OPINION
HAWKINS, Senior Circuit Judge:
The settlement agreement approved in this products liability class action provides the class $100,000 in cy pres awards and zero dollars for economic injury, while setting aside up to $800,000 for class counsel and $12,000 for the class representatives— amounts which the court subsequently awarded in full in a separate order. William Brennan and other class members (collectively “Objectors”) challenge the fairness and reasonableness of the settlement and appeal both the approval and fee orders, arguing the district court abused its discretion in failing to consider whether the gross disproportion between the class award and the negotiated fee award was reasonable. We agree that the disparity between the value of the class recovery and class counsel’s compensation raises at least an inference of unfairness, and that the current record does not adequately dispel the possibility that class counsel bargained away a benefit to the class in exchange for their own interests. We therefore vacate both orders and remand so that the district court may conduct a more searching inquiry into the fairness of the negotiated distribution of funds, as well as consider the substantive reasonableness of the attorneys’ fee request in light of the degree of success attained.
FACTS AND PROCEEDINGS
I. Background
Plaintiffs filed twenty-six putative class actions in courts around the country against Motorola, Inc., Plantronics, Inc., and GN Netcom, Inc. (collectively “defendants”), alleging defendants knowingly failed to disclose the potential risk of noise-induced hearing loss1 associated with extended use of their wireless Blue-tooth headsets at high volumes, in violation of state consumer fraud protection and unfair business practice laws.2 The Judi[939]*939cial Panel on Multidistrict Litigation coordinated these cases in In re Bluetooth Headset Products Liability Litigation in the Central District of California.3
Plaintiffs’ Second Amended Consolidated Complaint sought money damages on behalf of millions of individuals who had purchased Bluetooth headsets since June 30, 2002,4 purportedly in reliance on allegedly misleading representations about the safety and usability of the product. Plaintiffs alleged that defendants advertised “talk times” of three hours or longer, while in reality, consumers could not safely use the headsets for more than a few minutes each day without exposing themselves to the risk of noise-induced hearing loss — a risk that defendants failed to disclose in any of their marketing materials. The complaint did not state a claim for personal injury but asserted economic injury, alleging plaintiffs would not have purchased their Bluetooth headsets but for defendants’ willful false advertising. Plaintiffs sought actual damages in the amount paid for the product, which they claimed to be between $70 and $150 per headset, along with injunctive relief, restitution, punitive damages, attorneys’ fees and costs.
Class counsel spent considerable time researching legal and industry standards on acceptable noise levels, surveying warnings on other audio devices, obtaining acoustic test results and other documents from defendants, and working with experts to review this data and evaluate the risk of noise-induced hearing loss. The parties voluntarily exchanged discovery and held at least three in-person meetings to discuss the merits of the litigation and discovery issues before participating in a formal mediation session, overseen by a retired California Court of Appeal Justice.
Unable to reach a settlement at that time, defendants shortly thereafter filed a joint motion to dismiss on various grounds, insisting their products are safe and denying any wrongdoing on their part. The motion to dismiss was fully briefed by both sides, but before the court heard argument, and before any motion was made to certify a class for merits purposes, the parties successfully participated in another mediation session and filed a proposed class action settlement agreement purporting to resolve all claims.
II. The Terms of the Settlement Agreement
In exchange for plaintiffs’ general release and waiver of all asserted claims defendants agreed to: (1) post acoustic safety information on their respective websites and in their product manuals and/or packaging for new Bluetooth headsets; (2) pay a total of $100,000 in cy pres awards to be distributed among four non-profit organizations dedicated to the prevention of hearing loss;5 (3) pay notice costs up to $1.2 million; (4) pay documented costs to class counsel up to $38,000, or if notice costs fell below $1.2 million, no more than $50,000; (5) pay attorneys’ fees in an [940]*940amount set by the district court, not to exceed $800,000; and (6) pay an incentive award in an amount set by the district court, not to exceed $12,000, to be divided among the nine class representatives.6 Approval of the settlement was not conditioned on any minimum attorneys’ fee award, minimum costs award, or the payment of any incentive award to any plaintiff.
