Product Line Optimisation with Genetic Algorithms over Conjoint Simulations

Description

Searches for near-optimal product portfolios by running a genetic algorithm many times over conjoint market simulations, showing how many products to offer and which missing features matter.

A conjoint simulator can score any one product line, but the number of possible lines is far too large to test by hand. This method treats each product as a set of attribute levels (genes) and lets a genetic algorithm evolve portfolios whose fitness is their simulated preference share against a none option, or revenue, profit or several objectives, optionally with competitor products present. Because the search is stochastic, it is repeated many times for each line size, giving a distribution of good solutions rather than one answer. Plotting share by line size shows where adding products stops paying off, and summing share by attribute level across the optimal portfolios reveals features the current line lacks. Chapman's case used two conjoint datasets (CBC and ACBC) with identical attributes and checked that both told the same story.

Attributes

Loss type
Siloed
Effort
Weeks

Steps

  1. Collect high-quality part-worth utilities, ideally from two conjoint methods (for example CBC and ACBC) using identical attributes and levels.
  2. Define the objective: total preference share of your products against none, or revenue, profit or a multi-objective goal, and decide whether competitor products are in the simulated market.
  3. Encode each product as a vector of attribute levels and set up a genetic algorithm whose fitness function is the market simulation.
  4. Run the search many times (Chapman used 50 runs) for each candidate line size, for example 1 to 20 products.
  5. Plot the best share at each line size and look for the point of diminishing returns.
  6. For each attribute level, sum its share across the near-optimal portfolios to find valuable features that are missing from the current line.
  7. Compare the results across datasets and present the range of near-optimal lines, not a single answer.

Sources

  • Uncommon Choices: Novel Applications of Conjoint Analysis in Practice (Chapman, 2021)

Also known as

  • GA product line optimisation
  • Conjoint portfolio search
  • Near-optimal product line exploration
  • Product portfolio optimisation with genetic algorithms over conjoint simulations

Categories

  • Analytics & Behaviour
  • Statistics & Modelling