Budget the Program - Tooling, Headcount, Build-versus-Buy

AI SEO budgets fail in a specific direction: too much software, too little capacity to act on what the software says. The category is young, the vendors are loud, and a dashboard is easier to buy than an editorial process is to build. The corrective is a spend order — and the discipline to hold it.

The spend order

Tier zero: free and first-party. Before any procurement: Bing Webmaster Tools with the AI Performance report (platform-native citation data, free), IndexNow wired into the publish path, AI-referral segmentation in your analytics, and your own server logs — the ground truth for what AI crawlers actually fetch. This tier costs configuration time and produces the calibration data every later purchase gets judged against.

Tier one: the tools you already pay for. Enterprise crawlers, log analyzers, and SEO platforms have been growing AI-readiness and AI-visibility modules. Exhaust what current contracts cover before adding a vendor — the delta you need may be a feature flag, not a purchase order.

Tier two: one dedicated visibility platform, piloted. The AI visibility category is real but crowded and fast-moving, and enterprise coverage — multiple engines, multiple markets, corpus control — is exactly what entry tiers gate. Pilot against a fixed prompt corpus for a quarter and score four things: engine and market coverage, agreement with your Bing first-party data, corpus and export control, and whether findings convert into work your teams accept. Buy after the pilot, not the demo.

Tier three: build what cannot be bought. The workflows with the most enterprise leverage are too specific to your stack for vendors: schema validation gates in CI, nightly diffs of schema-declared prices against the commerce API, log-based crawler audits across market domains, answer-accuracy checks against your own product truth. Small, governed scripts — not a platform project.

Why log evidence outranks advocacy

The llms.txt episode is the budget lesson of the past two years. The file accumulated enthusiastic advocacy through 2025; when Ahrefs measured 137,210 domains in mid-2026, 97% of published llms.txt files had received zero requests — and no AI bot ever probed for the file on sites lacking one. Google states plainly that its search systems ignore it. Any enterprise with its own server logs could have verified this for the cost of one query.

That is the standing procurement rule: before funding any AI SEO artifact, check whether the systems it targets actually consume it — in your own logs. Vendor conviction is not uptake. See IndexNow: Reality vs Hype for the same test applied to a mechanism that does have real consumers.

Headcount versus tooling

The binding constraint in a mature program is almost never measurement — it is the capacity to act on measurement. Citation gaps convert into visibility only through editors restructuring pages, engineers shipping schema, and PR earning coverage. A useful planning posture for an enterprise delta-program: the majority of new spend goes to people and process — editorial throughput, technical implementation time, PR briefs — and tooling stays a minority line. If the plan’s tooling line exceeds its capacity line, the plan produces reports, not visibility.

Decision rules

• No paid visibility tool until tier zero is live; you cannot evaluate a sampler without first-party calibration data.
• Procurement criteria in order: market/language coverage, engine and mode coverage, corpus control, export/API access, agreement with first-party data. Dashboard aesthetics are not on the list.
• Every build-tier script gets an owner and a repository. Ungoverned personal scripts become next year’s silent failure.
• Re-run the budget split annually. The tool category is consolidating fast; a platform bought this year should have to re-win its line next year.