Time savings & ROI

Quantifies how much time the two pillars — auto-input and automation testing — save when building one website template, split by two scenarios.

All numbers are estimates

Every figure below is an estimate (est.), kept deliberately conservative. Replace them with measured data using the procedure in §6. Do not quote these as measured. Exceptions, which are grounded: the WordPress per-template table (§2.1) is design-team reviewed (real numbers), and the recurring task hours in §5 are real (from the workload sheet). The Duda/Astro build tables in §2–§3 are still estimates.

0. The two scenarios

• Design ready — a design/template already exists; you only convert, populate, and QA.
• From scratch — no design yet; the design phase is part of the work.

Automation doesn’t design for you, so the percentage saved is highest when the design already exists — there’s less fixed cost left for automation to compress.

1. The phase model

A template build is modelled as 5 phases. Automation touches phases 2–5:

#	Phase	Touched by
1	Design the template	— (from-scratch only)
2	Convert / build structure on the platform	partly auto-input
3	Auto-input content (text / images / SEO meta)	auto-input
4	QA / testing	automation-testing
5	Fix + re-verify	both

Per-platform caveat:

• WordPress & Duda — phase 3 (auto-input) is a distinct lever (MCP / Dev Mode).
• Astro — phases 2 and 3 merge: content goes in as code in Claude Code, and testing runs directly on that code. The Astro table below combines them.

2. Per-platform tables

Times in hours. M = manual, A = with automation. WordPress (§2.1) is design-team reviewed (real numbers); Duda and Astro are still raw estimates.

WordPress revised after design-team feedback

The original SOP estimate was too optimistic. The design team reported the real-world automated cost is higher: total 3.75h → 8.25h (design ready) and 11.75h → 16.25h (from scratch) — i.e. saved 75% → 45% and 49% → 29%. Two systematic corrections, which almost certainly apply to Duda and Astro too (still un-reviewed below):

• MCP is not 1-click — building structure still needs setup/tweak (2h → 4h).
• Auto-input output isn’t perfect — it needs review (3 → +0.5h) and ~2h of UAT fixing on top of the base fix pass (1h → 3h).

2.1 WordPress (auto-input via MCP) — design-team reviewed

Phase	Ready M	Ready A	Scratch M	Scratch A	Note
1 Design	0	0	8	8	automation doesn’t design
2 Build structure	6	4	6	4	MCP still needs setup/tweak — not 1-click
3 Auto-input content	4	1	4	1	still need to review the output after the MCP run
4 QA / testing	3	0.25	3	0.25	CI suite is fast — unchanged
5 Fix + re-verify	2	3	2	3	auto-input output isn’t perfect → ~2h extra UAT fix
Total	15	8.25	23	16.25
Saved	—	45%	—	29%

Duda (§2.2) and Astro (§2.3) below are still the original optimistic estimates — they have not had the design-team realism pass. Expect the same two corrections (build is not 1-click; output needs review + UAT) to pull their automated totals up once measured.

2.2 Duda (auto-input via Dev Mode, no API)

Phase	Ready M	Ready A	Scratch M	Scratch A	Note
1 Design	0	0	8	8
2 Build structure	5	2.5	5	2.5	Dev Mode layout build
3 Auto-input content	4	1	4	1	Dev Mode population
4 QA / testing	3	0.5	3	0.5	output-only audit on preview + live URL
5 Fix + re-verify	2	1	2	1
Total	14	5	22	13
Saved	—	64%	—	41%

2.3 Astro (everything in code; phases 2+3 merged)

Phase	Ready M	Ready A	Scratch M	Scratch A	Note
1 Design	0	0	8	8
2+3 Code + content	12	3	12	3	hand-code vs figma-to-astro / clone-to-astro in Claude Code
4 QA / testing	3	0.25	3	0.25	runs on source + CI gate
5 Fix + re-verify	2	1	2	1
Total	17	4.25	25	12.25
Saved	—	75%	—	51%

3. Summary — total hours / template

Platform	Design ready (M → A)	Saved	From scratch (M → A)	Saved
WordPress (reviewed)	15 → 8.25	45%	23 → 16.25	29%
Duda (est.)	14 → 5	64%	22 → 13	41%
Astro (est.)	17 → 4.25	75%	25 → 12.25	51%

The WordPress row is the realistic, design-team-reviewed figure; Duda and Astro are still optimistic estimates and will likely land lower once reviewed.

Contribution split (of the automated savings, roughly):

• Auto-input drives most of the phase 2–3 savings (the largest block).
• Automation-testing drives the phase 4–5 savings — smaller in raw hours but the highest leverage, because QA is the bottleneck that blocks shipping and is the most error-prone manually.
• Automation is not free. The design-team review showed auto-generated output needs human review + UAT fixing, so the Fix phase can be larger than the manual one (WordPress: 3h automated vs 2h manual). Budget for it — the win is in build + input, not in skipping QA/fix.

The percentage saved is highest when the design is ready: with the fixed design cost removed, automation compresses nearly everything that’s left.

4. One-time costs (not per-template)

These do not recur per template, so they’re excluded from the tables above:

• Visual baseline backfill across the fleet — ~1–2 weeks, one person.
• CI setup (reusable workflow, branch protection, Pages preview) — done once, ~0 per new repo via the starter template.
• Learning Playwright / the kit — ~1 week ramp for the QA owner.

5. Recurring QA/ops task hours (from the workload sheet)

These are real per-task hours the teams entered in the Workload & Package Builder sheet (tab Task Time Inputs) — the recurring SEO/QA/ops work that the automation suite targets. (Dollar costs are intentionally omitted: the sheet’s blended rate and package quantities are still blank, so Summary & Cost is $0 — fill those to cost it.)

Recurring task	Hours (manual)	Frequency	Automation coverage
Uptime / SSL check	0.15	per site / month	Fully automated — nightly uptime monitor (HTTP/TLS/DNS), free
Doorway test QA	0.60	per page	Automated — static auditor + design-auditor (overflow, headings, links)
Launch-readiness QA (robots/canonical/sitemap)	0.50 + 0.30/page	per site	Mostly automated — static auditor + design checks
Technical crawl + triage	2.15	per project / month	Partly — auditor flags; triage/fix stays human
Page speed / Core Web Vitals	4.50	per project	Partly — Lighthouse in CI measures; fixes human
QueryMind content audit + review	0.66	per page	Agentic — qa-verify judgment
Monthly performance report	1.30	per web / month	Data pull automatable; narrative human
WordPress maintenance (updates+backup+scan+SSL)	1.15	per month	SSL/uptime automated; updates/backup human

Highest-leverage automation wins, in order: uptime/SSL (0.15/site/mo → fully free & continuous, was manual & easily forgotten), then per-page doorway/launch QA (0.60–0.80 × every page → the design-auditor runs them in seconds on the free runner).

Source: workload sheet 1APbbV8oZtNd2CZ4uKqnWapkyfxs_j8ReKw_T65stMCE · tab Task Time Inputs. Read via the gws CLI. Numbers update when the teams revise their inputs.

6. Measure the real time

Replace every estimate above with measured data:

1. Pick 1–2 pilot templates (1 Astro with source, 1 Duda/WP output-only).
2. Log start/end per phase — manual vs automated — in a sheet or a dashboard column.
3. For the automated QA phase, read the CI job duration instead of timing by hand (GitHub Actions run time / Worker run). The verdicts and per-run history are on the QA console.
4. After ≥ 3 templates, replace the estimate cells with the measured averages.