III. Approval of the Settlement Agreement
Pursuant to the district court’s preliminary approval order directing the provision of notice, the parties implemented a comprehensive notice plan comprised of direct mailings, magazine advertisements, and a dedicated Internet website, reaching about 80% of potential class members an average of more than 2.5 times each. Of the millions of potential class members, 715 people validly elected to opt out of the settlement. Fifty people, including the seven Appellants before us now, sent in objections to the settlement and were given an opportunity to be heard at the fairness hearing. After considering the objections, the district court entered an Order (“Approval Order”) and Final Judgment certifying the class for settlement purposes only, pursuant to Federal Rule of Civil Procedure 23(b)(3), and approving the settlement agreement as fair, reasonable, and adequate.
IV. Award of Fees and Costs
After ordering class counsel to produce additional unredacted billing records and reviewing the files submitted, the district court later entered a separate order (“Fee Order”) awarding $850,000 to class counsel for fees and costs, based on a lodestar method calculation, and $12,000 to be distributed among the nine representative plaintiffs.
Objectors timely appealed both the Approval and Fee Orders.
STANDARDS OF REVIEW
We review a district court’s approval of a class action settlement for clear abuse of discretion. Rodriguez v. W. Publ’g Corp., 563 F.3d 948, 963 (9th Cir. 2009). Such review is “extremely limited,” and we “will affirm if the district judge applies the proper legal standard and his findings of fact are not clearly erroneous.” In re Mego Fin. Corp. Sec. Litig., 213 F.3d 454, 458 (9th Cir.2000).
We also review for abuse of discretion a district court’s award of fees and costs to class counsel, as well as its method of calculation. Lobatz v. U.S. W. Cellular of Cal., Inc., 222 F.3d 1142, 1148-49 (9th Cir.2000). Findings of fact underlying an award of fees are reviewed for clear error. Id. at 1148.
DISCUSSION
Objectors contest both the fee award and the approval order, but their objections are motivated by the same issue: what they claim are excessive attorneys’ fees, negotiated unfairly by class counsel and ultimately awarded unreasonably by the court. Objectors argue that the district court should not have approved as fair and reasonable a settlement agreement that, on its face, so disproportionately advances the interests of class counsel over those of the class itself. They further contend that, even if the approval order can be upheld, class counsel should not have been awarded eight times the amount of the class recovery. Rather, they argue, [941]*941the two negotiated sums should have been viewed as a “constructive common fund” and fees limited to an appropriate percentage thereof.
We would ordinarily begin our review with the Approval Order, whose vacatur would render moot the challenge to the Fee Order. However, because Objectors’ challenge to the fairness of the settlement agreement relies, in large part, on a determination that the requested fees were substantively unreasonable, we instead begin by temporarily assuming the Approval Order to be valid and proceed to examine the reasonableness of the fee award first, before then returning to review the Approval Order itself.
I. Attorneys’ Fee Award A. Applicable Legal Standards
While attorneys’ fees and costs may be awarded in a certified class action where so authorized by law or the parties’ agreement, Fed.R.Civ.P. 23(h), courts have an independent obligation to ensure that the award, like the settlement itself, is reasonable, even if the parties have already agreed to an amount. See Staton v. Boeing Co., 327 F.3d 938, 963-64 (9th Cir. 2003); Knisley v. Network Assoc., 312 F.3d 1123, 1125 (9th Cir.2002); Zucker v. Occidental Petroleum Corp., 192 F.3d 1323, 1328-29 & n. 20 (9th Cir.1999). The reasonableness of any fee award must be considered against the backdrop of the “American Rule,” which provides that courts generally are without discretion to award attorneys’ fees to a prevailing plaintiff unless (1) fee-shifting is expressly authorized by the governing statute; (2) the opponents acted in bad faith or willfully violated a court order; or (3) “the successful litigants have created a common fund for recovery or extended a substantial benefit to a class.” Alyeska Pipeline Serv. Co. v. Wilderness Soc., 421 U.S. 240, 275, 95 S.Ct. 1612, 44 L.Ed.2d 141 (1975) (Brennan, J., dissenting); accord Zambrano v. City of Tustin, 885 F.2d 1473, 1481 & n. 25 (9th Cir.1989).
