MEMORANDUM OPINIONS AND ORDER
ASHMAN, United States Magistrate Judge.
Plaintiffs have filed a Motion for Costs for Electronic Discovery.1 They argue that Defendant CB Richard Ellis, Inc. (“CBRE”) should bear the costs of searching CBRE’s email backup tapes to find documents containing pornographic terms and images, as well as documents relating to CBRE’s workplace environment generally, due to the large number of these types of documents that have been found in a controlled sampling. CBRE responds that only a small fraction of the emails that have been found contain arguably relevant material and that it should not be forced to pay for the search or production. For the following reasons the Court grants Plaintiffs’ motion in part and denies it in part.
I. Background
Plaintiffs filed this class action complaint against CBRE alleging a nationwide pattern and practice of sexual harassment at the [570]*570CBRE offices.2 As evidence of the hostile work environment prevalent at the offices of CBRE, Plaintiffs seek discovery of pornographic material that they claim was distributed electronically (i.e., via e-mail) and displayed on computers throughout the offices.
CBRE initially produced 94 monthly e-mail backup tapes from 11 offices.3 The backup tapes consist of the e-mails that existed on a given server at the time the backup is made. They are not a complete depiction of every email that existed on the CBRE system during a month. Kroll Ontrack, an electronic discovery service, was retained by Plaintiffs to restore and extract the user e-mails from the tapes, perform searches for keywords and file attachment types, and load the results of the searches onto Kroll’s ElectronicDataViewer (“EDV”), an Internet-based system, for review.
Kroll was instructed to process one monthly tape from each of three offices.4 It correctly searched the August 1999 tape for the Chicago office, the June 1999 tape for the St. Louis office, and inadvertently searched the June 1999 tape for the Columbus office, instead of the Oak Brook office. Kroll recovered over two hundred thousand documents from the tapes, referred to as the “processing set.” Next, Kroll searched the documents for a 92 pornographic term and six disciplinary term search list using a processing engine which is able to search in the text of the documents and in metadata (embedded data in an electronic document). The processing engine can find the terms at the beginning, middle or end of a word or series of symbols. Kroll searched the documents and provided the resulting review set to Plaintiffs’ counsel, who noted that spam had not been removed from the review set. For purposes of this motion, the parties have defined “spam” as anything received from outside the company, or sent solely to someone outside the company from inside the company so that the review set would contain only e-mails transmitted from a CBRE employee to at least one CBRE employee.
Kroll processed the documents again to remove spam from the review set. It also removed documents that did not contain a search term but that would otherwise be counted as a hit due to family cascading — a phenomenon whereby a document related to a document containing a hit are counted as two separate hits even if the related document does not contain a search term. For example, an e-mail that contains a search term with an attachment that does not contain a search term was counted as two hits even if the attachment did not contain a hit. After accounting for spam and family cascading, Kroll provided the parties with the new review set which contained 17,375 documents. At this point, Kroll also gave the parties a new estimate of costs to process the tapes from the 11 offices. Although the original estimates of the project ranged from $46,000 to $61,000, due to the large number of documents containing the pornographic and disciplinary search terms, Kroll revised its cost estimate and advised the parties it could cost up to $249,000 to perform the work.
In the meantime, before the parties had the opportunity to review the most recent processing set, the Court ordered the parties to each choose four terms from the list of search terms developed by Plaintiffs. Plaintiffs instructed Kroll to search for the eight terms and produce all of the documents containing search terms.5 Kroll was also instructed to use the process of de-duplication, the process whereby documents which appear in a user’s mailbox on multiple days are not counted as multiple hits. For example, if the same e-mail appeared in an inbox over a period of several months, only one copy of [571]*571the document would be produced. After deduplication, Kroll found 8,660 documents by searching for the 8 search terms, and by accounting for spam and family-cascading.
