Deriving a Formula to Estimate the Upside of Personalization

← Back to all posts

In the previous entry in this series (Part 3a here, with Part 1 here and Part 2 here), we explored some general thoughts on how one might derive a formula that estimates the upside of AI-driven personalization, considered how it may be impacted by way in which companies can achieve growth, and suggested some ways that AI-driven personalization might impact each of those approaches.  We ended by describing the impact of personalization as a function of a number of factors, including product usage, brand selection, and sophistication of personalization capabilities.

So to go one level deeper: can we develop a working formula that captures all these factors?  A general approach would be something like this:

Let’s attempt to build on this and see if we can develop a couple of workable formulas based on purchase frequency, and from the perspective of an individual consumer and a brand.

• Upside of personalization from an individual consumer for frequently purchased items.

a: Formula:

b. Example 1: soft drink brand, $100 baseline spend, 70% concentration, 3% growth in frequency, 20% likelihood to switch, 10% possible growth in consumption, 5% spend on expanded repertoire

Upside = ( (100 * .70) * (1+ 3% + (20%*(1-70%)) ) * (1 + 10%) + (100 5%) – (100 * .70)) /

(100 * .70) =  (83.98 – 70) / 70 = 19.9% potential upside from an individual consumer who is open to switching and has headroom to expand consumption

c. Example 2: toothpaste, $60 baseline spend, 90% concentration, 0% added for freq, 10% likelihood to switch, 0% growth in consumption, 3% spend on expanded repertoire

Upside = ( ( (60 * .9) * (1+ 0% + (10% * (1 - 90%)) ) * (1 + 0%) + (60 * 3%) )  - (60 * .9) ) /

(60 * .9) =  (56.394 – 54) / 54 = 4.4% potential upside from an individual consumer with limited likelihood to switch, static consumption

• Upside of personalization for a brand that is frequently purchased, where personalization sophistication score ranges from 0 to 5.  Also can be applied to larger items that are less frequently purchased.

a: Formula:

b. Example 1: fast food restaurant, with -0.5% category growth, 70% openness to switching, leader personalization score = 4 out of 5, laggard = 1

i.     Upside = 0.01 * (1+ .7) * 4 =  6.8% (leader)

ii.     Upside = 0.01 * (1 + .7) * 1 = 1.7% (laggard)

iii.     Upside = (3 * -0.005) * (1 + .7) * 1 = -2.56% (no personalization)

c. Example 2: health and personal care, 9.1% YoY growth (2023 – per https://www.oberlo.com/statistics/fastest-growing-consumer-products), 30% openness to switching, leader personalization score = 3 out of 5, laggard = 1

i.     Upside = 0.091 * (1 + .3) * (1 + 3/5) = 18.9% (leader)

ii.     Upside = 0.091 * (1 + .3) * (1 + 1/5) = 14.2% (laggard)

iii.     Upside = (1/5 * 0.091) * (1 + .3) * -1 = 2.4%

(no personalization)

d. Example 3: washing machine, 2.34% YoY growth (https://www.statista.com/outlook/cmo/household-appliances/major-appliances/washing-machines/united-states), 40% openness to switching, leader personalization score = 4, laggard = 1

i.     Update = 0.0234 * (1+.4) * (1 + 4/5) = 5.85% (leader)

ii.     Update = 0.0234 * (1+.4) * (1 + 1/5) = 3.9% (laggard)

iii.     Update = (1/5 * .0234) * (1 + .4) * -1 = -0.66%

(no personalization)

Now, I’m not saying this is the solution, only that the outputs are reasonable, and that suggests that coming up with a formula is possible.  And surely some academic somewhere must have solved for this already, correct?  Regardless, it has been fun to explore. Keeping in mind that this type of approach is all based on theory/hypothesis, and would require actual research and data to validate, one can still make the case that, for these examples at least, the outputs seem to be realistic and possibly could serve as a starting point for gauging the impact in a particular category. Alternatively, the particulars of the approach, especially the input values, can easily be varied to reflect different assumptions.

Now, this does raise some additional questions:

1. Would this differ over a short-term timeframe vs. a longer period? That seems to be one of the dimensions that hasn't been adequately discussed. Would the “growth from switching” variable, for example, diminish over time as laggards catch up?
2. If you replace “personalization” with “marketing” or “targeting” or “segmentation” or “promotions” is the equation much different?

Something has been nagging me as I’ve worked through this thought process, and it’s that last question that really made me pause. Is there something unique about AI-driven personalization that makes it different than any past marketing innovations that have tried to achieve sustainable competitive advantage?

Remember, according to the frequently-cited stats from McKinsey & Company, first suggested in 2015 and never updated, personalization:

• Reduces acquisition costs by as much as 50 percent. Hard to believe that is sustainable, because if so, we should be able to increase the number of new customers by double without increasing spend.  If only it were that easy.  Who’s doing that successfully today?
• Lifts revenues 5 to 15 percent. This may be possible in high-growth categories with low barriers to switching and expandable consumption, and over a short time horizon. Again, is there a current example of someone growing that quickly, period over period?
• Improves efficiency of marketing spend by 10 to 30 percent. This could mean a lot of things, but if this means we can increase revenue 10-30% with the same marketing spend, it’s too good to be true.

The projections from these “fun” formulas as described above are much lower than McKinsey suggests, and I still find it hard to believe that it can be that easy - you just turn on personalization and revenues magically increase, and that level of growth is fair game for anyone that wants to try.  Something is still missing.  In the next and last part in this series, we’ll come closer to some clarity.

View original on LinkedIn