The award of attorneys’ fees in a class action settlement is often justified by the common fund or statutory fee-shifting exceptions to the American Rule, and sometimes by both. See Staton, 327 F.3d at 972; see also Court Awarded Attorney Fees, Third Circuit Task Force, 108 F.R.D. 237, 250 (1985) (purpose of common-fund exception is to “avoid the unjust enrichment of those who benefit from the fund that is created, protected, or increased by the litigation and who otherwise would bear none of the litigation costs”). We have approved two different methods for calculating a reasonable attorneys’ fee depending on the circumstances.
The “lodestar method” is appropriate in class actions brought under fee-shifting statutes (such as federal civil rights, securities, antitrust, copyright, and patent acts), where the relief sought — and obtained — is often primarily injunctive in nature and thus not easily monetized, but where the legislature has authorized the award of fees to ensure compensation for counsel undertaking socially beneficial litigation. See Hanlon v. Chrysler Corp., 150 F.3d 1011, 1029 (9th Cir.1998); In re General Motors Corp. Pick-Up Truck Fuel Tank Prods. Liability Litig., 55 F.3d 768, 821 (3d Cir.1995).
The lodestar figure is calculated by multiplying the number of hours the prevailing party reasonably expended on the litigation (as supported by adequate documentation) by a reasonable hourly rate for the region and for the experience of the lawyer. Staton, 327 F.3d at 965. Though the lodestar figure is “presumptively reasonable,” Cunningham v. Cnty. of Los Angeles, 879 F.2d 481, 488 (9th Cir.1988), the court may adjust it upward or downward by an appropriate positive or [942]*942negative multiplier reflecting a host of “reasonableness” factors, “including the quality of representation, the benefit obtained for the class, the complexity and novelty of the issues presented, and the risk of nonpayment,” Hanlon, 150 F.3d at 1029 (citing Kerr v. Screen Extras Guild, Inc., 526 F.2d 67, 70 (9th Cir.1975)7). Foremost among these considerations, however, is the benefit obtained for the class. See Hensley v. Eckerhart, 461 U.S. 424, 434-36, 103 S.Ct. 1933, 76 L.Ed.2d 40 (1983); McCown v. City of Fontana, 565 F.3d 1097, 1102 (9th Cir.2009) (ultimate reasonableness of the fee “is determined primarily by reference to the level of success achieved by the plaintiff”). Thus, where the plaintiff has achieved “only limited success,” counting all hours expended on the litigation — even those reasonably spent — may produce an “excessive amount,” and the Supreme Court has instructed district courts to instead “award only that amount of fees that is reasonable in relation to the results obtained.” Hensley, 461 U.S. at 436, 440, 103 S.Ct. 1933.
Where a settlement produces a common fund for the benefit of the entire class, courts have discretion to employ either the lodestar method or the percentage-of-recovery method. In re Mercury Interactive Corp., 618 F.3d 988, 992 (9th Cir.2010) (citing Powers v. Eichen, 229 F.3d 1249, 1256 (9th Cir.2000)). Because the benefit to the class is easily quantified in common-fund settlements, we have allowed courts to award attorneys a percentage of the common fund in lieu of the often more time-consuming task of calculating the lodestar. Applying this calculation method, courts typically calculate 25% of the fund as the “benchmark” for a reasonable fee award, providing adequate explanation in the record of any “special circumstances” justifying a departure. Six (6) Mexican Workers v. Aria. Citrus Growers, 904 F.2d 1301, 1311 (9th Cir.1990); accord Powers, 229 F.3d at 1256-57; Paul, Johnson, Alston & Hunt v. Graulty, 886 F.2d 268, 272 (9th Cir.1989).