At this point, we note that discussing documents in terms of numbers is somewhat inexact. For example, an e-mail containing a search term that exists in a user’s outbox, and also exists in another user’s inbox, counts as two hits, even though it is really one document. A document containing a search term that is sent from one user to another, and returned under the “reply with history” option available on CBRE’s e-mail system counts as two hits. But, because of de-duplication, an e-mail that is present multiple times in one user’s mailbox is not counted multiple times. So although talking about documents in terms of numbers is not entirely accurate, the search system was designed to get an idea of how frequently the documents containing search terms were being passed around by CBRE users within or between the offices. Because spam was eliminated, it means the picture does not present an entirely accurate view of any other pornographic e-mails that maybe have been available on the CBRE e-mail system, or how often users are opening such documents in view of other people. The numbers also do not reflect e-mails that were not captured on backup tapes.
The parties are able to view the documents on Kroll’s EDV, a software program designed for viewing electronic documents such as these. One problem with the EDV, however, is that the search engine is not as advanced as the initial processing search engine that was used to find the 8,660 documents. The EDV search engine can find words with root extenders (e.g. “kiss!” finds kiss, kissing, kissed), but unlike the original search engine, the EDV search engine only finds search terms located at the beginning of words (so “moral” is a hit on the original search, but not through the EDV). The parties also discovered that family cascading was still a problem. Therefore, in the end, the parties reviewed approximately 1/3 of the documents (2,667), and have agreed that the remaining documents are “non-responsive.” The Court, therefore, likewise considers the remaining documents as non-responsive.
The parties have manipulated the numbers and categorized the 8,660 documents in various ways that supports their respective positions. Plaintiffs claim that 567 of the documents are responsive, i.e., are pornographic or are documents reflecting CBRE policies and procedures. Therefore they calculate that 567 of the 2,667 documents were responsive, for a 21.3% responsive rate. This is technically accurate — 21.3% of the documents that the parties reviewed were responsive. By agreeing that the remaining unreviewed documents were non-responsive, the parties effectively agreed, however, that the pertinent number for the denominator was the 8,660 documents. Therefore, (567/8,660) equals a 6.5% responsive rate. Defendants of course have calculated a much smaller responsive rate, and claim that the parties have identified only (142/8,660) documents as responsive, for a 1.64% responsive rate.
II. Discussion
A. General Principles
Guiding this Court’s decision are the overarching principles of Federal Rule of Civil Procedure 26. Under the familiar language, a party may seek discovery of “any matter, not privileged, that is relevant to the claim or defense of any party.... Relevant information need not be admissible at the trial if the discovery appears reasonably calculated to lead to the discovery of admissible evidence.” Fed.R.Civ.P. 26(b)(1). The court may limit discovery if it determines that the burden of the discovery outweighs its likely benefit. Fed.R.Civ.P. 26(b)(2)(iii). To make this determination, the Court will consider what has been dubbed the proportionality test of Rule 26(b)(2)(iii): the needs of the case, the amount in controversy, the resources of the parties, the importance of the issues at stake in the litigation, and the importance of the proposed discovery in resolving the issues. In this way, parties are protected from unduly burdensome or expensive discovery requests. Fed.R.Civ.P. 26(c).
The Court also begins this discussion with the general presumption in discovery that the responding party must bear the [572]*572expense of complying with discovery requests. Oppenheimer Fund, Inc. v. Sanders, 437 U.S. 340, 358, 98 S.Ct. 2380, 57 L.Ed.2d 253 (1978). However, if the responding party asks the court for an order protecting it from “undue burden or expense,” the court may shift the costs to the non-producing party, rather than just disallowing the requested discovery. Fed.R.Civ.P. 26(c); Oppenheimer Fund, Inc., 437 U.S. at 358, 98 5. Ct. 2380; Rowe Entm’t v. The William Morris Agency, Inc., 205 F.R.D. 421, 428 (S.D.N.Y.2002); Fed. R. Civ. Pro. 34, Advisory Committee Notes, 1970 Amendment (“courts have ample power under Rule 26(c) to protect respondent against undue burden or expense, either by restricting discovery or requiring that the discovering party pay costs”).
