The Literature Review as a Business Asset: how to find it, read it honestly, referee the fights, and why it beats commissioning another survey

Here is a slightly embarrassing fact about most business questions. Does hybrid work hurt productivity? Do engaged employees make more money? Does a price cut actually move volume in your category? Somebody with a PhD, a grant, and a lot of patience has probably studied it already, often five or six times. The work sits in journals with titles like “Heterogeneous Treatment Effects in Quasi-Experimental Panels,” which is about as inviting as a tax form. Meanwhile companies spend $20,000 to $50,000 asking 400 strangers the same question. What follows is a Q&A on turning that buried research into something you can use today.

Q1. What is a literature review, in business terms?

1. It is a structured read of what published research already says about one question, boiled down to a verdict, a confidence level, and the conditions under which the verdict holds.
2. Think of it as buying the results of dozens of studies you did not have to fund. Gallup’s employee engagement meta-analysis pools 736 studies, 347 organizations, 183,806 business units and 3.35 million employees across 90 countries (Gallup Q12 Meta-Analysis, 11th ed.). No single company survey gets near that sample.
3. The payoff is concrete: in that dataset, top-quartile engagement units showed 23% higher profitability and 78% lower absenteeism than bottom-quartile units (Gallup).

Q2. How much research is actually out there?

1. About 5.14 million scholarly articles were published in 2024, up 2.06% from 2023 and roughly 23% over five years (PublishingState, citing publication trackers).
2. Google Scholar indexes an estimated 389 million records, several times the roughly 70 to 90 million in Scopus or the Web of Science Core Collection (Gusenbauer, Scientometrics, 2019).
3. The output is also shifting geographically. China now authors 28% of global articles and the US 11% (STM OA Dashboard, 2025), so a search limited to American journals misses a lot of the evidence.

Q3. Isn’t all of it locked behind paywalls?

1. Less than you think. A large Unpaywall study found 45% of articles published in 2015 were free to read somewhere legal (Piwowar et al., PeerJ, 2018). Unpaywall now says its users read 52% of papers free, from a pool of about 49 million free articles.
2. Gold open access alone covered 40% of articles, reviews and conference papers published in 2024 (STM OA Dashboard).
3. For the rest, pay-per-view averaged $37.72 an article, ranging from $15 to $80 (Lemley & Li, 2015). Duke Libraries found 59 of the 100 most cited papers ever are paywalled, at about $33.41 each. Before paying, email the author. Most researchers are delighted that anyone wants to read their paper and will send a PDF within a day.

Q4. Where do I look, and in what order?

1. Start by hunting for an existing review or meta-analysis. Add “systematic review” or “meta-analysis” to your search. If one exists, someone else has done 80% of your work.
2. Translate your question into their vocabulary. “Do our salespeople quit because of the commute?” becomes “commute time AND voluntary turnover.” Then snowball: read the reference list backward and use “cited by” to go forward.

Figure 1 (left): shares of the literature free to read. Figure 2 (right): typical Pew telephone survey response rates, 1997 to 2018.

Source	Size / coverage	Best for
Google Scholar	~389M records (Gusenbauer 2019)	Broad first sweep; “cited by” snowballing
OpenAlex / Semantic Scholar	Free, open indexes	Bulk searching and exporting results
SSRN and NBER	Working papers	Economics, finance, management, often years before journals
PubMed	Biomedical literature	Health, benefits, workplace wellness
Unpaywall (browser add-on)	~49M free articles	Finding the legal free copy of a paywalled paper

Table 1: A starter kit for finding the research. All five are free to use.

Q5. Why does this beat running a new survey?

1. Price. A custom quantitative or qualitative project typically runs $20,000 to $50,000 (Farnsworth Group, 2026), and a US online survey of 400 to 1,000 completes runs $10,000 to $30,000 (Merren, 2026). Traditional firms take 4 to 12 weeks (Preuve, 2026).
2. Quality. Pew’s typical phone survey response rate fell from 36% in 1997 to 6% in 2018 (Pew Research Center, 2019). Your survey hears from the few who pick up. The published literature often draws on randomized trials, administrative records and panels you could not afford to build.
3. You are already paying for search time anyway. McKinsey estimated knowledge workers spend about 19% of the workweek searching for and gathering information (McKinsey Global Institute, 2012). Pointing that time at peer-reviewed evidence is a better use of it.
4. A rough comparison: two weeks of a fully loaded analyst (about $150,000 to $175,000 a year, per Preuve) plus 25 articles at $37.72 comes to roughly $7,300. That is my own back-of-envelope estimate, and it assumes a focused business review rather than a full academic one.

Approach	Typical cost	Time	What you get
Rapid desk literature review	~$7K (estimate)	1 to 3 weeks	Pooled findings from many studies, with caveats
US online survey (400 to 1,000)	$10K to $30K	4 to 8 weeks	Fresh data on your own customers
Custom mixed-method study	$20K to $65K	4 to 12 weeks	Tailored answers, one sample
Full academic systematic review	~$141K	~67 weeks	Gold-standard rigor (overkill for most firms)

Table 2: Sources: Merren 2026; Farnsworth 2026; Michelson & Reuter 2019; Borah et al., BMJ Open 2017.

5. A caveat worth stating plainly: a full systematic review in medicine costs about $141,194 (Michelson & Reuter, 2019) and averages 67.3 weeks with five authors (Borah et al., BMJ Open, 2017). You rarely need that. You need to borrow its discipline, write down your search, set inclusion rules before reading, and log what you tossed out.

Figure 3 (left): cost per answer by method. Figure 4 (right): publication bias in antidepressant trials.

Q6. What does synthesizing “honestly” mean?

