Methodology

How the calculator works

Short, transparent notes on every workflow — what we measure, the math we use, and where the percentages come from.

← Back to the calculator

Overview

What this estimates

The tool adds up recoverable hours per week across seven workflows, then converts them to money using your hourly rate. Percentages are net of human review where journalism standards require it — not raw vendor speed claims.

Where evidence is noisy we round down on savings. Use the output for planning, not as a promise of realized savings.

Quick reference

Seven workflows

WorkflowScope / inputNet time saved
TranscriptionInterview hours/day × journalists70%
DraftingStories/day × journalists45%
TranslationArticles translated/day (newsroom)65%
Social post writingPosts/day (all channels)45%
Social creativesVisual assets/day55%
ResearchResearch hours/day × journalists60%
Data & documentsDoc hours/day × journalists55%

Details

Formula by workflow

01 — Transcription

70%

AI returns a draft transcript quickly; journalists still verify quotes and names. Savings are computed on the manual baseline, then multiplied by the net reduction after that review work.

audioHoursPerWeek = interviewHoursPerDay × 5 × journalistsCount
manualHours = audioHoursPerWeek × 4        // 4:1 manual typing vs audio
hoursSavedPerWeek = manualHours × 0.70
  • 4:1 — about four hours of typing per hour of audio for manual transcription.
  • 70% net — room for quote checks and corrections (hybrid newsroom practice).
  • Interview hours — load follows recording volume, not story count.

Sources: Sonix (transcription benchmarks) · CNTI newsroom research (2025) · Nieman Lab

02 — Drafting

45%

Covers outlines, first drafts, headlines, and SEO rewrites. The journalist keeps responsibility for facts, sourcing, and final voice.

draftsPerWeek = storiesPerDay × 5 × journalistsCount
totalHours = draftsPerWeek × (60 / 60)        // 60 min per draft
hoursSavedPerWeek = totalHours × 0.45
  • 60 min — typical first-draft window for ~500–800 words before editing.
  • 45% — faster first drafts plus fact-check and voice overhead (conservative vs HBS-style AI studies).
  • Scope — excludes sub-editing and legal review.

Sources: Harvard Business School (2025) · newsroom AI practice surveys

03 — Translation

65%

Machine translation does the first pass; a human checks tone, accuracy, and cultural nuance before publish.

translationsPerWeek = translationVolume × 5
totalHours = translationsPerWeek × (30 / 60)   // 30 min/article manual
hoursSavedPerWeek = totalHours × 0.65
  • Newsroom-wide — do not multiply translation volume by journalist count.
  • 30 min/article — manual baseline for a typical news piece.
  • 65% net — hybrid AI plus human editorial pass for accuracy.

Sources: CNTI (2025) · Reuters Digital News Report

04 — Social post writing

45%

AI drafts captions, threads, and platform variants from a brief; an editor approves before publishing.

postsPerWeek = socialPostsPerDay × 5
totalHours = postsPerWeek × (12 / 60)
hoursSavedPerWeek = totalHours × 0.45
  • 12 min/post — ideation, writing, and platform formatting for text-first posts.
  • Split — visual and video pieces count under Social creatives.

Sources: Buffer / Sprout industry benchmarks (2025)

05 — Social creatives

55%

Templates, auto-resize, brand kits, and AI-assisted cuts reduce hands-on design time for social-native formats.

assetsPerWeek = socialAssetsPerDay × 5
totalHours = assetsPerWeek × (20 / 60)
hoursSavedPerWeek = totalHours × 0.55
  • 20 min/asset — branded graphic or short social clip — not long-form video.
  • 55% — reflects stronger AI design and short-video tooling in 2025–26.

Sources: Canva / Adobe Express / CapCut product trajectory

06 — Research

60%

Natural-language research tools compress search, scan, and synthesis time; journalists still verify claims against primary sources.

hoursPerWeek = researchHoursPerDay × 5 × journalistsCount
hoursSavedPerWeek = hoursPerWeek × 0.60
  • Desk research — applies to search and synthesis — not field reporting or cultivating sources.

Sources: McKinsey State of AI (2025) · HBS AI productivity work

07 — Data & documents

55%

Document AI speeds up extraction, summarisation, and structured questions across large file sets; cross-checking remains manual where stakes are high.

hoursPerWeek = dataDocumentsHoursPerDay × 5 × journalistsCount
hoursSavedPerWeek = hoursPerWeek × 0.55
  • Docs & data — PDFs, spreadsheets, CSVs, data-heavy pages — not final verification of sensitive numbers.

Sources: NotebookLM / Claude document benchmarks (2025–26)

Money

From hours to euros

We sum weekly savings across all workflows, then:

monthlyCostSavings = totalHoursSavedPerWeek × hourlyRate × (52 / 12)
// 52÷12 ≈ 4.33 weeks/month — using 4 would undercount by ~8%

annualROI = monthlyCostSavings × 12

Hourly rate should be a fully loaded staff cost if you can (salary, tax, benefits). The result is time value, not net profit — subtract tool subscriptions if you need net cash impact.

Built-in averages

Constants in the model

  • Manual transcription ratio4:1 (typing vs audio)
  • Draft baseline60 min / story
  • Translation baseline30 min / article
  • Social post baseline12 min / post
  • Social asset baseline20 min / asset
  • Working week5 days
  • Weeks per month52 ÷ 12 (≈ 4.33)

Savings assume active, trained use of appropriate tools — licenses alone do not recover time.

Out of scope

Not included (yet)

These can mean real extra savings; the calculator intentionally ignores them today:

  • AI-assisted editing, house style, and sub-editing
  • Distribution automation and push/notification copy
  • SEO metadata and structured data at scale
  • Analytics digests for editors
  • Alt text and high-volume image captioning
  • Subscription cost of tools (gross savings only)

Last updated April 2026