Though courts have discretion to choose which calculation method they use, their discretion must be exercised so as to achieve a reasonable result. See In re Coordinated Pretrial Proceedings, 109 F.3d 602, 607 (9th Cir.1997) (citing In re Wash. Pub. Power Supply Sys. Sec. Litig., 19 F.3d 1291, 1294-95 n. 2 (9th Cir.1994)). Thus, for example, where awarding 25% of a “megafund” would yield windfall profits for class counsel in light of the hours spent on the case, courts should adjust the benchmark percentage or employ the lodestar method instead. Six Mexican Work[943]*943ers, 904 F.2d at 1311; see In re Prudential Ins. Co. America Sales Practice Litig. Agent Actions, 148 F.3d 283, 339 (3d Cir. 1998) (explaining that basis for inverse relationship between size of fund and percentage awarded for fees is that “in many instances the increase in recovery is merely a factor of the size of the class and has no direct relationship to the efforts of counsel” (internal quotation marks omitted)).
B. Approval of the Attorneys’ Fee Award in the Present Case
The district court here applied the lodestar method, although it never announced a lodestar figure. After finding numerous defects in class counsel’s proposed computation of its $1.6 million lodestar, including duplicative entries, excessive charges for most categories of services, a substantial amount of block billing, and use of an inflated hourly rate, the court announced that its own analysis revealed the lodestar still “substantially exceeds” the $800,000 defendants agreed to pay.
Citing our previous observation that a “defendant is interested only in disposing of the total claim asserted against it,” Sta-ton, 327 F.3d at 964, Objectors argue the district court should have treated this settlement as producing a “constructive common fund” and employed a percentage-of-recovery method to assess the reasonableness of the $800,000 fee award, rather than relying exclusively on a lodestar calculation. Several courts have embraced the constructive common fund approach, warning that “private agreements to structure artificially separate fee and settlement arrangements” should not enable parties to circumvent the 25% benchmark requirement on “what is in economic reality a common fund situation.” In re Gen. Motors, 55 F.3d at 821; see Johnston v. Cornerica Mortg. Corp., 83 F.3d 241, 246 (8th Cir.1996) (“[I]n essence the entire settlement amount comes from the same source. The award to the class and the agreement on attorney fees represent a package deal.”); cf. Manual for Complex Litig. § 21.75 (4th ed. 2008) (“If an agreement is reached on the amount of a settlement fund and a separate amount for attorney fees ... the sum of the two amounts ordinarily should be treated as a settlement fund for the benefit of the class.”). Plaintiffs insist this is not a common-fund case because the relief obtained was primarily injunctive in nature and because, as prevailing parties under California Civil Code § 1750’s fee-shifting provision, they are entitled to attorneys’ fees calculated under the lodestar method.
Whether or not we view this as a common-fund case, we agree with Objectors that the district court needed to do more to assure itself — and us — that the amount awarded was not unreasonably excessive in light of the results achieved. Notably, the district court made (1) no explicit calculation of a reasonable lodestar amount; (2) no comparison between the settlement’s attorneys’ fees award and the benefit to the class or degree of success in the litigation; and (3) no comparison between the lodestar amount and a reasonable percentage award. On this record, we lack a sufficient basis for determining the reasonableness of the award.