B. Standards for Discovery of Electronic Data
Electronic data, such as e-mails, are discoverable. As contrasted with traditional paper discovery, e-discovery has the potential to be vastly more expensive due to the sheer volume of electronic information that can be easily and inexpensively stored on backup media. Zubulake v. UBS Warburg LLC, 217 F.R.D. 309, 316 (S.D.N.Y.2003) (“Zubulake I”); Byers v. Ill. State Police, No. 99 C 8105, 2002 WL 1264004, at *10 (N.D.Ill. June 3, 2002). Depending on how the electronic data is stored, it can be difficult, and hence expensive, to retrieve the data and search it for relevant documents. Theoretically, as technology improves, retrieving and searching data will become more standard and less costly. See e.g., Discovery By Keyword Search, 15)No. 3 Prac. Lit. 7 (2004).
In the meantime, until the technology advances and e-discovery becomes less expensive, cost will continue to be an issue as parties battle over who will foot the bill. In the electronic arena, three main tests have been suggested to determine when it is appropriate to shift the costs of searching and producing inaccessible data to the requesting party in order to protect the producing party from unduly burdensome e-discovery requests.6
First, under the marginal utility approach, the more likely it is that the search will discover critical information, the fairer it is to have the responding party search at its own expense. McPeek v. Ashcroft, 202 F.R.D. 31, 34 (D.D.C.2001). Next, the court in Rowe created eight factors for consideration in the cost-shifting analysis, one of which incorporated the marginal utility test.7 205 F.R.D. at 429. Finally, the court in Zubulake I modified the Rowe test to account for the fact that it interpreted the Rowe test as generally favoring cost-shifting, which had ignored the presumption that the responding party pays for discovery.8 217 F.R.D. at 320. We agree with both the Rowe court and the Zubulake court that the marginal utility test is the most important factor. Furthermore, while we are guided by the remainder of the Rowe and Zubulake factors, we find that the proportionality test set forth in Rule 26(b)(2)(iii) must shape the test. Thus, we modify the Zubulake rules by adding a factor that considers the importance of [573]*573the requested discovery in resolving the issues of the litigation.
Therefore, we will consider the following factors: 1) the likelihood of discovering critical information; 2) the availability of such information from other sources; 3) the amount in controversy as compared to the total cost of production; 4) the parties’ resources as compared to the total cost of production; 5) the relative ability of each party to control costs and its incentive to do so; 6) the importance of the issues at stake in the litigation; 7) the importance of the requested discovery in resolving the issues at stake in the litigation; and 8) the relative benefits to the parties of obtaining the information. At all times we keep in mind that because the presumption is that the responding party pays for discovery requests, the burden remains with CBRE to demonstrate that costs should be shifted to Plaintiffs. See Zubulake II, 216 F.R.D. at 283.
C. Application of the Eight Factors
Marginal Utility
1. The likelihood of discovering critical information.9
When the matter is initially brought to the court’s attention, the extent to which the request appears to be specifically tailored to discover relevant information may help the court weigh this factor. Zubulake I, 211 F.R.D. at 323 (first factor); Rowe, 205 F.R.D. at 429 (first factor); Byers, 2002 WL 1264004, at *12 (shifting costs to requesting party where it was unlikely the search would uncover relevant information). If a test run is ordered, as in this case, the actual results of the test run will be indicative of how likely it is that critical information will be discovered.
