a substantial body of work with a measurable outcome, provided somebody recorded the baseline before it started.