Does Personalisation Actually Work

Answer to the question

Does personalisation actually increase sales?

A little, where it has been measured properly. Randomised field experiments raised email open rates from about nine per cent to about eleven by adding the recipient's name. In the cleanest randomised head-to-head, a personalised travel advertisement converted worse than the generic one from the same brand.

Key takeaways

• The two trillion dollar prize traces to one unpublished consultancy index with no method, sample frame, model or rubric, and among its stated criteria are the growth and depth of a company's digital customer relationships.
• The best positive randomised evidence is real and small: a recipient's name in a subject line moved open rates from about nine per cent to about eleven, with leads up about thirty-one per cent, worth thirty-five extra leads at about one hundred dollars each in one campaign.
• In the cleanest randomised comparison, across 77,937 people, the personalised advertisement converted worse than the generic one from the same brand, and the ordering reversed only for the roughly forty per cent who had already visited a review site.
• No independent, multi-firm evaluation of what a personalisation programme returns at firm level was found, so every randomised estimate available is one campaign, one channel or one platform.

Everybody asks whether personalisation pays. At firm level that question has no measured answer. The answerable one is narrower: when a personalised version was randomised against a generic version of the same thing, which won?

Here is where we land. Personalisation works, and it works best at the cheap end, where the intervention is nothing more than the customer's own name. The expensive end, the bought segments and the dynamic creative, is where the evidence thins and where the one clean randomised head-to-head ran the wrong way. The two trillion dollar prize on the jacket is a projection rather than a measurement.

The consultancy's two trillion dollar claim

The figure sits on the jacket of Personalized, by Mark Abraham and David C. Edelman, published by Harvard Business Review Press in 2024 and copyrighted to The Boston Consulting Group. It invites readers to capture their share of the two trillion dollar personalisation prize, before any evidence is offered.

Follow it and it traces to one object, then stops. BCG's press release of 15 October 2024, announcing the book, states that over the next five years two trillion dollars in revenue will shift to companies that understand how to create personalised experiences and communications. The release names the instrument, a measurement system the firm built itself called the Personalization Index, and the projection rests on a gap of ten percentage points of annual revenue growth between the firms it classes as leaders and laggards.

That is the whole trail.

An instrument that scores a firm partly on having already grown cannot then be used to explain its growth.

No published method, no sample frame, no model, no rubric, no statement of how leaders were cut from laggards. Among the index's stated criteria are the depth, retention and growth of a company's digital customer relationships, so it scores a firm partly on having already grown, then shows that high scorers grow faster.

Our reading of the book splits in two. The architecture is good, and the learning loop, in which a company records a response and changes its next action for that individual rather than for a bucket, is worth keeping. The business case is not evidence. The book itself warns that adding up revenue from concurrent personalisation campaigns grossly overstates their impact, then reports unaggregated lifts for eleven chapters.

That architecture makes this a different object from [what a recommendation box adds](https://thewhyweproject.com/articles/how-much-recommendations-really-sell) to one page. A personalisation programme is an operating model with a budget.

What Sahni's field experiments actually earned

Navdeep Sahni and colleagues, in Marketing Science in 2018, ran randomised field experiments with three organisations, sending experimentally tailored emails to millions of people. Adding the recipient's name to the subject line raised the probability of an open by about twenty per cent. Sales leads rose by about thirty-one per cent and unsubscribes fell by about seventeen per cent.

The relative lift is the number the trade quotes; the base is what makes it mean anything, and the published abstract gives it. The open rate moved from 9.05 per cent to 10.80 per cent, leads from 0.39 per cent to 0.51 per cent, unsubscribes from 1.2 to 1.0. A secondary institutional summary from the business school of one author adds that one campaign gained thirty-five extra leads, valued by the company at about one hundred dollars each.

Three and a half thousand dollars, across a whole list.

