How we validated Fairfirst's online revenue channels with an award-winning experiment
Fairfirst had the product for years. They were contemplating whether to kill it. We gave them a definitive answer.
- Portfolio
- Principal
- Client
- Fairfirst Insurance
- Sector
- General insurance
- Scale
- An estimated 2 million potential customers
- Where
- Sri Lanka
- Engagement
- Product feasibility validation, run through a paid-media funnel
- Over
- Nine weeks in market, 26 August to 31 October 2024 · 2024
The product was finished, category-first, and almost nobody would buy it.
Insurance in Sri Lanka is bought from a person. The standard route was a branch or an agent and roughly two hours of paperwork. Fairfirst built Digishop to remove all of it: motor and travel cover, bought end to end, entirely online, which no one in the category had done.
- 01Call an agent
- 02Wait for the form to come back
- 03Print it
- 04Fill it in and scan it
- 05Email it back
- 01Open Digishop
- 02Complete the purchase end to end, online
- 03Policy issued. No call, no agent, no paper
On paper the value exchange was not arguable. Fairfirst put the waste at four million hours a year across an estimated two million potential customers.
People came to the page, and then went and did it the two-hour way instead.
That is a harder problem than low awareness, and a more expensive one. A company that has built a channel nobody uses has to decide whether it built the wrong thing. The brief we were given was not “make this popular”. It was “find out whether this is real, before we spend another year on it”.
They were not rejecting the product. They were rejecting being the first person to try it.
Both audiences already transacted online everywhere else. Travel buyers were digitally fluent professionals who used apps for transport, banking and food, and then bought insurance by printing and scanning a form. Motor buyers used digital services sometimes, and reserved agents for anything that felt important.
Neither group described the product as bad. People who had used Digishop described it as simple. The refusal sat entirely at the starting line: what if I enter the wrong thing, who do I call if I get it wrong, is this actually secure. And insurance is a category where reassurance has to come from other buyers rather than from the brand.
A closed loop like this is the useful kind of diagnosis, because it tells you what not to spend money on. More reach pours more people into the top of a funnel that is leaking for a reason reach cannot touch.
The known route felt safer than the faster one, and cost was no argument against it. Two hours with an agent was not experienced as two hours wasted; it was experienced as two hours of someone else being responsible.
In a high-trust category with no visible precedent, going first is not a convenience decision. It is an exposure. The behaviour had no social proof because it had no adopters, and it had no adopters because it had no social proof.
First, we turned it into a trial with a number that could kill it.
Before any media was booked, the business fixed two figures: a ceiling on what it would risk finding out, and the level of return below which Digishop would not get a channel budget.
“LKR 1,000,000 was determined as the maximum spend that the business would incur to trial this new sales channel. A ROI of at least 1 would signal the management to move forward.”
That is a go/no-go threshold of 2.0 ROAS, two rupees back for every one spent, declared in advance by the people who could have stopped the work. It is the single most important thing on this page, and it is the reason the result at the bottom means anything. A number you can only beat is not a test.
Everything after this point is instrumentation. The advertising is how we got traffic into the funnel; it is not what we were measuring.
Then we tested three explanations for the drop-off at the same time.
The funnel was leaking between quote-start and checkout. We had three competing accounts of why, and rather than argue about them we served all three, by pixel trigger, to the people who had actually abandoned.
Variants were served to quote-starters, abandoners and cart initiators separately, so a message could be judged against the stage it was meant to rescue rather than against the campaign average. Weekly, the ones that moved funnel progression took budget from the ones that did not.
The product pages and the checkout were being changed at the same time and on the same evidence: friction removed where the analytics said people were leaving, and the three-step journey mapped on the page so the length of it was visible before anyone started.
Only then did the campaign itself run: a hero film, cutdowns, testimonials and creator content, built to make a first purchase look like something other people had already survived.
What the work did.
- 8.92ROAS, against a go/no-go threshold of 2.0Revenue over ad spend, not a profit multiple. The entry describes both products as low-margin and highly price-competitive.
- 597completed end-to-end digital purchasesFrom a base the entry describes as negligible. No insurer in the market offered a comparable fully online flow.
- ~2%of Fairfirst's total insurance revenue is now digitalReported in the entry as “to date” and not tied to a stated period. From effectively zero.
A ceiling of LKR 1,000,000 on trial spend, and LKR 2,000,000 of revenue inside three months, a return of 2.0. Below that, Digishop would not get a channel budget.
8.92. Four and a half times the threshold that would have closed it.
The channel was approved and made permanent. Digital purchases now account for roughly 2% of Fairfirst's total insurance revenue, from a base of near zero, and Digishop sits in the product roadmap, the budget and the distribution plan rather than in an innovation slide.
Every figure on this page is Fairfirst's own, submitted by the brand to the 2025 Effie awards; “the entry” below means that submission. The strongest thing about this case is the order events happened in. The threshold was written down before the spend was committed, by people who could have shut the project down, and the result is reported against that threshold rather than against a benchmark chosen afterwards. That is the difference between a trial and a campaign with a number attached.
The attribution is also unusually clean, and by the client's own account. Across the period the product, the pricing and the purchase process were unchanged; no parallel Fairfirst campaign was driving the same KPIs; no regulatory or pricing change occurred that would have lifted digital insurance demand on its own. Only the communications moved. Purchases were counted as completed online checkouts on the two product pages, and funnel progression by pixel events at quote-start, add-to-cart and checkout.
What we will not do is dress up the rest. The entry reports that motor conversions “grew significantly”, travel conversions “surged”, drop-offs “declined meaningfully” and reach “exceeded prior campaigns”, and attaches a figure to none of them. Those are not results and they are not printed here as results. Three numbers on this page are real: 597 purchases, a ROAS of 8.92, and digital at about 2% of total revenue. Everything else in this write-up is description.
Four limits, stated rather than waited for. ROAS is revenue over ad spend, so it is not a profit multiple, and the entry itself calls these products low-margin, though the client's own threshold was set on revenue too, so the comparison is a fair one. The effort ran nine weeks against a rule written for three months. The 2% figure is given as “to date” with no period attached. And every source in the entry is the agency's own research: nothing here has been verified by anyone outside the work.
Who can vouch for this.
A product nobody buys is not always a product problem. Sometimes the only thing wrong with it is that nobody has gone first, and that is a cheaper thing to fix than a rebuild, if you are willing to find out before you commit.