Using AI to Generate Promotional Brochures and Mailers

Design in creative industries is like an iceberg: there's 10% of highly visible work (popular billboards and award-nominated visuals) and then another 90% under water that nobody notices. During creative.ai's VC-funded phase, we spent most of our time on the latter, since AI and generative techniques are particularly well suited there.

One specific example is the design of promotional brochures and mailers that you find in supermarkets and pharmacies, or those that get dropped into your mailbox. Many people rely on these brochures to get good deals on products they need, or to find coupons for cheaper alternatives.

It takes 2 to 3 professionals to produce just one multi-page brochure every week plus its format variations, and that's for a single client! There's a lot of repetitive and manual design involved, and due to the tight deadlines it means the quality suffers if anything comes up — such as last minute client changes or staff sickness/shortages.

The plan was to deploy AI in a way that could alleviate the burden and overtime with minimal disruption, allow creative teams to spend their time improving the quality rather than meeting deadlines, and give freedom to clients to make late changes. The first step was to derisk the technology with a prototype to the point where the tasks left for pre-production were mostly predictable. It went better than expected!

PILOT PROTOTYPE

Procedural generation of a single product cell for soap and shampoo, with placeholder text, product IDs, and images. Other prototypes (not shown) looked into connecting generator to use the client assets directly, and reproducing the agency's existing templates. Of all the visual design prototypes we produced at creative.ai during its VC-funded phase, this was one of the fastest! [2018]

This pilot was succesful in derisking the techology side, how we could apply our systems to produce a stream of random cells each with sensible layouts. B2E development can be slow at first, but in this case moved faster than expected and the agency was keen to move forward with closer integration.

We learned multiple lessons from working on this prototype:

1. Content Management is the biggest piece of work! Many clients didn't use databases or revision control (the situation slowly improving since), and creative.ai had hired more developers to take on these backend challenges — which is exactly what VC funding is for!

2. Agency Morale: the leadership of the creative agency was concerned about HOW to roll out this technology, and its impact on the team's morale — rightly so! It took us months to figure our they were all afraid, and when we switched our approach to education/coaching then everything clicked...

When seen in the context of industry-wide transformation of work due to automation (read this previous article), you can understand why applying such technology can easily cause tremendous disruption when it's not done carefully.

Unfortunately, we couldn't find a way to make that work within the constraints we had as a VC-funded startup the time... Not the ending you'd expect from a blog post, but that's startup reality for you!