Content repurposing and automated reporting

0:00 Repurposing + reporting

If you create content or report on marketing results, you probably spend hours every week on work that follows the same steps every time: turning one webinar into a dozen posts, or pulling numbers from five dashboards into one report. These are two of the highest-return automations a marketer can build, and they're low-risk when designed with review steps. In this lesson you'll build a content repurposing pipeline and an automated reporting pipeline, and learn the one principle that keeps AI reports trustworthy.

0:36 Lesson roadmap

Here's the plan. First, why these two automations deliver such reliable returns. Then an analogy for repurposing, the repurposing pipeline and its design rules. Then the reporting pipeline and the key principle that keeps AI reports trustworthy. Then two examples: a consultant's weekly blog flow and a restaurant group in Riyadh doing both content and reporting.

1:00 Why it matters

Why does this matter? Because consistency beats bursts. Teams that publish regularly across channels and review their numbers weekly make better decisions, but both jobs are tedious, so they slip. Automating the repetitive parts keeps the rhythm going, and frees your time for the creative and strategic work that actually needs you.

1:23 Analogy: one stock, many dishes

An analogy for repurposing. Think of a chef with one great stock. From it they make soup, sauce, risotto and gravy, each adapted to the dish, but all built on the same base. Your long-form content is the stock. The extraction step skims out the key points, quotes and facts. Then each channel prompt cooks a dish for its audience: a LinkedIn post, a thread, an Instagram caption, a newsletter blurb, a short video script. And a head chef, your reviewer, tastes before anything goes out.

2:00 Repurposing pipeline

Here's the repurposing pipeline. A trigger fires when a source is published: an RSS item, a YouTube upload, a podcast episode or a file in Drive. Get the content, transcribing audio or video if needed. Then an AI extraction step pulls key points, verbatim quotes with timestamps, statistics with their sources, the audience and the call to action. From that, channel drafts are written in parallel, each with its own prompt, examples of your best posts, and length rules. Brand and compliance checks run, a human approves with edits allowed, and approved drafts are scheduled and logged.

2:42 Design rules

A few design rules make the difference. Extract first, then write: grounding drafts in extracted facts reduces made-up content. Never invent statistics: tell the model to use only figures that appear in the source, and flag anything missing. Create Arabic or Urdu versions as separate drafts with native review, not word-for-word translations. And handle disclosures: sponsored content needs hashtag ad or the platform's paid partnership label, and realistic AI imagery may need labeling.

3:14 Reporting pipeline

Now reporting. A schedule fires, say Monday at eight forty-five local time. Pull metrics from analytics, ad platforms, the CRM and your store. Normalize everything into one table: date, channel, campaign, spend, clicks, leads, revenue. Compute KPIs in code or a spreadsheet: cost per acquisition, return on ad spend, conversion rate and week-over-week changes. Then let AI write the narrative from that computed table, and deliver it where people work, like Slack, email or a dashboard.

3:47 Key principle

Here's the principle that keeps reports trustworthy: compute numbers in code, use AI only to explain them. Language models are unreliable at arithmetic across many rows, but good at narrative once the numbers are fixed. So the prompt says: use only the numbers in the table, don't calculate new totals, say if something's missing. Structure the summary as three headline changes with exact figures, two likely drivers labeled as hypotheses, two questions to investigate, and one recommended action, in under a hundred and fifty words.

4:24 Example 1: blog -> LinkedIn + newsletter

Example one, simple. A consultant publishes a weekly blog post. An RSS trigger fetches it, AI extracts three key points and one quote, and drafts a LinkedIn post and a newsletter blurb. The consultant approves both in Slack each Monday, and they're scheduled automatically. Ten minutes of review replaces an hour of rewriting.

4:47 Example 2: Riyadh restaurant group

Example two, realistic. A restaurant group in Riyadh. Each monthly chef video on YouTube triggers a workflow: the transcript is used to extract dishes, quotes and tips, which become Arabic Instagram captions and English LinkedIn posts. A brand check blocks health claims and only mentions halal certification when verified. The marketing manager approves, and posts are scheduled. For reporting, every Sunday at eight forty-five Riyadh time, the workflow pulls ad spend, reservations and delivery orders, computes cost per reservation and weekly changes per branch in code, and AI writes the narrative. It goes to opted-in managers on WhatsApp through an approved template, and by email with the sheet link.

5:34 Common mistakes

Three mistakes to avoid. Letting AI compute totals or percentages from raw rows, which produces confident, wrong numbers. Publishing repurposed content without review, especially in a second language, where tone and meaning can drift. And reports nobody reads. Deliver them where people already work, keep them short, and always include one recommended action.

5:57 Watch me do it: weekly lead report

Watch me do it. I build the weekly lead report for Crescent in n8n. A schedule trigger every Monday at eight forty-five, Karachi time. Three data pulls: the CRM for last week's leads with source, band and first response time; the ads platforms for spend by campaign; and the CRM again for deals won. A Merge node combines them by source. Then a Code node computes the numbers: leads per source, cost per lead, share of A-band leads, median first response time, and week-over-week change, and outputs a CSV table. No AI touches the arithmetic. Next, a Basic LLM Chain with the analyst prompt: use only the numbers in the table, don't calculate new totals, give three headline changes with exact figures, two hypotheses clearly labeled, two questions, one action, under a hundred and fifty words. I run it. The summary mentions a percentage that isn't in the table, so I tighten the prompt with: if a number is not in the table, do not mention it, and add the week-over-week column so the model doesn't need to calculate. Rerun: clean. Finally, a Slack node posts the summary to the leadership channel with a link to the full table in Google Sheets, and the table is appended to an archive sheet for trends.

7:30 Recap + try this now

Recap. Repurpose by extracting first, drafting per channel, checking and approving. Report by pulling, normalizing, computing KPIs in code, and letting AI explain the fixed numbers. Try this now: use the Python KPI function in the lesson text on last week's numbers from two channels, feed the CSV output to the analyst prompt, and compare the AI's summary with your own reading. If it invents anything, tighten the prompt.

8:00 Try this now

Try this now. Take last week's numbers from two channels, for example ads and your CRM, and put them in one table with channel, spend, leads, revenue and previous week's leads. Run the Python KPI function from the lesson text to produce a CSV. Paste that CSV into the analyst prompt. Compare the AI summary with your own reading of the numbers. If it invents a figure or calculates something new, tighten the instructions and run it again.