Your engine makes more drafts than it used to. Whether it makes better work with less of your time is a separate question, and the answer sits in numbers you probably aren't logging yet.
This final part of the build adds a dashboard with 2 sections, production and commercial, plus an error log. You'll read 4 weeks of illustrative data from Ridgeline Ops (the fictional consultancy that runs through this series), find its weakest stage, and set up a fair comparison for the next 4 weeks.
Full commercial results, revenue and return included, belong in a separate results tracker. Here the commercial section has 1 job: showing whether the pieces reach buyers and give them a reason to act.
Jump to the free AI Content Engine Review Dashboard
AI content performance: measure editing time, defects and cost per approved piece
AI content performance starts with 3 production measures: minutes of human editing per approved piece, defects found per draft, and cost per approved piece with your own time included. Log them against each content ID on the board, total them weekly, and you'll see which stage is costing you before a month of guessing goes by.
What to log for each piece
| Field | How to record it |
|---|---|
| Edit minutes | From opening the draft to approving it; a phone timer is enough |
| Brief and prompt minutes | Time spent writing the brief and running the prompts |
| Draft versions | The final draft version number on the board |
| Defects | Each problem you fixed, logged with its type and stage; sharpening a line that was already correct doesn't count |
| Approved | Yes, or No with the reason it went back |
Defect types
Give every defect 1 type, so patterns show up after a few weeks:
- Unsupported claim: a fact with no source ID behind it
- Generic line: a sentence any competitor could post
- Voice miss: a break from the voice brief
- Repetition: another piece's main point, aimed at the same reader
- Meaning drift: a claim that changed meaning on its way from the source
- Wrong next step: missing, vague or different from the brief
- Format miss: wrong length or shape, like a script that runs long
Log the stage where each defect started, too: source, brief, draft, voice pack or repurposing. The stage tells you where the fix belongs.
The production section, 4 weeks of illustrative data
All figures below are illustrative. Maya Okafor's time is valued at $150 an hour, and tool costs are $8 a week for the assistant in weeks 1 and 2, then $30 a week once the Make scenarios from part 4, automating the workflow in Make, start in week 3. Both numbers are invented for the example.
| ID | Week | Format | Versions | Edit minutes | Defects | Approved |
|---|---|---|---|---|---|---|
| RL-1 | 1 | Post | 2 | 35 | 4 | Yes |
| RL-2 | 1 | Post | 2 | 50 | 3 | Yes |
| RL-3 | 2 | Post | 2 | 25 | 1 | Yes |
| RL-4 | 2 | Post | 3 | 70 | 4 | No, back to Idea |
| RL-5 | 3 | Post | 2 | 20 | 2 | Yes |
| RL-6 | 3 | Article outline | 1 | 15 | 0 | Yes |
| RL-7 | 3 | 3 | 65 | 4 | Yes | |
| RL-8 | 4 | Video script | 2 | 30 | 2 | Yes |
| RL-9 | 4 | Carousel outline | 2 | 25 | 1 | Yes |
| RL-10 | 4 | Post | 2 | 20 | 1 | Yes |
The weekly roll-up uses this formula: cost per approved piece = (Maya's minutes ÷ 60 × $150 + tool cost) ÷ approved pieces. Maya's minutes add her brief and prompt minutes (70, 70, 65 and 40 across the 4 weeks) to her edit minutes.
| Week | Drafts | Approved | Edit minutes per approved | Defects per draft | Maya's minutes | Cost per approved piece |
|---|---|---|---|---|---|---|
| 1 | 2 | 2 | 42.5 | 3.5 | 155 | $198 |
| 2 | 2 | 1 | 95 | 2.5 | 165 | $421 |
| 3 | 3 | 3 | 33 | 2.0 | 165 | $148 |
| 4 | 3 | 3 | 25 | 1.3 | 115 | $106 |
Week 2's spike is RL-4: 70 minutes of editing on a piece that never got approved. Rejected work still costs money, and cost per approved piece shows it, where cost per draft would hide it.
The error log
Here are 10 of the 22 logged defects, by ID; RL-4's 3 missing figures share a row.
| ID | Defect | Type | Stage | Fix |
|---|---|---|---|---|
| RL-1 | Invented "up to 40%" statistic | Unsupported claim | Draft | Drafting prompt gains part 2's [NEED FACT] rule |
| RL-2 | Opened with a question | Voice miss | Voice pack | Opening rule added to the voice brief |
| RL-3 | Called a 2-year-old change "recent" | Unsupported claim | Source | Facts register gains a When column |
| RL-4 | 3 claims with no figure behind them | Unsupported claim | Source | Brief back to Idea; note N-15 needs data |
| RL-5 | Same opening story as RL-7 | Repetition | Repurposing | Matrix gains a Used in column |
| RL-7 | "Finance ruined the meeting" | Meaning drift | Draft | Rewritten to match the interview at 11:40 |
| RL-7 | Asked for a call in the first email on a topic | Wrong next step | Brief | Changed to a reply question |
| RL-9 | Said 9 orders caught; the notes say 7 | Unsupported claim | Source | Drafts now run from the corrected source sheet |
Across all 22, the stages break down as source 8, draft 7, voice pack 5, repurposing 1 and brief 1.
