Expert articles in English and Cantonese, on schedule

You need expert tips articles for two markets, in two languages, each with its own tone and featured merchant. Writing them by hand doesn't scale.

Regional online marketplace · SG and HKBuilt in-house by our founder
2languages written at the same time
8AI agents in the pipeline
70minimum score before a draft passes
In their shoes · A content team serving two markets

Before and after.

A merchant education platform needed regular expert tips articles for Singapore in English and Hong Kong in Cantonese, each with local context and a featured merchant. Manual writing couldn't scale, and generic AI text was not good enough.

  • Manual writing that couldn't keep up
  • No formal review before publishing
  • Generic AI text that missed local tone
How it works

What we built, step by step.

Step 1

Research merchant and local news

Each article starts with research on the featured merchant's profile and website, plus fresh local context for that market.

Step 2

Write in two languages

Singapore gets warm, conversational English. Hong Kong gets colloquial Cantonese, checked against Hong Kong language markers.

Step 3

Score against 4 criteria

A review agent scores every draft out of 100: brand voice, practical value, structure and how naturally the merchant story fits.

Step 4

Revise or ask a person

Drafts under 70 go back with specific notes, up to 3 times. If still short, the team is asked to review on Slack.

2languages written at the same time
8AI agents in the pipeline
70minimum score before a draft passes
The full story

From busywork to running on its own.

The problem

A merchant education platform needed regular expert tips articles for Singapore in English and Hong Kong in Cantonese, each with local context and a featured merchant. Manual writing couldn't scale, and generic AI text was not good enough.

What we built

We built two parallel pipelines that research the featured merchant, pull fresh local context, write the draft and score it out of 100 on brand voice, practical value, structure and merchant story. Drafts under 70 go back to the writer with notes, up to 3 times, before a person is asked to step in on Slack.

What changed

The content team gets polished Google Docs drafts in both languages on schedule. Hong Kong articles read as natural spoken Cantonese, and most drafts pass on the first or second try.

Built withn8nOpenAIFirecrawlTavilyGoogle Docs

Built in-house by our founder, before or alongside Workchestra. Names and numbers in the screens are samples; the results are real.

Keep reading

More case studies.

Case study 17 · Internal team

Every Slack request acknowledged and scheduled

You're the only one who can build automations, so requests land in Slack all day. Each one pulls you out of deep work.

Read the story
Case study 01 · Philippine manpower agencyWorkchestra client

Every applicant in one fair, traceable pipeline

Each recruiter keeps their own sheet and form, with different questions. Leads get assigned by hand, and nobody can see who interviewed whom.

Read the story

Want this for your team?

A free 20-minute call with the founder. We will tell you honestly if it is worth automating.

Book a free call
Case studies

Keep exploring the work.

18 case studies from real teams. Each one started with a team doing the same task by hand, every day.

See all 18
Case study 01Workchestra client

More than 1,200 candidates are routed and traceable from first contact to orientation, with nothing retyped by hand.

Philippine manpower agencyRecruiters and the agency's management1,200+candidates routed and trackedCase study 02Workchestra client

Management can compare any period, refreshed every hour, while the operations team keeps the same workbook.

Philippine manpower agencyManagement at a manpower agencyHourlyrefresh, fully unattendedCase study 03

Response time dropped from hours to under 60 seconds, with no manual data entry.

Regional online marketplaceAccount managers and their sales lead< 60 sfrom ad form to CRM recordCase study 04

Every thread, meeting and client session now has one source of truth.

US executive coaching firmA CEO and his executive assistant3ranked time slots per requestCase study 05

802 at-risk accounts, 35% of the 2,267 in scope, got a timely personal message in a single run.

Regional online marketplaceAccount managers with hundreds of accounts802at-risk accounts messaged in one runCase study 06

Research that took 20-30 minutes now finishes in under 2 minutes.

Regional online marketplaceAccount managers preparing for first calls< 2 minto research one companyCase study 07

Decisions take seconds because each card carries its own context, so anyone on the team can act.

US executive coaching firmThe CEO's team reviewing AI workHourlyAI runs reviewed in one placeCase study 08

3.04 TB and 36,118 files moved, preserving about seven years of client history in the firm's own storage.

US executive coaching firmA firm with years of session recordings3.04 TBmigrated end to endCase study 09

Every lead gets an instant reply, at 2 PM or 2 AM.

Regional online marketplaceSales reps and account managers5qualifying questions asked automaticallyCase study 10

Nobody pulls report data by hand anymore, and the team saves 8-10 hours a week.

Regional online marketplaceA marketing team running ads in two markets0manual data pullsCase study 11

The mastersheet updates itself, so nobody runs VLOOKUPs.

Regional online marketplaceThe ops person who owns the mastersheet3data sources joined automaticallyCase study 12

Coaches get overviews the same day and can turn one into a client email draft with a single button, now rolled out to every coach.

US executive coaching firmCoaches following up after client sessions8kinds of AI mistakes checkedCase study 13

A spoken note becomes a review-ready draft in under 30 seconds.

Internal teamAn executive who thinks faster than they type< 30 sfrom voice note to email draftCase study 14

Both markets start the day with an enriched, cleaned lead list, and existing clients never reach sales.

Regional online marketplaceSalespeople who should be selling300leads discovered per runCase study 15

Every submission is classified within seconds, before the sales team opens the file.

Regional online marketplaceSales reps calling new sign-ups< 2 sto verify each submissionCase study 16

A 2-day, 4-person process became one click.

Regional online marketplaceA sales manager approving monthly results1 clickto approve, down from 4 people and 2 daysCase study 17

Every request gets an answer within seconds and lands on the calendar, so nothing falls through the cracks.

Internal teamThe one builder everyone messages0requests left unacknowledged