First, our discomfort is not with the choice of the lodestar method as a primary basis for calculation, but rather the absence of explicit calculation or explanation of the district court’s result. The district court “s[aw] no need” to calculate a precise lodestar amount in light of defendants’ willingness to pay and because reducing the award below $800,000 would in no way benefit the class or enhance the cy pres award. But a defendant’s advance agreement not to object cannot relieve the district court of its duty to assess fully the reasonableness of the fee request. See [944]*944Staton, 327 F.3d at 963-64; Knisley, 312 F.3d at 1125. Under the lodestar method, the district court must calculate the lodestar figure based on the number of hours reasonably expended on the litigation, adjusting the figure to account for the degree of success class counsel attained, along with other factors. From the face of the Fee Order, however, we do not have sufficient information from which to conclude that the district court included a reasonable number of hours in its lodestar “calculation” or that it “considered the relationship between the amount of the fee awarded and the results obtained.” Hensley, 461 U.S. at 437, 103 S.Ct. 1933 (“It [is] important ... for the district court to provide a concise but clear explanation of its reasons for the fee award.”); accord, McCown, 565 F.3d at 1102 (“Once the district court completes its analysis of the final lodestar amount, it must explain how it arrived at its determination with sufficient specificity to permit an appellate court to determine whether the district court abused its discretion in the way the analysis was undertaken”). Instead, we know only that the district court believed the lodestar figure to be less than $1.6 million but greater than $800,000.
Second, the district court declined to reduce the award because the injunctive relief and cy pres payment provided “at least minimal benefit,” even while acknowledging that “the settlement did not achieve all the goals of the suit.” The district court appears to have conflated the Rule 23(e) standard for approval of a settlement agreement (which requires consideration of whether the settlement agreement offers plaintiffs more than they are likely to achieve at trial) with the requirement that the fee award be “reasonable in relation to the results obtained.” Hensley, 461 U.S. at 440, 103 S.Ct. 1933. Although the court stated that the “substantial reduction” below the lodestar accounted for the fact that “there were a number of claims for which class counsel achieved no relief,” we remain in the dark both as to how substantial a reduction it was and what level of success plaintiffs in fact achieved. The district court made no findings in its Approval Order with regard to the value vel non of the injunctive relief, noting only that the cy pres award was an appropriate form of relief given the large class size. Nor did it discuss in the Fee Order the value of the injunctive relief or whether it in fact was “socially beneficial” as would justify a fee award under California Civil Code § 1750’s fee-shifting provision. Cf Fed.R.Civ.P. 23(h), 2003 Advisory Cmte. Notes (“Settlements involving nonmonetary provisions for class members also deserve careful scrutiny to ensure that these provisions have actual value to the class.”); In re TD Ameritrade Accountholder Litig., 266 F.R.D. 418, 423 (N.D.Cal.2009) (“[T]he standard [under Rule 23(e) ] is not how much money a company spends on purported benefits, but the value of those benefits to the class.” (citing O’Keefe v. Mercedes-Benz USA LLC, 214 F.R.D. 266, 304 (E.D.Pa.2003))). With neither a lodestar figure nor a sense of what degree of success this settlement agreement achieved, we have no basis for affirming the fee award as reasonable under the lodestar approach.
Third, even though a district court has discretion to choose how it calculates fees, we have said many times that it “abuses that ‘discretion when it uses a mechanical or formulaic approach that results in an unreasonable reward.’ ” In re Mercury Interactive Corp., 618 F.3d at 992 (quoting Powers, 229 F.3d at 1256). Thus, even though the lodestar method may be a perfectly appropriate method of fee calculation, we have also encouraged courts to guard against an unreasonable result by cross-checking their calculations against a second method. See, e.g., Vizcaino v. Mi[945]*945crosoft Corp., 290 F.3d 1043, 1050-51 (9th Cir.2002); see also In re Gen. Motors, 55 F.3d at 820. Just as the lodestar method can “confirm that a percentage of recovery amount does not award counsel an exorbitant hourly rate,” the percentage-of-recovery method can likewise “be used to assure that counsel’s fee does not dwarf class recovery.” In re Gen. Motors, 55 F.3d at 821 n. 40. “If the lodestar amount overeompensates the attorneys according to the 25% benchmark standard, then a second look to evaluate the reasonableness of the hours worked and rates claimed is appropriate.” In re Coordinated Pretrial Proceedings, 109 F.3d at 607.
