Skip to main content
Work with us

Case study

The social media team that runs itself

Kwanda funds community projects. Partners film the work on their phones. We built the system that turns those clips into social media content at scale.

At a glance

Service
Operations
Built with
Google Drive, Notion and AI
Output
Up to three posts a week
Running cost
Under $20 a month

Kwanda's partners film their projects on their phones and send the clips in from different countries, in different ways. Kwanda needed a steady way to turn that footage into social media content that shows donors the work on the ground. Now every clip has one home, and new posts arrive ready to approve.

How it works

Six steps. The agent does the work, with a human in the loop to approve.

Step 1

Agent

A new folder for every partner, every month

On the 1st, the agent makes a dated folder for each partner and emails them where to send that month's clips. Partners always know where their footage goes.

On the first of the month, a new November folder appears inside each of 4 project folders.

Step 2

Agent

Every clip is checked

The agent checks each new clip as it arrives. Landscape clips, clips over a minute and broken files are set aside, so only usable footage goes forward.

Three clips arrive. Two pass the check. One is set aside because it was filmed in landscape.

Step 3

Agent

Then labelled and filed

The agent watches each clip that passed and labels what is in it and how good it is. Everything lands in one media library, ready to search.

The clips that passed appear in a media library, each labelled by project and subject, with a quality score.

Step 4

Agent

Three stories picked each week

Each week the agent picks three posts to make. Each is a project and an angle: the build, the result or the people. It remembers what it used before, so stories don't repeat and fresh clips come first.

Seven partner clips play side by side. Three are picked, one from each project, each for a different angle: the build, the people and the result.

Step 5

Agent

Each story becomes a finished post

The agent cuts the chosen clips together, adds captions in Kwanda's style, lays music underneath and closes on Kwanda's end card. Three posts, ready to review.

A video editor: four clips are placed on the timeline, captions and music are added, an end card closes it, then the finished post plays in the preview.

Step 6

HumanAgent

Posts schedule themselves

A human in the loop watches each post and approves it. It is then scheduled straight to Instagram. Nobody uploads files, copies captions or remembers to post.

The approved posts appear in the posting queue, each marked as scheduled.

Results

20 seconds

To approve a post. Editing and scheduling one by hand used to take up to two hours.

Averages per post from the first month.

Average reach per post
1,360
Average engagement
9%
Average shares per post
13

Got a job like this?

Tell us which job is taking your team too much time. We'll show you what it looks like automated.

Work with us