This is the fair case for personalisation, randomised, at scale, with revenue-proximate outcomes. And nothing was inferred about the recipient at all. The intervention was their own name, the cheapest personalisation available and the opposite of a data programme costing tens of millions a year. The mediating measure is the open rate, a contaminated metric, and the lead valuation is the company's own.

Lambrecht and Tucker's losing advertisement

Anja Lambrecht and Catherine Tucker published the comparison the industry claims to run and almost never prints. A travel firm randomised 77,937 people over twenty-one days between a generic advertisement, showing a beach image and the brand logo, and a dynamic one showing a hotel the person had browsed.

The personalised version lost.

In the cleanest randomised comparison of a personalised against a generic advertisement from the same brand, the personalised version lost.

Against a daily purchase probability of two hundredths, the dynamic advertisement came in at about minus one hundredth, while retargeting in general came in at about plus nine thousandths. The authors volunteer the calibration: relative effects between advertisement types are large while absolute effects are small. Click-through averaged under one hundredth of one per cent.

The ordering reverses in one place, which is the strongest case for personalisation inside the same paper. Generic advertisements worked best for people who had not yet sought out product quality information at a review site. Dynamic retargeting won only after such a visit, which about forty per cent of these consumers made. A specific advertisement earns its place once a person has narrowed their own choice.

Their own warning about baselines matters as much as their result, and they printed it:

> "This suggests that the large gains for retargeting that many industry studies report may be because these consumers are already more likely to purchase, because they have already sought the product out."

The arithmetic underneath is 14.6 per cent of review-site visitors buying against 8.6 per cent of everyone else. Two limits travel with it, both stated by the authors. The advertisements differed in design as well as content, so the test compares the two as commonly practised rather than two identical creatives, and offline purchases were not observed.

The meta-analysts, and Tucker's reframe

The backfire thesis is weaker than the trade's anxiety, and the two largest pooled estimates say so. Martin Eisend and colleagues, in the Journal of Marketing, integrate 1,536 effect sizes from 290 studies and find personalisation leads to persuasion success, with negligible negative effects. Tien Ee Dominic Yeo and colleagues, across 53 experiments, find that perceived relevance, but not perceived intrusiveness, explained the persuasiveness of personalised advertising.

What the paradox still holds is a mechanism, and the mechanism has a precondition. Christopher Summers and colleagues show that a behaviourally targeted advertisement can work as an implied social label, because the recipient works out that an inference has been made about them, and the effect holds only when the targeting is at least moderately accurate.

Nico Neumann and colleagues measured it. Across three field tests of more than 90 third-party audiences from 19 data brokers, bought profiles improved identification of a person with one desired attribute by between zero and 77 per cent against picking at random, and by about 123 per cent once optimisation software was added.

The lower bound is zero.

Tucker's reframe on control

Catherine Tucker's 2014 study is the reframe. A social network gave users more control over their personal information midway through a randomised field experiment, changing nothing about advertiser targeting. Clicks on personalised advertisements nearly doubled, with the largest gains for those using the most unique private information. What the model returns depends less on the quantity of data than on whether the person believes they handed it over.

A trade summary of Yeo and colleagues, secondary and labelled as such, reports that personalisation helped in commercial advertising while showing no persuasive advantage in political or charity contexts, which is where this piece stops and [what can be inferred from a digital footprint](https://thewhyweproject.com/articles/what-psychographic-targeting-can-do) begins. Nor does any of it price [what you are actually paying for a free app](https://thewhyweproject.com/articles/what-you-actually-pay-for-free-apps).

Then the finding our searching returned. We looked for an independent, multi-firm evaluation of what a personalisation programme returns at firm level and found none. Every randomised estimate above is one campaign, one channel or one platform. The firm-level return the two trillion dollar figure projects has not been independently measured by anybody.

Three checks separate a personalisation claim that can be inspected from one that cannot. The first is what the comparison was, a generic version of the same thing or nothing at all. The second is what the base rate was, since an eighty-three per cent lift on four tenths of one per cent and on twelve per cent are different objects. The third is who would have bought anyway, since the people a retargeting system can see already came looking.

