CoChat

Computer Estimation of Customer Similarity With Facebook Lookalikes: Advantages and Disadvantages of Hyper-Targeting

Open access
Semeradova, Tereza, Weinlich, Petr · 2019 · IEEE Access

The DOI matches its Crossref record; 92 of the 98 references it cites were found in a registry or an index; 1 of the 15 sources its claims cite were checked for support.

Abstract

The advertising systems and the algorithms they use are constantly evolving and expanding the possibilities for reaching potential customers. Hyper-targeting (also called microtargeting) is the use of detailed customer data and marketing automation to deliver highly targeted and personalized messages across a large number of channels. These campaigns are designed to appeal to specific people or small groups of customers. By using the ability to process large amounts of data through innovations, such as predictive analytics, marketers can gain a deeper understanding of their audiences, focusing on specific accounts and not on the entire segments. This reportedly allows B2B brands to target customers directly and provide unique personal and highly relevant experiences. However, the scientific evidence to support this claim is missing. Some previous studies even suggest a negative impact of highly personalized advertising content on user reactiveness and purchase behavior. In this article, we test the effects of different levels of personalized advertisements using the advanced campaign targeting tool called Facebook Lookalike Audiences. Facebook Lookalike Audiences works on the basis of the estimation of customer similarity based on the characteristics of a custom audience, as defined by the advertiser. We examine the performance of various targeting settings using data from 840 Facebook ads with different personalization levels. These advertisements are compared in terms of reach, number of reactions, frequency of impressions, number of clicks, average time spent on a website, number of viewed pages, number of conversions, and profitability. We believe that the findings presented in this article help clarify the factors that influence user reactiveness toward personalized online advertising using evidence from actual Facebook ad sets.

Overview

CoChat Summary

## TL;DR

The paper tests whether Facebook Lookalike Audiences and varying personalization levels improve ad performance, using 840 real Facebook ads from an e-shop. It finds that moderate personalization performed best, while the most highly personalized ads were not always the most effective and could trigger less favorable reactions.

## Key findings

  • The study analyzed 840 Facebook ads and found that hyper-targeting has limits; the authors report that it may lead to negative attitudes toward the advertiser.
  • For customer segmentation, Facebook performance data processed by Facebook’s internal algorithm performed best, reaching 11.68 EUR average profitability.
  • Across segmentation methods, newsletter subscribers processed by Apriori produced 10.84 EUR average profitability, and average customer value processed by Apriori produced 10.53 EUR; the lowest averages were web traffic (9.95 EUR) and clustering algorithm (10.04 EUR).
  • Personalized ads were most effective at the conversion stage (12.00 EUR), followed by consideration (10.23 EUR), and were least effective for awareness (8.84 EUR).
  • The paper reports that placement did not influence the impact of personalized ads: Facebook newsfeed, Marketplace, and Messenger were similar, while Instagram newsfeed was lower at 9.98 EUR.
  • By gender, the average profitability was 9.76 EUR for females and 9.91 EUR for males; the authors say males reacted more positively at several personalization levels, while females were more sensitive to highly personalized content.

## Methods

  • The authors ran an online field experiment using the advertising accounts of a Czech e-shop selling home accessories and kitchen equipment, with real-life campaigns.
  • They created 840 Facebook ads using Facebook Custom Audiences (CA) and Lookalike Audiences (LA), with 10 levels of personalization, 3 conversion stages (awareness, consideration, conversion), and 4 placements (Facebook newsfeed, Messenger, Facebook marketplace, Instagram newsfeed).
  • They used two segmentation methods (Apriori and clustering) and four data sources (average customer value, newsletter subscription, web traffic, internal Facebook data), then analyzed the results in SPSS Statistics v26 using average profitability as the main outcome.

## Limitations

  • The paper explicitly says the experiment “may be considered as a pilot study” because the authors could not find comparable prior research using actual Facebook campaign data with Lookalike Audiences.
  • It also states that the results cannot be compared with other findings on personalized advertising because similar research was not available in the scientific literature.
  • Evident from the text: the study is based on one e-store and Facebook ads only, so generalization beyond this setting is not directly supported by the paper.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/access.2019.2948401

Cites

92 of 94 references verified0 citation errors · 2 not checkable

Statistics

Nothing to recompute.

Figures

FIGURE 1 Business intelligence and analytics: research directions.
FIGURE 1 Business intelligence and analytics: research directions.CC-BY · From the PDF
FIGURE 2 Data inputs and customer similarity estimation.
FIGURE 2 Data inputs and customer similarity estimation.CC-BY · From the PDF
Image 3
CC-BY · From the PDF
Image 4
CC-BY · From the PDF

Verification

Claims the paper makes

0 Valid · 0 Not valid · 8 Not checked · 34 Own results and interpretations

24 more claims

Create a free account to see all 42 claims and to verify the 1 reference still open.

See all claims — free account
  • The quote only partly states this claim. 1
  • Probably not a substantial claim. 7
  • It may state two separate claims as one. 7

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Research works printed but not linked: 14

  1. rosemann, 2012, Leveraging social network data for analytical CRM strategies—The introduction of social BI, Proc ECIS, 1
  2. wiener, 1956, Cybernétique et Société
  3. 2019, Understand Attribution Results
  4. huang, 2015, The improvement of customer satisfaction by adopting CRM technology, JTB J Technol Bus, 120, 1
  5. 2019, About Advertising Objectives on Facebook
  6. 2019, Facebook Active Advertisers 2019
  7. 2019, About Lookalike Audiences for Facebook ADS
  8. shmueli, 2017, Data Mining for Business Analytics Concepts Techniques and Applications in R
  9. ládrová, 2018, New Facebook Algorithm How Does It Affect Sponsored Posts? IDIMT-2018 Strategic Modeling in Management Economy and Society, 225
  10. 2019, Facebook Advertising Targeting Options
  11. swan, 2019, The New Facebook Algorithm |5 Tips to Master It In 2019
  12. 2019, Custom Audiences
  13. 2019, Create a Lookalike Audience |Facebook Ads Help Centre
  14. 2019, Create A Custom Audience From A Customer List |Facebook Ads Help Centre

Cited by

See all 39

Metadata

Authors
Semeradova, T., & Weinlich, P.
Affiliations
Technical University of Liberec (CZ)
Venue
IEEE Access
Published
2019
Volume
7
Pages
153365-153377
ISSN
2169-3536 (ISSN-L)
OpenAlex
W2981405027
Cited by
39
References
98
Citations by year
2026: 5 · 2025: 7 · 2024: 4 · 2023: 10 · 2022: 4 · 2021: 5 · 2020: 4
FWCI
6.55
Type
journal article
Language
en
Open access
Locations
  • IEEE Access (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Technical University of Liberec through competitiveness of the company in the digital environment (SGS-2019-1068)
  • Technology Agency of the Czech Republic under the Program of Applied Research ZETA through developing the skills necessary for the digital business transformation (TJ02000206)
  • Technology Agency of the Czech Republic under the Program of Applied Research ZETA through developing skills necessary for digital business transformation
Topics
  • Social Sciences › Business, Management and Accounting › Marketing › Digital Marketing and Social Media
  • Social Sciences › Business, Management and Accounting › Marketing › Marketing and Advertising Strategies
  • Social Sciences › Business, Management and Accounting › Marketing › Consumer Market Behavior and Pricing
Keywords
microtargeting, personalized advertising, customer similarity, customer data, predictive analytics, B2B marketing, conversion rate

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.