Connect published work with relevant inquiries and calls
Connecting published work to inquiries means tagging each response with the content ID that prompted it and separating buyers from everyone else. Count buyer replies, resource requests and booked calls per piece. Those counts show audience fit and next-step strength, which are production questions; revenue and return belong in the results tracker.
Attribute each response to a content ID
Maya's reply keywords do most of the attribution: "handoff" belongs to RL-1, "form" to RL-2, "cutoff" to RL-3 and "huddle" to RL-5. A reply without a keyword gets the ID she judges most likely, marked as a guess.
A buyer reply comes from someone who matches her buyer: an operations director or owner at a 20-to-80-person distribution company. Peers, vendors and job seekers count as other replies. The split takes a minute per reply, and it's the most useful number in this section.
The commercial section, illustrative
| ID | Week live | Buyer replies | Other replies | Resource requests | Calls booked |
|---|---|---|---|---|---|
| RL-1 | 1 | 2 | 6 | 3 | 0 |
| RL-2 | 2 | 2 | 8 | 2 | 0 |
| RL-3 | 3 | 3 | 5 | 0 | 0 |
| RL-7, email | 3 | 4 | 0 | Reply ask, no resource | 1 |
| RL-5 | 4 | 1 | 14 | 1 | 0 |
2 rows stand out. RL-5 drew the most replies of the month, and only 1 came from a buyer; most came from warehouse supervisors and fellow consultants. RL-3 drew 3 buyer replies and 0 requests for its change-cutoff memo.
Carry these content IDs into the personal brand results tracker when you want to follow a reply through to a client and its revenue. This dashboard stops at the booked call.
Choose 1 improvement and compare the next cycle
Choosing 1 improvement means naming the weakest stage from the dashboard, writing down the change, and holding everything else steady for the next 4 weeks. The Ridgeline data supports 4 diagnoses, each with a specific test. Maya can run only 1 cleanly, so she picks the stage that costs the most and sets a comparison she can't argue with later.
4 diagnoses and a test for each
| Problem | Signal in the data | Test |
|---|---|---|
| Weak source material | 8 of 22 defects started at the source; RL-4 (70 minutes, never approved) and RL-7 (65 minutes) both had evidence gaps | A brief reaches Ready only when each claim it allows points to a note with a figure, a date or a named example |
| Heavy editing | RL-2, RL-4 and RL-7 took 50 to 70 minutes; the other 7 pieces had a median of 25 | Add 1 of Maya's client emails to the voice pack, which holds only posts, as the model for email briefs; then compare edit minutes on the next 2 emails |
| Poor audience fit | RL-5: 15 replies, 1 from a buyer | Rewrite the next huddle post's buyer question in an owner's words, then compare the buyer share of replies |
| Weak next step | RL-3: 3 buyer replies, 0 memo requests | Offer the cutoff rule as a 1-page sheet for the dispatch board, then compare requests per buyer reply |
The 1 change Maya picked
Source material is the weakest stage. It produced the most defects and the most expensive piece, RL-4, which was also the only rejected one. Fixing it also happens before any drafting time is spent.
The change is a stricter Ready rule: a brief moves to Ready only when every claim it allows points to a note with a figure, a date or a named example. The prompts, voice pack, Make scenarios and publishing rhythm all stay as they were in week 4.
Compare against weeks 3 and 4 only. The automation changed the setup after week 2, so a 4-week baseline would credit the new rule with savings the automation made.
| Measure | Baseline, weeks 3 and 4 | Target, weeks 5 to 8 |
|---|---|---|
| Source-stage defects per draft | 0.67 (4 in 6 drafts) | Under 0.2 |
| Edit minutes per approved piece | 29 | Under 25 |
| Cost per approved piece | $127 | Under $115 |
| Briefs held at Idea for missing evidence | Not tracked | Count them; no target |
Her decision rule, written before week 5 starts: keep the new Ready rule if source-stage defects and edit minutes both fall and cost per approved piece doesn't rise. If more than 2 briefs a week stall at Idea, she adds a monthly hour for pulling figures into her notes.
The AI Content Engine Review Dashboard has these tables with each formula printed beside them, plus the error log and diagnosis sheet.
Add the 5 production fields to your board this week and log every piece for 4 weeks before you change anything. Then pick the stage with the most expensive defects, change 1 thing, and compare the next 4 weeks against a baseline that ran on the same setup. Deciding which buyer questions earn a slot on the board is still a strategy call, and the AuthorityOS guide to personal brand content strategy covers it.