Content-ops AI on an existing comparison stack
Whistleout · Agentic workflows · comparison publishing
Plan changes, plan tables, and editorial updates still depended on specialists moving between the CMS, spreadsheets, and partner feeds. Volume increased. The operating model did not.
We added an agentic layer to the content operations they already ran: ingest partner updates, flag mismatches, draft table and copy changes, and leave publishing with editors.
Editors keep the CMS. AI handles the repetitive updates. Fewer stale plans reach the page.
27% less time from partner update to published table.
Context
Comparison publishers depend on freshness. Plans, prices, and inclusions change constantly, and a mid-size editorial team cannot scale in line with every partner feed. The CMS, the spreadsheet used as the source of truth, and the live tables were already in place. The work between them was manual.
Replacing the publishing platform would have stopped updates to the site. The brief was to place AI inside the workflow that already published 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 change an editor could trust. A miss left a stale table. Too much automation risked publishing the wrong inclusion.
We prioritized the plan families with the highest change volume and the longest handling time, so the first insertion would show up in cycle time rather than in a trial environment.
What we built
An agent monitors the feeds and files the team already used, proposes a structured change, and drafts the CMS fields and short copy that usually accompany a table update. Editors review the change in the same publish path as before.
Controls included no automatic publishing, a confidence label when a source is incomplete, and a clear “a person 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 editorial judgement, including which comparison to feature, rather than on transcription.
The CMS remained the system of record. AI took on defined steps in a process the team already owned.
Lessons for mid-size publishers
Do not replace the CMS in order to introduce AI. Add detection, drafting, and quality checks to the publish path you already trust. Measure time to publish and error rate, and keep a person on any change that could 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.