Gen-3 I: Introduction

Introduction to Gen-3

Welcome

BASES is Nielsen Innovation Practice. Go through the module to learn more about your role & responsibilities in Nielsen Innovation.

Learning Objectives

Gen-3 is an innovative forecasting model. During this training you will:

  • Learn the importance of choice share
  • Discover how to estimate volume from the state transition probabilities table
  • Understand the forecasting challenge when we have a heterogeneous population and use "cohorts" to solve this dilemma
  • Recognize Gen-3's calculation sequence

What is Gen-3?

Gen-3 is a forecasting model which is currently used for general restage forecasting studies and is also used for all DecisionPoint forecasting. 
 

Choice Share

What is Choice Share?

Choice share is the probability consumers will purchase the tested item the next time they purchase the category assuming the product is available at base price with base level shelf presence. Gen-3 uses the NBD-Dirichlet theory to combine the two probabilities.

It is important to differentiate choice share and the probability to purchase.

Choice Share Probability to Purchase

Choice Share is:

The probability to purchase the item given the consumer is already purchasing the category

Probability to Purchase combines:

  • The probability consumers will purchase the category
  • The choice share (probability)

Choice Share Criteria

Choice share is just probability to purchase the item... when consumers shop the category!

Choice Share depends on:

  1. If consumers are aware of the product.
    For a new item, when consumers shop for the category, the probability that they become aware at shelf and immediately buy (Non-aware trial) is lower than if they are already aware of the product via previous media or shelf exposure (aware trial).
  2. If consumers have tried the product.
    For example, choice share among triers, non-repeaters (1st Repeat choice share) will be higher than among non-triers if the concept is weak but the product is strong, and vice versa.

Choice Share Definitions

Click the images below to understand Choice Share definitions.

Choice Share Calculations

How do we estimate different choice shares at the base price (with average level shelf presence) for new items? 

State Choice Share Choice Share Calculation
State 1:
Non-Aware
Non-Aware Trial Choice Share For new item, the choice share for state 1 ( non-aware ) will be small, and default is the aware trial  choice share adjusted downwards by the probability of becoming aware at shelf.
State 2:
Aware, Non-Triers
Aware Trial Choice Share We show consumers the concept stimulus, thus they are aware of the concept and it can be used as a surrogate to reflect the aware trial  choice share. 
For Decision Point/Virtual Launch, choice share can be found in the simulator.
State 3:
Triers
First Repeat Choice Share In a BASES II forecast, we let consumers try the product, and they have tried once, thus the key measures can be used to reflect the 1st repeat choice share.
State 4:
Repeaters
2+ Repeat Choice Share We assume this is the same as State 3 (First repeat choice share). If you have Extended usage Option KM test results, you can run through Step 4 tool to obtain 2nd repeat choice share.

State Transition Probability

State Transition Probability Introduction

For New Items, everyone starts as "Non-Aware..."
Over time, driven by both internal factors and external factors, consumers progress from State 1…State 4 (or regress if they forget) based on a set of probabilities called the State Transition Probability. 

What factors drive consumers between states?

Aware

What's the probability consumers will become aware of the product?

Similar to the Global model, Gen-3 has an awareness sub-model to estimate the probability that a non-aware HH will become aware of the product.

Distribution awareness is based on Quantity, Quality and Purchase Frequency. It can be influenced by:

  • External Factors - Item availability: Distribution quantity (PCW/ACV) and quality
  • Internal Factors - The frequency of consumers to visit the store and shop for the category

Advertising awareness is based on Quantity and Quality and can be influenced by:

  • External Factors - Advertising Quantity (eGRPs) and Advertising Quality (copy quality, it's potential to generate awareness)
  • Internal Factors - Consumers' media viewing behavior


Forget

What is the probability aware consumers will forget?

Without ‘reminders’, consumers may forget (become non-aware). The forgetting factor is similar to awareness decay in the Global Model.

The likelihood of respondents to forget also depends on:

  • External Factors - The absence of further GRP's leads to awareness decay
  • Internal Factors - whether they have tried the sample and whether they have purchased the product

In Gen-3, we assume respondents who have purchased the product at least once will not forget the product.


Non-Aware Trial

What's the probability consumers will try the product without previous media exposure?

The probability of purchasing an item (which can occur in states Non-Aware Trial, Aware Trial or Repeat) is governed by two factors: Propensity and Opportunity.

Opportunity to purchase is based on Category and Distribution. It can be influenced by:

  • External Factors - Distribution (item availability)
  • Internal Factors - Category Purchase combining the probability consumers will purchase the category with choice share (probability that the consumers will purchase the item when they shop for the category) yields an important piece to estimate, the probability that consumers will purchase the product

Purchase Propensity is based on State, Choice Share at the corresponding state and Appeal on Shelf. It can be influenced by:

  • External Factors - Persuasiveness of in-store presence (Price, Displays,etc)
  • Internal Factors - State of Consumers (Non-Aware, Aware Non-Trier, Previous Purchasers)


Aware Trial

What's the probability consumers who are aware of the product will try?

The probability of purchasing an item (which can occur in states Non-Aware Trial, Aware Trial or Repeat) is governed by two factors: Propensity and Opportunity.

Opportunity to purchase is based on Category and Distribution. It can be influenced by:

  • External Factors - Distribution (item availability)
  • Internal Factors - Category Purchase combining the probability consumers will purchase the category with choice share (probability that the consumers will purchase the item when they shop for the category) yields an important piece to estimate, the probability that consumers will purchase the product

Purchase Propensity is based on State, Choice Share at the corresponding state and Appeal on Shelf. It can be influenced by:

  • External Factors - Persuasiveness of in-store presence (Price, Displays,etc)
  • Internal Factors - State of Consumers (Non-Aware, Aware Non-Trier, Previous Purchasers)


Repeat

What's the probability consumers who have tried the product will repeat?

The probability of purchasing an item (which can occur in states Non-Aware Trial, Aware Trial or Repeat) is governed by two factors: Propensity and Opportunity.

Opportunity to purchase is based on Category and Distribution. It can be influenced by:

  • External Factors - Distribution (item availability)
  • Internal Factors - Category Purchase combining the probability consumers will purchase the category with choice share (probability that the consumers will purchase the item when they shop for the category) yields an important piece to estimate, the probability that consumers will purchase the product

Purchase Propensity is based on State, Choice Share at the corresponding state and Appeal on Shelf. It can be influenced by:

  • External Factors - Persuasiveness of in-store presence (Price, Displays,etc)
  • Internal Factors - State of Consumers (Non-Aware, Aware Non-Trier, Previous Purchasers)

State Transition Probabilities Table

Combining these factors gives us the State Transition Probabilities Table. 

Legend:

A= The probability that consumers will become aware of the product.

F= The probability that consumers will forget.

T1= The probability that consumers will try the product without previous media exposure.

T2= The probability that consumers who are aware of the product will try.

R= The probability that consumers who have tries the product will repeat.


State Transition Probabilities Example

The Example demonstrates a State Probabilities Table for a new item. 

Legend:

A= 10% (The probability that consumers will become aware of the product)

F= 5% (The probability that an aware consumer will forget)

T1= 0% (The probability that consumers will try the product without previous media exposure)

T2= 5% (The probability that consumers who are aware of the product will try)

R= 15% (The probability that consumers who have tried the product will repeat)

Heterogeneity

Introduction to Heterogeneity

Applying Heterogeneity

Using the State Transition Probability table we will isolate how a factor, for example Awareness Generation, is applied.
 

Example: Awareness Build (with Homogeneous Population)

Homogeneous Population where all population members:
Pr {become Aware} = 10% 

Chart demonstrates a Homogeneous Awareness Build Curve.

Example: Awareness Build (with Heterogeneous Population)

We know in reality, consumers differ in viewing behavior and probability of becoming aware of the product, for example:

  • 20% are “heavy viewers” of TV and have Pr {become Aware} = 30%
  • 80% are “light viewers” of TV and have Pr {become Aware} = 5%

So if the population average still has Pr {become Aware} = 10
In this case, the awareness build is different … This chart provides comparison of Heterogeneous and Homogeneous Awareness build. 

Heterogeneity Nature of Population



We observe this whenever we try to take a simple average for any incidence measure, including:

  • Awareness
  • Category Purchase Frequency
  • Trial Purchase Propensity
  • Repeat Purchase Propensity
  • etc


These are all important in our forecast, but we need to introduce one more concept to ensure we correctly forecast. Cohorts...

Cohorts

Introduction to Cohorts

Gen-3 divides households into small, homogeneous sub-groups (i.e. “Cohorts”) with similar characteristics relevant to buying behavior. 

Criteria to design Cohorts





When creating cohorts, we group Households with similar Awareness, Trial and Repeat probabilities together to apply the same state transition probabilities table within each cohort.




Households are grouped based on the following drivers:

  1. Media viewing behavior (probability of becoming aware via media)
  2. Category purchase frequency (probability of awareness at the shelf and purchasing)
  3. Non-Aware trial propensity
  4. Aware trial propensity
  5. Repeat purchase propensity

Each Cohort is defined as a combination of levels of these drivers.

2,500 Cohorts in Gen-3

Households with similar Awareness, Trial, and Repeat probabilities are grouped, with a State Transition Probabilities table for each Cohort. 
The five criteria are divided into 4-5 groups, then 'crossed' (in a cross-tabulation) to form a cohort. The behavior of each cohort over time is then treated almost as if it were one individual, except that the individuals within a cohort can be spread across the different states: 1,…,4.

Click each image to see criteria used to define Cohorts.

Calculation Sequence of Gen-3

We are ready to calculate volume. 
Steps Volume Calculation
Step 1: 
At launch of a new item
All cohorts are assumed to be in State 1 ="non-aware" of the item.
Step 2: 
At Period (t)
For each Cohort the State Transition Probability table is calculated, based on marketing variables at Period (t).The state transition probabilities table is unique for each cohort and each period.
Step 3: 
These Transition Probabilities are applied
To the State Probabilities for the Cohort at the end of Period (t-1) to give the State Probabilities for the Cohort at the end of Period (t).Steps 2 and 3 are repeated for t = 1, 2, … as required.
Step 4: 
Volume is modeled separately for each Cohort
Based on its unique state transition probabilities table for each period with results aggregated at the end.Because the model uses distributions such as NBD, it can calculate how many households there are at each level of media viewing, category purchase, item choice etc.
Step 5: 
Reported results are the aggregation of the cohorts
 Representing the population.

Gen-3 Outcome...

...results in Awareness Build and Trial and Repeat curves for the population, deriving usual measures. 

While we know a lot, it's important to pair this with heterogeneity.

Using Gen-3 for Restage Forecasts

Restage Forecast Overview

Restager utilizes a two-step process to estimate the sales potential for the restaged product:

  1. Calibrate/Customize the restager model to predict known past-year volume of the parent brand.
  2. Use the calibrated model to estimate volume for the restage.

Click the launch sequence stages to learn more.

Calibration

In a restage study, we have historical by-period in-market sales data and marketing plan for the current cell. The model will run many times and readjust the settings so the calibrated results best fit the actual sales data.

The key parameters the Gen-3 model adjusts during calibration are:

  1. Average Choice Share
  2. Price elasticity/sensitivity to trade promotion
  3. Base trend and equity decay rate

This iterative process is called auto-calibration.

Calibration Step 1: Average Choice Share

Choice shares for different states differ for new items. However, for established brands, since it is difficult to distinguish first time buyers from repeat buyers (and majority of consumers are repeat) we assume choice shares are the same for all states for the control estimate.

While you may have entered buyers/non-buyers, concept/product choice shares, you will only see one calibrated choice share.

Gen-3 runs the forecast over & over to calibrate choice share with actual sales and marketing plans, therefore there is no need to wait for key measure scores to autocalibrate. 
As soon as you receive the marketing plan and historical sales, you are ready to calibrate!

 

 

  • Interim Period: 

Since there is no change in concept proposition, the interim choice share remains the same as the current item.

  • Non-Aware Trial Choice Share: 

For a restage, non-aware consumers have a propensity to buy the Restage similarly to the Current line.

  • Aware Trial/ Repeat Choice Share: 

(Calibrated choice share for the period) x (Estimated trial (or repeat) choice share from Step 4 for restage vs. current). 
 

Gen-3 Example

Calibration Step 2: Calibrate Promotion Sensitivities

Through matching the by-period fluctuations in sales volume and the marketing plan, Gen-3 calibrates the marketing plan support for each cohort to move from states (e.g. from non-aware to aware, from aware to trial), thus the volume lift effect.

This calibration process is how Gen-3 can provide volume sensitivity to different trade promotions, for example:

  • Price elasticity
  • Display
  • Feature




In the restage forecast, the model uses the calibrated sensitivity with by-period marketing plans and the heterogeneity, to estimate the promotion lift effect of various marketing plan inputs.

Calibration Step 3: Calibrate Equity Decay Rate

Without advertising, choice share tends to decline due to competitive pressure.

  • Over time, due to competitive pressure, if a brand is left unsupported, the probability for consumers to buy your brand will decline.
  • The choice share declines at a steady rate, leading to a decline in equity stock. We call this steady decay the equity decay rate.

This can be seen in the following Gen-3. 


The auto-calibration process finds the equity decay rate that best fits final volume.



The equity decay rates for the interim and the restage estimates are assumed to be the same as the control estimate. Without advertising support, the model assumes the choice share for current, interim and restage will all decline at the same rate.

 

Gen-3 Example

Autocalibration Results


Adjusting Gen-3 parameters occurs by choosing “Auto-Calibrate” in Gen-3. It iteratively runs forecasts to optimize the parameters:

  • Choice share
  • Price-effect (elasticity)
  • Feature effect
  • Display effect
  • Trends


This can be seen in the following Gen-3. 


We need to be patient running auto-calibration, due to the amount of forecasts being run simultaneously.

Gen-3 Results

Results = Restaged Volume Forecasts

Using the NBD-Dirichlet theory to account for the heterogeneity, Gen-3 puts together the by-period marketing plan, the category dynamics and by-state information to generate a volume estimate!

In this training we have summarized the "black box" of the Gen-3 model. Additional Gen-3 training is available on:

  • Gen-3 inputs
  • Analyzing a Gen-3 forecast