Marketing strategy
A/B testing on a small budget: make three versions, keep one
Chrisphine Otieno · 19 Oct 2026 · 2 min read
Large companies rarely run one version of an ad. They make several, test them on small audiences, and put their money behind the winner. Small businesses usually could not afford to: one shoot, one video, fingers crossed.
AI video changes the economics. When a new version costs a few credits and a few minutes instead of a new shoot, testing becomes possible for anyone.
What to test
Test one thing at a time, so you know what made the difference:
- The hook: the first two seconds. This usually matters most. Hook, proof, offer
- The model or style: grounded and real (AK1) against bold and cinematic (AK2)
- The language: Kiswahili against English as Kenyans speak it, or the mix
- The offer: a discount against free delivery
- The length: 10 seconds against 20
How to run a simple test
- Make the versions. In Tari Studio, make your first ad, then use Another take for a variation with the same settings, or Improve to change one thing and get a new quote. Each quote shows the credit cost before anything is made. How the Studio works.
- Give each version its own tracked link. When you publish, add a tracked WhatsApp link to each post, so you can see which version brought conversations. WhatsApp and leads.
- Post them under similar conditions: the same platform, similar times, ideally similar audiences.
- Wait long enough. A day is rarely enough. A week usually is for a small business.
- Compare what matters: conversations and sales, not just views or likes.
- Keep the winner, learn why, and test the next thing.
Measure the right thing
Views are easy to get and easy to misread. A funny ad may get thousands of views and no sales; a plain one may get fewer views and more orders. The tracked link tells you which version people acted on. Measuring what matters.
Learn as you go
Rate results in the Studio with a thumbs up or down and a reason ("hook too slow", "love this lighting"). The Studio keeps short notes about your brand from that feedback, and the AI director reads them on every future request, so your next ads start from what worked. Brands and brand memory.
Keep a simple log
A notebook or spreadsheet with the date, what you tested, the versions and the result is enough. After a few months you will have a playbook of what works for your customers that no agency could sell you.
Be honest in every version
Testing is about presentation, not truth. Never test a version with a price, claim or offer you cannot honour. Every version has to be one you would be happy for customers to hold you to.