‘It's the thought that counts.’
The situation
Gift-giving is a universal human practice and most people are bad at it. Published research puts Australian spending on unwanted gifts at more than $900 million a year, with an estimated 270,000 tonnes of them ending in landfill, while three-quarters of Australians say they enjoy buying gifts for others. The cause is information asymmetry: we do not know what other people want, so we guess. The tools on offer, gift cards, wish lists, generic search and general-purpose AI assistants, each solve part of the problem and none of it. Large language models on their own invent products and links, run on stale training data, and cannot carry the context that makes a gift right for one person on one occasion.
What we were asked to do
Nobody asked. We set out to find whether AI could make anyone a good gift-giver, and to prove a commercial model for how to scale such a service. As part of the work we created social content to validate consumer demand and interest.
What we did
We researched how and why people give gifts, from the anthropology to the market data, and interviewed retailers and brands on what gift shopping costs them in missed sales. From that we designed the business: the proposition for givers and receivers, the gift profile at its centre, the revenue model built on affiliate commission so the core product could stay free for users, and the roadmap. We then built the recommendation engine. It combines large language models for understanding what a user asks for, predictive models of preference, and live search across the web for products, experiences and donations with links to buy, so a recommendation is real, current and purchasable. At its core is a probabilistic scoring framework, adapted from the Drake Equation, that scores every candidate gift across personal fit, budget, occasion, emotional impact, trend, intrinsic value, timing and novelty into a single relevance score. We published the framework as a research paper. We released the alpha Gift Finder to the market under our own brand, measured it, and designed the engine so it could be embedded in a retailer's online store as well as used directly by consumers.
The results
- 01Twice the engagement of the shopping-app category average. In alpha testing, users spent twice as long with Givving as the published average for shopping apps, and came back for repeat sessions.
- 02A working recommendation engine that returns real gifts with links to buy, across products, experiences and donations, rather than a list of ideas.
- 03A published research framework for probabilistic gift relevance scoring, which is the intellectual property behind the engine.
- 04Two routes to market designed in from the start: the consumer app, and the engine as a plug-in for retailers' online stores.
- 05Expressions of interest from some of Australia's largest retailers for the engine inside their online stores.