The 8 term search list chosen by the parties contains five pornographic terms, four of which were selected by Plaintiffs, and three disciplinary terms, all three of which were chosen by CBRE. The 8 term search resulted in 8,660 hits. Whether many of the resulting documents are actually responsive is subject to debate by the parties, and they have gone to great lengths to characterize the documents in various ways that support their respective positions. Plaintiffs have identified at least 567 documents as being responsive which is a 6.5% responsive rate.10 These documents include graphic pornographic images, sexual correspondence and jokes, and CBRE policies and procedures regarding the circulation of inappropriate email and the visitation of inappropriate websites, as well policies relating to sexual harassment. Plaintiffs also claim they recovered documents demonstrating the demeaning attitude of CBRE towards its female employees pervading the work environment. Some of these documents appeared in multiple user accounts which supports Plaintiffs’ theory that these types of documents were being spread throughout the offices. On the other hand, some of these documents were sent from a single user to another single user. There is nothing to indicate that any CBRE employee indicated that these e-mails were offensive or that any employee refused any future such e-mails. Furthermore, Plaintiffs do not explain how one-to-one emails support their theory of a hostile work environment.
CBRE disputes these classifications, arguing ' that the majority of the 8,660 documents are not even relevant, extrapolating the results of the documents that were reviewed to all the documents that were found. It asserts that many of the e-mails were not considered offensive to the women who received them, or were sent from men to men.11 CBRE’s interpretation of the wom[574]*574en’s tolerance to sexually explicit e-mails, however, is necessarily slanted in its own favor, and there is no information regarding whether women routinely viewed other people’s e-mails, such as secretaries who accessed their bosses’ e-mails as part of their job requirements. CBRE would further decrease the number of relevant documents by excluding documents concerning CBRE’s policies and procedures. It was CBRE, however, who chose the three disciplinary search terms so it cannot now complain now that the terms it selected successfully found the documents they were intended to find. Therefore, under CBRE’s analysis, even excluding what CBRE has characterized as male-to-male e-mails, there are 386 responsive documents.12 This is a 4.5% responsive rate.
Before determining whether this factor supports cost-shifting, we will consider whether this information is available from other sources.
2. The availability of such information from other sources.13
If the information is available from another source, the marginal utility from the e-discovery is low, and would support cost-shifting. These first two factors which comprise the marginal utility test are the most important, because the more likely it is that relevant information will be discovered, the fairer it is to make the responding party pay for the information. Zubulake I, 217 F.R.D. at 323.
Clearly, relevant information on CBRE’s backup e-mail tapes is only available through restoring and searching the backup tapes unless it has been previously produced by CBRE. The test search did result in relevant documents that had not been produced by CBRE. Specifically, although CBRE argues that it has produced relevant policy documents, we note that the search did find policy documents that had not been previously produced, as well as e-mails containing pornographic material which had not been produced and are not available through any other source. In addition to confirming the existence of relevant documents on backup tapes that have not been restored, it is likely that there were relevant documents that have already been destroyed as CBRE failed in its duty to preserve relevant electronic information.14
Given that the search resulted in a 4.5 to 6.5% responsive rate, and that the majority of the documents are not available from another source, we consider whether the marginal utility factors support shifting the cost to Plaintiffs. As a comparison, in Zubulake II, the test run ordered by the court resulted in 1,075 e-mails after duplicates were eliminated, 600 of which were deemed responsive by the defendant (or 55.8%), and 68 of which were identified by the plaintiff as directly supporting her gender discrimination claim (6.3%). 216 F.R.D. at 282, 285. The court found that although all the documents had some relevancy to the claims of the case, not one e-mail provided any direct evidence of discrimination. Id. at 286. The most the emails demonstrated was the “dysfunctional atmosphere” in which the plaintiff worked. Id. at 285. The court, nevertheless, found that there was a “speculative” chance that the remaining backup tapes would provide evidence of direct discrimination, especially because some of the restored evidence indicated that the plaintiffs supervisor had intentionally deleted especially relevant e-mail. Id. at 287. It therefore found that the marginal utility was potentially high due to the types and number of documents that had been recovered, and that the test tipped slightly against cost-shifting even though the [575]*575actual number of recovered relevant documents was low. Id.
