Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

CBC vs ACBC vs MBC for finding the optimal product portfolio for TV packages

I want to find the optimal product portfolio for my TV packages, think: SKY TV in the UK.

There's an array of packages today with bundled channels based on their genres, e.g. sports, movies, kids, etc.

There's also other non-channel related features like broadband, cloud recording, multiroom services that I can bundle together in my to-be portfolio.

My question is: what is the optimal conjoint type for me to get the output I need (utilities, attributes importances, etc) and also easy for the respondents to take and relate to in a real world?

I am thinking of CBC, ACBC, and MBC. Listing down my potential concerns with each below:

- Is this realistic to show to respondents? I've read briefly about how MBC is widely used in telco, and am wondering if a traditional ACBC setup will look very unrealistic to respondents

MBC (disclaimer: I've never done an MBC before):
- Does the output allow me to dynamically and flexibly play around with product features to find the optimal portfolio in the simulator?
- I might have many "levels" for certain attributes, e.g. Broadband, I want to split them by brand and speed. In ACBC I can easily split this into 2 different attributes, but with items in MBC, I am constrained

Thanks in advance, Sawtooth team!
asked Jun 15, 2020 by Troubled_Nerd

1 Answer

0 votes
A bit overly-simplistic but probably ok way to think about things is with a traditional conjoint analysis (CBC or even ACBC) we're going to show predetermined packages to people and ask them to choose the one they like the best, or accept or reject those bundles.  With a menu-based approach, we're going to invite the respondent to build their own bundles and vary pricing, availability, bundling discount, etc.

All approaches could be used, but there's an attractiveness to MBC if it more closely matches what a respondent would end up doing in the real world to build their own bundle.  A traditional or adaptive CBC could definitely be used as well, though.  Adaptive CBC is potentially a nice option because it allows you to treat price more like a continuous range and apply price adjustments based on levels.  This means you can, on average, show bundles with lots of things at higher prices.

MBC projects are a definite step up in difficult for a few reasons.  The first is that you typically need someone with some web skills, like HTML, CSS and JavaScript because there is no MBC exercise type in Lighthouse Studio.  The typical approach would be to use custom "Free Format" questions and chop up an existing CBC design to control what shows up and where on your menu.  Traditional conjoint exercises don't require this because you simply enter attributes and levels and the software makes the choice tasks for you automatically.

The second skill set needed would be some data manipulation capabilities so you can take the CBC design and choices from the software and put them together in the right format to begin working with the data in the MBC software.  Think scripting capabilities to read in two CSV files and spit out a joint CSV file of both.

The third skill is being familiar with experimental designs and regression analysis in general.  The MBC will do the hardest work on running the models for you, but when you import the raw design and answers you have to tell the software which variables are independent and dependent variables, and specify the structure of the model (variable 1 was hamburger price and should be an input into variable 10, hamburger choice, but also variable 2, cheeseburger price, is an input into hamburger choice as well because we expect them to be substitutes).  The MBC software does contain some tools to help you decide on specifying the structure/effects of the choices, but at the end of the day you start out with a blank slate and have to structure things from scratch.  In a traditional CBC, at most we tell the software we want an interaction effect or a constraint, but most of the time we just hit the "Estimate" button and run a model.

The MBC software does contain a simulator that works very similarly to a traditional conjoint simulator, and your primary output would be the share predictions given a specified menu of options.  Although utility scores are generated, because of the complex nature of the model (inherently desirability of an item plus a price utility plus a substitution effect, for example) we generally don't report average utilities, try to figure out attribute importance, etc.

Hopefully that gives a nice overview.  We definitely don't want to scare anyone away from an MBC project, but it is definitely a lot more difficult to pull off than a traditional conjoint.  We'd be happy to set up trial licenses for you for any option if you wanted to take a stab at things first before deciding what direction to go.
answered Jun 15, 2020 by Brian McEwan Platinum Sawtooth Software, Inc. (52,630 points)
Thank you for your detailed clarification, Brian!

I am particularly concerned about how I can use the output of an MBC model, and whether I am familiar enough with MBC in general to use the data wisely and sufficiently.

In CBC/ACBC, I can easily run estimates and get the attributes importances and utilities for me to analyze the results and build a simulator using the utilities.

From what you're saying, it seems like I cannot do the same with MBC. I've been trying to do more research on the output typically derived from MBC and what can be done with it, without much success. Could you shed more light on the data output of an MBC and how to use them?

Thank you!
Shoot me an email at brian@sawtoothsoftware.com and I'll share a link to a webinar we did on MBC. I think that would be the best starting point to see some real examples and wrap your head around how an MBC project works with us.