Rajeev Gill

AI for agency operations: seven workflows that are already useful

Seven AI workflows I have run inside real agency operations — taxonomy governance, tag QA, attribution reconciliation, guidelines, rollouts, scoping, and turning calls into assets — described anonymised.

8/29/2026

Most "AI for agencies" content describes pilots. This article describes seven workflows I have run inside real operations, with the unglamorous details: what they replace, what they still need a human for, and where they fall short. Client details are anonymised throughout.

Four of seven workflows highlighted: taxonomy governance, tag and pixel QA, attribution reconciliation, and scoping with calls-to-assets
All seven run in live agency contexts. Four more are in pilot and not claimed here. Evidence date: 29 August 2026. Download the editable Excalidraw source.

The seven workflows

1. Global taxonomy governance. One event contract — names, properties, owners — maintained across markets. AI drafts and checks taxonomy changes against the contract; humans approve. What it replaced: spreadsheet archaeology and Slack archaeology.

2. Tag and pixel QA. Automated browser journeys exercise real consent flows and record what actually fires, compared against the contract. What it replaced: spot checks that missed consent-blocked tags for weeks.

3. Attribution reconciliation. Platform-reported conversions versus backend truth, with the gap measured and explained rather than argued about. What it replaced: trusting the platform number by default.

4. Guidelines maintenance. Living standards documents with named owners and review dates, checked automatically for drift. What it replaced: PDF guidelines nobody opened twice.

5. Rollout packs. One governance change produces the same deployable, auditable artifact set for every market. What it replaced: per-market improvisation.

6. Product and commercial scoping. Messy calls and decks become structured PRDs and scoped decisions with explicit open questions. What it replaced: the "we'll figure it out during the build" meeting.

7. Workshops and calls into operating assets. A call ends with an artifact, not just minutes. What it replaced: recordings nobody rewatched.

Governance stack: contract layer, QA layer, reconciliation layer, delivery layer, with review gates between every layer
Review gates sit between every layer. Nothing ships unreviewed. Evidence date: 29 August 2026. Download the editable Excalidraw source.

A rollout end to end

Five steps: contract updated, pack built, deployed, QA'd in browser, reconciled
This sequence runs per market, per change. The browser QA is the step that catches what config review misses. Evidence date: 29 August 2026. Download the editable Excalidraw source.

When platform and backend disagree

Before: platform claims conversions and backend shows fewer, with spend decided on the platform number. After: gap measured and explained, deduplication applied, decision on the reconciled number
Both sources stay visible; neither is silently overwritten. Known from anonymised reconciliation work across markets. Evidence date: 29 August 2026. Download the editable Excalidraw source.

What the human still does

Approve contracts. Decide which discrepancies matter. Sign off anything client-facing. The workflows remove retrieval, checking, and formatting labour; they do not remove judgment. That is the design, not a limitation.

Where to go next

Related reading

Building maintainable knowledge systems from messy business material

Files, email, decks, sheets, and meetings become a governed knowledge layer — with named owners, review dates, and provenance — so answers trace to sources instead of folklore.

Proof, not prompts

A delivery standard for AI-assisted work: evidence packs, provenance, editable deliverables, reconciliation, and review gates — so a claim can be checked, not just believed.

Why agent systems become slow, expensive and fragile

The measured mechanisms behind agent bloat: context resends, oversight cost, cumulative failures, and the attribution traps that make dashboards lie about all of it.