Lambrecht and Tucker answered all three about their own study, in print, and the answer was that their personalised advertisement underperformed the generic one, which is why a twenty-one day test on 77,937 people, reporting a base rate and a selection warning alongside a result that embarrassed the practice it examined, carries more weight here than an unpublished index scoring firms partly on the growth it was later used to explain.

Common questions

Does personalisation actually increase sales?

A little, where it has been measured properly. Randomised field experiments reported by Sahni and colleagues in 2018 raised email open rates from 9.05 per cent to 10.80 per cent. In a randomised test reported by Lambrecht and Tucker in 2013, the personalised advertisement converted worse than the generic one from the same brand.

Where does the two trillion dollar figure come from?

From a consultancy's own Personalization Index, announced in the press release for the book that uses it, with no published method, sample frame, model or rubric. Among its stated criteria are the depth and growth of a company's digital customer relationships, so it scores firms partly on having already grown.

Is personalised advertising creepy enough to backfire?

Less often than the phrase suggests. Across 53 experiments, Yeo and colleagues traced persuasion to perceived relevance, with perceived intrusiveness carrying no explanatory weight, and Eisend and colleagues found negligible negative effects across 1,536 effect sizes. Accuracy is the sharper issue, since bought audiences beat random selection by between zero and 77 per cent.

Do privacy controls hurt personalisation?

The best available evidence runs the other way. Tucker's 2014 study found that when a social network gave users more control over their personal information and changed nothing about how advertisers targeted, clicks on personalised advertisements nearly doubled, with the largest gains for those using the most unique private information.

Sources

Abraham, M., & Edelman, D. C. (2024). Personalized: Customer Strategy in the Age of AI. Boston: Harvard Business Review Press

Boston Consulting Group (2024). 'Capturing the $2 Trillion Personalization Opportunity with AI.' Press release, 15 October 2024

Sahni, N. S., Wheeler, S. C., & Chintagunta, P. (2018). 'Personalization in Email Marketing: The Role of Noninformative Advertising Content.' Marketing Science, 37(2), 236-258. doi:10.1287/mksc.2017.1066

Finkel, E. (2018). 'Open This, [NAME]! Why Personalized Email Subject Lines Work.' Chicago Booth Review, 12 October 2018

Lambrecht, A., & Tucker, C. (2013). 'When Does Retargeting Work? Information Specificity in Online Advertising.' Journal of Marketing Research, 50(5), 561-576

Neumann, N., Tucker, C. E., & Whitfield, T. (2019). 'Frontiers: How Effective Is Third-Party Consumer Profiling? Evidence from Field Studies.' Marketing Science, 38(6), 918-926. doi:10.1287/mksc.2019.1188

Eisend, M., Niewiadomska, D., & van Noort, G. (2026). 'Personalization in Marketing Communication: A Meta-Analysis.' Journal of Marketing. doi:10.1177/00222429261460888

Yeo, T. E. D., Chu, T. H., & Li, A. Q. (2025). 'How persuasive is personalized advertising? A meta-analytic review of experimental evidence of the effects of personalization on ad effectiveness.' Journal of Advertising Research, 65(4), 616-631. doi:10.1080/00218499.2025.2467763

Randhawa, N. (2025). 'Personalised ads slightly more effective, says study.' MediaCat UK, 16 April 2025

Summers, C. A., Smith, R. W., & Reczek, R. W. (2016). 'An Audience of One: Behaviorally Targeted Ads as Implied Social Labels.' Journal of Consumer Research, 43(1), 156-178. doi:10.1093/jcr/ucw012

Tucker, C. E. (2014). 'Social Networks, Personalized Advertising, and Privacy Controls.' Journal of Marketing Research, 51(5), 546-562. doi:10.1509/jmr.10.0355

Related concepts

• Base-Rate Neglect
• Borrowed Certainty

People

• Nico Neumann