In this case, the search has resulted in a small number of relevant documents, although substantially less than in Zubulake, as would be expected. In Zubulake, the plaintiff was able to narrow her search to five employees and used only search terms involving her name and initials. In this case, Plaintiffs are attempting to prove that discrimination existed in CBRE offices across the nation. Although CBRE argues that the number of pornographic and policy documents are only a minuscule fraction of the total documents on the backup tapes,15 we decline to determine at this juncture exactly what percentage of documents on the e-mail system would prove a hostile environment. But, because the test results are partially based on Plaintiffs’ selection of what Plaintiffs believed are words or terms most likely to produce evidence of a sexually hostile atmosphere, the Court is of the opinion that the percentage of sexually objectionable emails is substantially lower than 4.5%. It will be Plaintiffs’ onus to demonstrate that hostile environment existed in CBRE’s offices; in the meantime they are entitled to relevant information as long as it is reasonably calculated to lead to the discovery of admissible evidence. Fed.R.Civ.P. 26(b)(1). Nevertheless, because the search also revealed a significant number of unresponsive documents, we find that the marginal utility test weighs slightly in favor of cost-shifting.16 Factors 8-5 The Cost Factors
3. The amount in controversy, as compared to the total cost of production.17
Plaintiffs’ expert estimated that the total cost of the production would range from $183,500 to $249,900 for ten offices considering the initial results of the 92 term search list. The parties have not addressed whether the results of the 8 term search allowed the expert to more accurately estimate the production costs.
Plaintiffs claim that should a class be certified, their class recovery could extend into the tens of millions of dollars. While the Court cannot completely accept Plaintiffs’ speculative estimate of its potential damage award, neither can it accept that their claims are worthless, especially considering that the burden of proving that cost-shifting is warranted falls on CBRE’s shoulders. Furthermore, even if the class is not certified, there are still five named plaintiffs, and their recovery is potentially high enough to justify some of the costs of discovery. See Xpedior Creditor Trust v. Credit Suisse First Boston (USA), Inc., 309 F.Supp.2d 459, 466 (S.D.N.Y.2003) (noting that even if a class were not certified, the named plaintiffs claims were potentially worth tens of millions of dollars). See also Zubulake II, 216 F.R.D. at 288 (noting that the case was not a nuisance value case, a small case, or a frivolous case).
CBRE would have the Court find that the total cost of production for all of its offices nationwide would stretch into the millions of dollars, ignoring the fact that discovery has been limited to only a small fraction of the 125 offices. Nevertheless, several hundred thousand dollars for one limited part of discovery is a substantial amount of actual dollars to pay for such a search. Therefore, this factor weighs in favor of cost-shifting.
4. The parties’ resources, as compared to the total cost of production.18
According to its website, CBRE is “the global leader in real estate services.” As stated on its most recent from 10-K/A, [576]*576CBRE had net revenues of 1.6 billion dollars for fiscal year 2003. Plaintiffs, who include former employees of CBRE, are clearly at a serious financial disadvantage. Plaintiffs’ counsel, however, obviously has been willing to front substantial amounts of money, and probably could contribute to these discovery costs. See Zubulake II, 216 F.R.D. at 288 (noting that it is common for plaintiffs firms to front huge expenses when the recovery is potentially millions of dollars). CBRE is now put in the awkward position of reversing its previous argument for factor - number three that this case is not worth much money, because now it must argue that the case is potentially worth millions, so that costs should be shifted to Plaintiffs. Because we found that the recovery in this case is potentially high, but we also now find that CBRE’s resources are large compared to the total cost of production, we find that this factor weighs against cost-shifting. See Xpedior, 309 F.Supp.2d at 466 (finding that this factor weighs against cost-shifting where defendant’s assets dwarfed plaintiffs assets even if plaintiffs attorney could contribute).
5. The relative ability of each party to control costs and its incentive to do so.19
In most cases, both parties will likely have the ability and desire to control costs. The requesting party should have an incentive for limiting its requests, and the responding party may have an incentive for using accessible searchable media, or for pressuring its software contractors to create such media or software. The responding party may also employ cost-saving measures, such as performing a key word search for a privilege review rather than examining each document.
