TL;DR: We tested AI-generated FB/IG ad creative to grow qualified leads for Fairmart’s SMB retail ICP. Anime-style, revenue-focussed creative (e.g. ‘selling while you sleep’) significantly outperformed realistic AI imagery, delivering a 3.48x ROAS, while also surfacing new lead-qualification challenges.

Problem & Context

At Fairmart, we sell eCommerce and POS software to SMB retailers such as minimarts, grocery stores, bike stores, pet shops, and similar specialist retailers. Our growth channels to date have been outbound sales, referrals, and inbound organic when prospects see our logo and equipment out in the field. Top of funnel has always been challenging for us, which leads our sales team (especially juniors) to spend 30 - 50% of their time just prospecting. We wanted to test paid channels, specifically Facebook + IG ads to see if we could grow our top of funnel, get more qualified leads to our sales team, and ultimately accelerate our revenue growth.

A second facet of the problem is that Singapore is already a structurally expensive advertising market due to its comparatively small audience size. We were targeting niche ICPs within that, pushing CPMs up even if we used interest-based targeting. Our ad budget was 1k SGD / month while experimenting and we did not have the budget to outsource for professional creative. What prompted us - pun fully intended - to try was the new AI tools available that opened new content options.

Objective

The project had two key objectives:

1. Test whether AI-generated creative can generate qualified leads within our ICP
2. Demonstrate that we could achieve a minimum ROAS of 2x from this channel

Solution

Our first approach was to use Midjourney to create realistic images of our target customers. We tried a broad range of variations, such as various value propositions, CTA variations, image vs video, and graphic overlays. Overall, the technology for generating AI creative has got very good and we were able to try a lot of things with internal resources only. We were able to get some leads but our CPM was consistently in the S$20 - S$25 range, making it prohibitively expensive.

During this time the OpenAI Studio Ghibli meme was all the rage so we decided to try it out for our audience as well. From our first attempt, we felt that AI-generated humans never look quite right, nor was the creative that visually engaging. Perhaps a more playful approach would draw better engagement.

From our ICP research, we understood that the #1 reason for any SMB retailer to sell online is to drive sales. We framed this in a variety of ways, but the most effective was ‘selling while you sleep’. Our sales team used this positioning on demos and we saw how it resonated with our target audience. We tried alternative value propositions focussing on the benefits of specific features, but those never achieved anywhere near the results we got with the revenue-focussed pitch.

Lastly, on bidding: each creative was applied to an awareness campaign coupled with a retargeting campaign. We used this combination to keep costs down while finding relevant audiences outside of interest-based targeting. When assessing creative performance, we combined the awareness and retargeting performance to get a consistent picture.

Results

Our results achieved our objectives and demonstrated that we could source relevant leads for our ICP through AI-generated creative, and that these leads could convert to real revenue. However, along the way we learnt additional important details that were arguably more interesting:

• While we addressed the top of funnel in terms of raw leads, we created a new bottleneck in qualifying those leads. A common reason people submitted interest was to learn how they could open a minimart and generate income versus already running one. Sifting through these leads took significant time from our sales team. We need to address this to scale, either by improving targeting or qualification to sift out aspirational business owners.
• We struggled to get raw volume from pet creative. One theory is that the segment is just too small, which is supported by business registration data showing there are roughly five times more minimarts than pet stores. But the gap felt too big for this to be the only factor. We know from anecdotal experience that pet stores as a category were generally under less pressure, so perhaps the ‘revenue boost’ value proposition resonated less with that audience because they felt less pressure to drive sales. In any case, we would not have expected such a big gap between the two.
• Kopitiam-themed creative showed the highest engagement but delivered unqualified leads (coffee shop owners) outside of our ICP. The creative depicted a person at a kopitiam monitoring their store, so we thought it would resonate with other SMB owners. In practice we saw mostly coffee shop owners or aspirants engage with that creative.
• ‘Realistic’ creative is not effective in terms of engagement. Our theory is that the people depicted fall into the uncanny valley of just looking not quite right. The anime creatives also had the advantage of being warmer and being on-trend with the Studio Ghibli meme.