Zovalor
Recent project · Agentic workflow

An agentic workflow that runs marketplace reporting every morning, so the team reviews instead of compiling.

  • ~90%less time on reporting
  • 4 → 14 marketplaces, 1 automated pipeline
  • 5coordinated AI agents
  • Human review at every key step

The challenge

The team logged into four retail marketplaces, downloaded spreadsheets by hand and ran calculations across them to understand sales, reviews and product performance. It took hours every cycle, and the numbers fed real decisions on pricing and advertising.

4 marketplaces
Manual spreadsheets
hours

How we approached it

  1. 1

    Leadership interviews

    Understood the overarching business objectives and what decisions the data needed to support.

  2. 2

    Team interviews

    Mapped how each functional team really did the work, including the spreadsheets and calculations they relied on.

  3. 3

    SOP review and refinement

    Compared what teams actually did with their written standard operating procedures, corrected the SOPs to match reality, and marked where automation and AI could save time, cut cost and improve accuracy.

  4. 4

    AI policy alignment

    Checked every part of the workflow against the company's AI policy, covering which data the agents could access, where it could go, and where a person must approve, so the automation stayed compliant from day one.

  5. 5

    The agent-ready SOP

    Rewrote the process as an agentic SOP, covering rules, guardrails, data sources, API and MCP connections, files, skills and instructions, so every agent knew exactly what to do and what not to do.

  6. 6

    Team training and handover

    Trained the team to work with the agentic workflow: how to review and approve the agents' output, ask the Q&A agent questions, and spot when something needs a human. That way the team owns the system, not just the reports.

The system

An orchestrated multi-agent workflow

Step 1Data Engineer agent

Extracts data from all 4 marketplaces, then cleans and pre-processes it.

Step 2Data Analyst / BI agent

Analyses performance and builds the Power BI visualisations and the report.

Step 3Communications agent

Emails results to the right team members, collects their review and feedback, and reports back.

Step 4Q&A agent

Answers anyone's questions about the finished report.

Scheduled · 9:30 a.m. daily

Orchestrator agent

The main agent. Starts the workflow on schedule and hands work to each sub-agent in order.

Data flow

  1. 4 retail marketplaces
  2. Data Engineer
  3. BI analysis
  4. Human reviewvia Communications agent
  5. Final report
  6. Pricing and ad decisions

A day in the workflow

  1. 9:30 a.m.

    The scheduled trigger starts the Orchestrator.

  2. Step 2

    The Data Engineer agent pulls and cleans the data from all 4 marketplaces.

  3. Step 3

    The BI agent analyses the data and drafts the visuals and report.

  4. Step 4

    The Communications agent emails the summary to the team for review.

  5. Step 5

    The team approves or flags issues (human in the loop).

  6. Step 6

    The final report is produced and shared, and the Q&A agent is ready for questions.

What the reporting tracks

  • Sales by product
  • Review ratings and count
  • Keyword performance
  • Month-over-month trends
  • Best performers and slipping products
  • Profit margins by product and channel

Used to protect margins, adjust pricing and guide marketplace advertising.

How it was built

  • Claude Code
  • VS Code
  • Agent skills and guardrail files
  • MCP and API connections
  • Scheduled tasks
  • Power Automate
  • Power BI
  • AI policy alignment
  • Team training

Everything lived in one structured project: instructions, skills, guardrails and source descriptions, so the agents run consistently and stay within clear rules.

The result

~90%

less time spent preparing reports. The team now steps in only at review points, checking the summary of sales, products and trends before the report goes out.

  • FasterReports ready the same morning
  • More preciseThe same rules applied every day
  • Better decisionsPricing and ads informed by fresh data
  • Compliant and adoptedAligned with the company's AI policy, and a team trained to run it

Wow, I didn't know AI could do that. We could have been spending all this time on other tasks. We're very pleased.

— Operations team, consumer-products brand

We would like to know how you do this.

— Leadership team, consumer-products brand

Want this for your sales channels?

Get this builtSee more use cases

Tell us about yourself or your business

Choose a package and share a little about yourself or your business. We'll reply within 2 business days with full details.