1. Weigh studies by design and size, then count votes. One randomized trial with 1,612 people outranks ten opinion surveys of 200.
2. Watch for the file drawer. When Turner and colleagues compared FDA records to journals for 12 antidepressants, 94% of published trials looked positive while the FDA judged only 51% positive, and published effect sizes ran 32% larger (NEJM, 2008). Negative results tend to stay home.
3. Discount single studies. The Open Science Collaboration reran 100 psychology experiments: 97% of originals were significant, only 36% of replications were, and average effects shrank by half, from r = .40 to r = .20 (Science, 2015). A lone striking study is a lead, and a cluster of consistent ones is evidence.

Q7. What do I do when the studies disagree?

Treat the disagreement as information. Remote work is a perfect example, because three careful studies landing in three different places look contradictory until you read what each one measured.

Study	Setting	Result	Why it differs
Bloom et al., QJE 2015	Ctrip call center, full WFH, RCT	+13% performance	Simple, measurable solo work; volunteers
Gibbs et al., JPE Micro 2023	10,000+ IT staff at HCL, forced pandemic WFH	Productivity -8% to -19%	Team coordination costs; kids at home; no choice
Bloom et al., Nature 2024	1,612 Trip.com grads, 2 days home, RCT	No change in output; quits down 33%	Hybrid, not full remote; planned rollout

Table 3: Resolving conflicting findings by looking at population, treatment and measurement.

1. Check the three usual suspects: who was studied, what exactly the “treatment” was, and how the outcome was measured. Here they explain almost everything.
2. Then write a conditional verdict: hybrid schedules for knowledge workers appear roughly neutral on output and good for retention; fully remote, team-heavy work carries real coordination costs. Trip.com acted on exactly this, since each quit cost about $20,000 in recruiting and training (Bloom et al., 2024).

Figure 5 (left): three rigorous remote-work studies. Figure 6 (right): replication results from the Open Science Collaboration.

Q8. Can AI tools do the heavy lifting now?

1. Increasingly, yes, for the grunt work. One AI pipeline, Otto-SR, hit 96.7% screening sensitivity and 93.1% data-extraction accuracy across 32,357 citations, reproducing 12 Cochrane reviews in two days (as reported in the LUMEN preprint, arXiv 2026).
2. Even before AI, cheaper screening worked. Crowdworkers on Mechanical Turk caught 95% to 99% of relevant citations at up to 88% lower cost than trained experts (Mortensen et al., via PMC).
3. The judgment still belongs to you. Language models can invent citations that look perfect, so open every paper you cite and confirm the number is really in it.

Q9. What should the final deliverable look like?

1. One page. The question, the verdict in a sentence, a confidence grade (strong, moderate, thin), the three best studies with links, where the evidence disagrees, and what it would take to change your mind.
2. Keep the search log. Next year, when the question comes back wearing a new hat, you update the file in an afternoon. That reusable file is the asset, and the global insights industry, now worth more than $150 billion a year (ESOMAR, 2025), has little incentive to tell you about it.

Sources

1. Gallup, Q12 Meta-Analysis, 11th ed.: https://www.gallup.com/workplace/321725/gallup-q12-meta-analysis-report.aspx

2. PublishingState, 5.14 million articles in 2024: https://publishingstate.com/more-than-5-million-scholarly-articles-were-published-in-2024-and-here-are-more-interesting-facts/2025/

3. Gusenbauer, Scientometrics 2019: https://doi.org/10.1007/s11192-018-2958-5

4. STM Association, OA Dashboard 2025: https://stm-assoc.org/oa-dashboard/oa-dashboard-2/uptake-of-open-access/

5. Piwowar et al., The State of OA, PeerJ 2018: https://peerj.com/articles/4375/

6. Unpaywall: https://unpaywall.org/

7. Lemley & Li 2015, reviewed in EBLIP: https://doi.org/10.18438/b8132v

8. Duke Libraries, paywall factoids: https://sites.duke.edu/library101_instructors/category/scholarly-communications

9. Pew Research Center, phone response rates 2019: https://www.pewresearch.org/short-reads/2019/02/27/response-rates-in-telephone-surveys-have-resumed-their-decline/

10. Farnsworth Group, market research cost 2026: https://www.thefarnsworthgroup.com/blog/market-research-cost

11. Merren, market research cost 2026: https://merren.io/blogs/how-much-does-market-research-cost/

12. Preuve, market research cost breakdown 2026: https://preuve.ai/blog/how-much-does-market-research-cost

13. McKinsey Global Institute, The Social Economy 2012: https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-social-economy

14. Michelson & Reuter, Contemp. Clin. Trials Comm. 2019: https://www.sciencedirect.com/science/article/pii/S2451865419302054

15. Borah et al., BMJ Open 2017: https://pubmed.ncbi.nlm.nih.gov/28242767/

16. Turner et al., NEJM 2008: https://www.nejm.org/doi/full/10.1056/NEJMsa065779

17. Open Science Collaboration, Science 2015: https://www.science.org/doi/10.1126/science.aac4716

18. Bloom, Liang, Roberts & Ying, QJE 2015: https://doi.org/10.1093/qje/qju032

19. Gibbs, Mengel & Siemroth, JPE Micro 2023: https://www.journals.uchicago.edu/doi/full/10.1086/721803

20. Bloom, Han & Liang, Nature 2024: https://www.nature.com/articles/s41586-024-07500-2

21. ESOMAR data via Market Xcel 2026: https://www.market-xcel.com/us/blogs/market-research-cost-usa

22. LUMEN preprint (cites Otto-SR results), arXiv 2026: https://arxiv.org/pdf/2606.28362

23. Mortensen et al., crowdsourced citation screening: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5589498/