Content-ops AI on an existing comparison stack
Whistleout · Agentic workflows · comparison publishing
Plan changes, plan tables, and editorial updates still depended on specialists jumping between CMS, spreadsheets, and partner feeds. Volume went up. The operating model did not.
We dropped an agentic layer onto the content ops they already ran: ingest partner updates, flag mismatches, draft table and copy diffs, and leave publish with editors.
Editors keep the CMS. AI does the grind. Fewer stale plans reach the page.
27% less time from partner update to published table.
Context
Comparison publishers live and die on freshness. Plans, prices, and inclusions change constantly, and a mid-size editorial team cannot scale linearly with every partner feed. The CMS, the spreadsheet of “source of truth,” and the live tables were already there — they were just too manual.
Standing up a new publishing platform would have frozen the site. The brief was to put AI inside the workflow that already shipped pages.
Problem framing
The expensive work was not writing from scratch. It was detecting that a partner had changed a field, reconciling it with what was live, and producing a diff an editor could trust. Misses meant stale tables; over-automation meant publishing the wrong inclusion.
We prioritised the plan families with the highest change volume and the longest human handling time so the first insertion would show up in cycle time, not in a lab.
What we built
An agent watches the feeds and files the team already used, proposes a structured diff, and drafts the CMS fields and short copy that usually accompany a table update. Editors review in the same publish path as before.
Guardrails: no auto-publish, confidence labels when a source is incomplete, and a clear “human must confirm” state for anything that changes price or eligibility.
Outcomes
Elapsed time from partner update to published table fell by about 27% on the targeted plan families, with fewer stale-field incidents in the same window. Editors reported spending the saved time on judgement — which comparison to feature — not transcription.
The CMS remained the system of record. AI became another pair of hands in a process they already owned.
Lessons for mid-size publishers
Do not replace the CMS to “do AI.” Insert detection, drafting, and QA into the publish path you already trust. Measure time-to-live and error rate, and keep a human on anything that can mislead a customer.
“Our editors still own the site. AI just stopped us copying the same plan change into three places by hand.”
Liked this insight?
See how these concepts apply to your business. Book a short fit call and we'll map a practical next step.