Avantia, the technology-enabled insurer behind HomeProtect, has updated its retail pricing model, employing Bayesian optimisation as an alternative to traditional A/B testing.

The new approach uses machine learning to more quickly identify the optimal price for its products, achieving this with fewer rounds of iteration and with less reliance on human involvement and the biases that entails.

The company anticipates a 10 per cent increase in yield, coming from both incremental sales and an increase in revenue per sale.

Chief executive Mark Eastham said that traditional methods of retail pricing can be costly as they take too long to hone in on an optimal value. “This new technique changes the game by allowing us to test more prices more quickly, and identify otherwise unexplored price points with potentially higher yields.

“We’re at our best when we blend an intense focus on data science with our high volume, dynamic trading culture – too often insurers are like scientists in a lab waiting for the experiment to end; we can’t afford to work like that.”

Bayesian optimisation is a statistical method for finding the highest point on an unknown function or curve. Applied to retail pricing, this means the method will find the optimal price point on a yield curve based on sales volume and revenue.

The Bayesian strategy is to treat the unknown curve as a random process that captures its ‘beliefs’ about all possible shapes this curve could take.

After deploying a price and observing its yield, the uncertainty about the possible shape of the curve is reduced around that deployed price point, but will still be large for untested points. The next price point to try is determined by selecting the point with the biggest expected improvement, in other words, the price point that maximises the expected increase in yield compared to the previously observed optimum.

Bayesian optimisation keeps repeating these steps of updating its ‘beliefs’ of the shape of the curve based on new observed data, and suggesting potentially optimal prices to deploy next, until the observed yields do not increase any further, indicating that the maximum has been found.

Share Story: