Simulating SA consumer reaction to ‘cow-free’ milk to find the right price & features
Published by Craig Kolb in Market simulation · 16 December 2022
Tags: Finding, the, best, price, conjoint, analysis, UHT, Milk, Cow, Free, Milk, Lab, Milk
Tags: Finding, the, best, price, conjoint, analysis, UHT, Milk, Cow, Free, Milk, Lab, Milk
Dairy farmers, biotech startups and beverage brands may be affected by the impending entry of ‘cow-free’ milk technologies onto the SA market.
Ballpark forecasts of consumer response under different future market scenarios may be helpful in providing an indication of how demand for traditional cow milk may change.
A conjoint analysis survey was conducted amongst South African grocery shoppers in February 2022 to assess consumer reaction to two new ‘cow-free’ milk technologies. Conjoint analysis is a methodology that provides ‘ballpark’ forecasts based on specific feature and pricing configurations.
The first technology uses cells in a bioreactor to produce milk. These are the same cells cows use to make milk (derived from stem cells). The second technology uses genetically modified flora (e.g. fungus) to produce milk.
In this article I will show the results of simulations for two sets of scenarios – the first set focused on the viability of launching a new biotech producer (without a dairy herd) and the second set focused on incumbent adoption of cow-free technology, either as an outright replacement for cow milk or as a product line extension.
This research takes no position on whether or not this is desirable or healthy. The aim is to provide ballpark forecasts and to demonstrate the usefulness of conjoint analysis in anticipating consumer reaction.
Conjoint analysis survey
Conjoint analysis is a survey-based technique that allows you to quantify how important product features and pricing are to a consumer, without asking directly. Simulators can then be built to estimate consumer response under different scenarios.
There are many forms of conjoint analysis. In this case metric conjoint analysis was used, as it has numerous advantages over choice-based conjoint (discrete choice models). A frequent objection - not being able to estimate a ‘none’ option - has been overcome through a recent methodological improvement introduced by the author.
The first step involves breaking products down into their component parts – called attributes which are further broken down into mutually exclusive levels.
It was decided to focus on the UHT milk market – in particular the most widely available 1 litre SKU. Five different attributes were the focus of the study: brand, price, type, cap and pack. A decision was made to restrict the brand levels to the leading non-store brands. A mockup of a hypothetical new brand ‘Kinder Green’ was also included to illustrate how the potential of a new biotech entrant could be assessed using conjoint analysis.
Table 1 Attribute levels and map
An algorithm was then used to construct artificial product profiles. The algorithm is designed to collect as much information about consumer reaction with the minimum number of product profiles. In this study it was only necessary to test a subset of 17 of the possible alternatives using a main-effects assumption.
The profiles were shown along with images of each brand’s milk packaging (1L SKU). A packaging mockup was created for Kinder Green.
Figure 1 Example of one of the profiles shown in the survey
After asking various awareness and behavioural questions, participants were exposed to a textual description of the new ‘cow-free milk’ concept (see appendix A) and their reaction recorded. Participants were then shown the product profiles – each participant seeing them in a different order to mitigate order effects. The orders were generated randomly. Measurements are then taken of participants overall reaction to each profile. Conjoint analysis does not require survey participants to indicate the importance or impact of individual attributes and levels; instead estimates of the impact (part-worths) are derived statistically.
The conjoint analysis model
Once the survey completed, a ‘part-worth’ was estimated for each attribute level (for each survey participant). This indicates the average impact the level had on purchase intention ratings across the profiles shown to the survey participant. The total utility of a specific product profile in the simulator is simply the sum of these part-worths and intercept.
A decision rule was then used to translate the utilities for all of the competitors into share estimates at the aggregate level. This model also incorporated external effects (usually ignored in conjoint analysis studies) in the form of perceived product availability in store. This was accounted for at the individual level to improve accuracy.
Before deciding on the final approach, a variety of alternatives were tested, including: different decision rules, approaches to negating the IIA assumption (by accounting for differential substitutability), rescaling to emulate choices made in choice-holdouts as well as comparisons to measured market share. It was decided not to force alignment with measured market share using calibration factors, as is common practice, since the model was near enough to measured share and it is preferable to explicitly incorporate external effects.
It should be noted that market share estimates should be treated as ‘ballpark’ estimates. While the aim was to align with measured share in sample, this does not necessarily align with actual volume markets share as no attempt was made to apply sample weights to correct for demographic representation (see appendix B) nor were calibrations conducted to force alignment with scanner data or diary panel data.
Figure 2 Schematic of the conjoint analysis model implemented
Results
Preference for ‘cow’ versus ‘cow-free’ milk
The chart below shows the percentage of survey participants preferring each type of milk. This is derived from each participants calculated part-worth for each technology. Cow milk has a narrow majority of 52.9% – all else held equal. However, as the scenarios below show, this can be compensated for by price.
Figure 3 Percentage of participants part-worth is largest for
Price insights
An example of the types of price analysis that can be done.
The simulator suggests that different brands and milk technologies yield different share and revenue responses to price increases. Take Kinder Green (Cow-Free Milk) for instance. While share diminishes, revenue does not over most of the price range tested – even increasing slightly in the middle of the range. Contrast that with Dewfresh, which exhibits a decline in both share and revenue as price increases. Note: the below is fixed in the baseline scenario, while price is allowed to vary.
Figure 4 Dewfresh – price vs share and revenue*
Figure 5 Kinder Green (cow-free) price vs share and revenue*
· Per 1,000 refers to per 1,000 buyers of the UHT milk category
· Revenue indicator = share proportion x 1,000 x average litres purchased per annum per buyer x price per litre
New entrant scenario simulations
The key question guiding the selection of the new entrant scenarios was: “Would it be possible for a biotech entrant without any existing dairy herd to enter the market and achieve a market share that’s likely to be sufficiently profitable?” For this exercise I have defined a sufficiently profitable market share for this industry as 10%.
The new entrant, Kinder Green, is shown in the first column in most of the scenarios.
As a cautionary note, the scenarios assume that cow-free milk delivers on its promise and that the actual product doesn’t fail to live up to expectations. See the appendix A to see how cow-free milk was defined.
Baseline scenario
The baseline scenario mirrors the actual market in early 2022, and acts as a reference point for the exploration of the alternative scenarios to follow. In this scenarios Kinder Green is switched off as it does not exist yet on the market.
Figure 6 Baseline
Scenario 1: new cow-free brand (Kinder Green) enters at a 10% price premium
In this scenario we answer the question, will the new brand be able to enter the market at a 10% price premium to Clover, with market support at similar levels to the weakest brand in the simulator (Dewfresh) and still be able to achieve its market share goal?
As can be seen below, the projected market share is far below the 10% share goal.
Figure 7 Scenario 1
Scenario 2: enter at 10% premium, with bioplastic cap and recyclable packaging
Would switching to a bioplastic lid and recyclable packaging bring the share to at least 10%? As can be seen, while there is a small increase in share, this is still far off the mark.
Figure 8 Scenario 2 
Figure 9 Cap part-worths
Scenario 3: discount to reach target share – with bioplastic cap and recyclable packaging
In addition to the bioplastic cap and recyclable packaging, would offering a discount help? Various discounts were applied until finally reaching the lowest tested level of 9.99. Even so, the projected share of 8.97% was short of the target of 10%.
Figure 10 Scenario 3
Scenario 4: discount to needed, marketing support to Clover level - with bioplastic cap and recyclable packaging
In the 4th scenario, marketing support levels are increased to match those of market leader, Clover, while the bioplastic cap and recyclable packaging is maintained. As the simulation shows, no discount is needed, in fact price can be set at a premium relative to Clover (17.65) to exceed the 10% target.
Figure 11 Scenario 4
Incumbent scenario simulations – replacement or line extension
Scenario 5: No new entrant. Incumbent brand Dewfresh replaces cow with cow-free milk.
What about a situation where there is no new entrant, but an existing brand switches to cow-free milk?
In the baseline scenario Dewfresh is priced at 14.74 and has a ballpark market share of 6.22%. After switching to ‘cow-free (from cells)’, its share is projected to drop in the simulation to 5.34%.
To recover lost share, the price only needs to decrease by 0.24 to 14.5 to push share back to just over parity (6.22%) to 6.5%. This suggests that cost of sales would need to diminish by at least 0.24 to make cow-free (from cells) a viable proposition for this brand.
Figure 12 Scenario 5
Scenario 6: No new entrant. Incumbent brand (Dewfresh) adds cow-free milk to product line
In this scenario, incumbent brand (Dewfresh) keeps cow milk and adds cow-free as an additional option, at the same price (see the first column). Summing the simulator projections below, the combined share equals 8.67%, exceeding Dewfresh’s original simulated share of 6.22%. This is likely due to cow-free milk appealing to a different segment of consumers, some of whom may previously have rejected Dewfresh.
Figure 13 Scenario 6
In conclusion
Discussion of the business implications and business models that will be viable in future are beyond the scope of this article. But hopefully this will stimulate some thought and interest in planning for the entrance of a potentially disruptive technology onto the SA milk market.
In part two, brand equity, pricing and further scenarios will be explored.
Ready to launch a conjoint analysis survey of your own? Acentric is offering the new conjoint analysis survey module at a promotional price. Try it now before the offer expires!
Appendix A: cow-free milk concept
‘Cow-free’ milk is now possible. Various technologies have either been
developed or are being developed to make milk that is identical to cow
milk, except no cows are involved in producing the milk. It tastes the
same as cow milk.
Advantages include kindness to animals, environmental advantages
(far less water and land use), possible health advantages and
potentially lower production costs.
How it’s made – the two alternative technologies:
Cow-free milk (flora-based): milk produced by combining milk
protein from genetically modified flora (e.g. fungus such as
mushrooms or yeast which include cow DNA to produce milk protein)
with other ingredients. It doesn't contain lactose, hormones, or
cholesterol. Cow-free milk (cell-based): is milk produced by cells in a
bioreactor. These are the same cells cows use to make milk, so this is
actual milk not just a milk protein. These cells are grown from stem
cells.
Appendix B: technical note – sample representation
No attempt was made to apply sample weights to correct for demographic representation. Owing to the exploratory nature of this research, only a small sample was used, which would have meant that sample weights would have inflated standard errors to unacceptable levels.
Footnote:
The ‘cow-free (from flora)’ option was not evaluated in the scenarios included in this article, suffice it to say that the flora option produces only slightly more negative reactions, and so it was not considered worthwhile running separate scenarios.