In this case, it appears that the ability to control costs partially pivots around the selection of the vendor. Plaintiffs have agreed to work with CBRE to jointly select a new electronic discovery service and to minimize costs to the extent it is possible to do so. ■ The costs of the search, however, are also driven to some extent by the scope of the search as selected by Plaintiffs. A smaller search term list would result in less hits, and less documents that must be transferred to an electronic viewer. For example, words that did not result in a large number of documents could be taken off the term search list. Documents that are non-responsive and appear multiple times (such as an email discussing a virus with the words sexy-virgin) could easily be searched for and eliminated. However, Plaintiffs’ search must necessarily be broad, due to the nature of the information for which they are searching. Therefore, we find that this factor slightly weighs in favor of cost-shifting.
Remaining Factors
6. The importance of the issues at stake in the litigation,20
Plaintiffs cite Zubulake II for the proposition that this factor “will only rarely come into play ... [and that] discrimination in the workplace ... is hardly unique.” 216 F.R.D. at 289 (quotation omitted); see also Xpedior, 309 F.Supp.2d at 466 (finding that a potential class action involving manipulation of the securities market does not raise the kind of public policy issues that might affect cost-shifting). Because the parties agree that this factor is neutral, we find that it does not weigh in favor of or against cost-shifting.
7. The importance of the requested discovery in resolving the issues at stake in the litigation.
As this is a factor explicitly set forth in Rule 26(b)(2)(iii), we add it to the Rowe and Zubulake analysis. See In re Gen. Instrument Corp., No. 96 C 1129, 1999 WL 1072507, at *6 (N.D.Ill. Nov.18, 1999) (denying plaintiffs’ motion to compel where plaintiffs did not identify any specific factual issue for which additional discovery would help them prove their case).
CBRE argues that Plaintiffs’ claims are not primarily based on pornographic material that may have been circulating through its offices. Plaintiffs respond that this information supports their hostile environment claim. Plaintiffs are entitled to obtain discovery rel[577]*577evant to their claims, as long as the discovery is reasonably calculated to lead to the discovery of admissible evidence. See Fed.R.Civ.P. 26(b)(1). If relevance is in doubt, courts should err on the side of permissive discovery. Channelmark Corp. v. Destination Prods. Inti, Inc., 99 C 214, 2000 WL 968818, at *2-3 (N.D.Ill. July 7, 2000). As there is reason to believe that the requested discovery would assist in resolving the issues at stake in this case, but because there is also other evidence to support Plaintiffs’ claims, we find that this factor weighs slightly in favor of cost-shifting.
8. The relative benefits to the parties of obtaining the information.21
This factor is the least important because in discovery the requested information is more likely to benefit the requesting party. See Zubulake I, 217 F.R.D. at 323 (finding that this is the least important factor because the requesting party usually benefits from its requests). In some cases, the information may aid the producing party, in which case it is more fair to require the producing party to pay for the discovery. In this case, the information requested will benefit Plaintiffs more than CBRE. Therefore, this factor is neutral.
D. Summary
Factors 1 and 2, the most important factors, weigh slightly in favor of cost-shifting to Plaintiffs. For the cost factors: factor 3 weighs in favor of cost-shifting; factor 4 weighs against cost-shifting; and factor 5 weighs slightly in favor of cost-shifting. Factor 6 is neutral; factor 7 weighs slightly in favor of cost-shifting; and factor 8 is neutral.
Therefore, because the factors favor cost-shifting, but the presumption is that the responding party pays for discovery costs, we find that CBRE should bear 25% and Plaintiffs 75% of the discovery costs of restoring the tapes, searching the data, and transferring it to an electronic data viewer. Each party will bear their own costs of reviewing the data and printing documents, where necessary.
III. Conclusion
For the foregoing reasons, Plaintiffs’ motion for costs is granted in part and denied in part as stated in this opinion.