KOJI Figma plugin cuts 300 ad resizes from 30 hours to 4

We resize ad campaigns for global brands, often into hundreds of formats in several languages at once, and every round used to be done by hand. Our in-house Figma plugin now does the repetitive part. A batch costs 2–5x less, and designers spend their hours on the master creatives.

The problem

Performance marketing runs the same loop for almost any brand: generate hypotheses, test them, scale. Scaling has two meanings here. One is testing more hypotheses. The other is expanding a single winning scenario into dozens or hundreds of placement formats. The latter is exactly where our process broke.

One campaign could need 700–900 formats across several languages. Getting from an approved idea to a live campaign took between a few days and a week.

The solution: an AI plugin for Figma

We built a plugin for our designers that runs inside Figma and generates the variants. The old chain had 10 handoffs and approvals. The new one has four: build the master banner, generate the resizes, get one round of client sign-off, export. The client has 200–300 resizes ready for the platforms within 2–3 hours.

Ad platforms rarely send work back now. Disclaimer contrast and text height, borders, file weight in kilobytes: we used to settle these one format at a time, and now the plugin applies them on every export.

Step 1: Cluster the formats

The plugin reads the technical specs from every file they arrive in and groups formats by proportion. Tall verticals go together, then near-squares, wide horizontals, and extra-narrow strips.

Step 2: Build one master creative per cluster

That makes 5–8 master banners, which KOJI designers build by hand. Everything after this step runs automatically.

Step 3: Generate and export the full batch

The plugin turns the 5–8 masters into 200–300 resizes in one click. On export it creates the folders and names each file by platform, placement city, and pixel size.

The first batch for a brand still gets manual QA. Once that brand's rules and its platforms' rules are in the templates, later batches ship without a second check.

The results

Turnaround per batch fell from a week to a day. Team time fell from about 30 hours to about 4. Production no longer caps how many hypotheses we test in a month. The marketer's ideas do, and we estimate that allows up to 10x more tests.

The scaling budget drops 2–5x, and the savings grow with volume. The 5x figure appears when many campaigns run through the pipeline.