If the district court here took that second look, the record does not reflect it. Absent any explanation from the district court, we are concerned that the amount awarded was 83.2% of the total amount defendants were willing to spend to settle the case, viewing the $800,000 allotment for attorneys’ fees, the $12,000 allotment for an incentive award, the $100,000 cy pres award, and the $50,000 allotment for fees as a “constructive common fund.” Twenty-five percent of this $962,000 fund, by contrast, would have yielded only $240,500 in attorneys’ fees. Even if we included the $1.2 million notice costs in the constructive fund (and accordingly reduced the fees to $38,000), the attorneys’ fees awarded would constitute 37.2% of this $2.15 million fund, in contrast to a 25% benchmark figure of $537,500. Plaintiffs urge us to find that the fee award is justified because the injunctive relief confers a valuable benefit and was the primary objective of the lawsuit, but the district court did not make findings on the value of the injunctive relief, so we cannot evaluate whether it justifies an otherwise disproportionate award.8
While we cannot say the disproportion between the fee award and the benefit obtained for the class was per se unreasonable, in the absence of an adequate explanation of the award, “we have no choice but to remand the case to the district court to permit it to make the necessary calculations and provide the necessary explanations.” McCown, 565 F.3d at 1102 (citing Tutor-Saliba Corp. v. City of Hailey, 452 F.3d 1055, 1065 (9th Cir.2006)).
On remand, the district court should (1) decide whether to treat the settlement as a common fund; (2) choose the lodestar or percentage method for calculating a reasonable fee and make explicit calculations; (3) ensure that the fee award is reasonable considering, inter alia, the degree of success in the litigation and benefit to the class; and (4) if standard calculations yield an unjustifiably disproportionate award, adjust the lodestar or percentage accordingly.
II. Approval of the Settlement Agreement
A. Applicable Legal Standards
Approval of the settlement agreement was not conditioned on the award of attorneys’ fees and costs or an incentive award, and therefore our vacatur of the fee award does not necessitate invalidation of the approval order. See, e.g., Rodriguez, 563 F.3d at 969 (affirming approval of the settlement but reversing and remanding the award of attorneys’ fees). Nonetheless, because the parties expressly negotiated a possibly unreasonable amount of fees, and because the district court did not take this [946]*946possibility into account in reviewing the settlement’s fairness the first time around, we must vacate and remand the Approval Order as well, so that the court may appropriately factor this into its Rule 23(e) analysis. On remand, the district court should reconsider its approval of the settlement after recalculating a reasonable amount of fees for class counsel.
Courts have long recognized that “settlement class actions present unique due process concerns for absent class members.” Hanlon, 150 F.3d at 1026. One inherent risk is that class counsel may collude with the defendants, “tacitly reducing the overall settlement in return for a higher attorney’s fee.” Knisley, 312 F.3d at 1125; see Evans v. Jeff D., 475 U.S. 717, 733, 106 S.Ct. 1531, 89 L.Ed.2d 747 (1986) (recognizing that “the possibility of a tradeoff between merits relief and attorneys’ fees” is often implicit in class action settlement negotiations); In re Gen. Motors, 55 F.3d at 805.
To guard against this potential for class action abuse, Rule 23(e) of the Federal Rules of Civil Procedure requires court approval of all class action settlements, which may be granted only after a fairness hearing and a determination that the settlement taken as a whole is fair, reasonable, and adequate. Fed.R.Civ.P. 23(e)(2); see Staton, 327 F.3d at 972 n. 22 (court’s role is to police the “inherent tensions among class representation, defendant’s interests in minimizing the cost of the total settlement package, and class counsel’s interest in fees”); Hanlon, 150 F.3d at 1026.
The factors in a court’s fairness assessment will naturally vary from case to case, but courts generally must weigh:
(1) the strength of the plaintiffs case;
(2) the risk, expense, complexity, and likely duration of further litigation; (3) the risk of maintaining class action status throughout the trial; (4) the amount offered in settlement; (5) the extent of discovery completed and the stage of the proceedings; (6) the experience and views of counsel; (7) the presence of a governmental participant; and (8) the reaction of the class members of the proposed settlement.
Churchill Vill, L.L.C. v. Gen. Elec., 361 F.3d 566, 575 (9th Cir.2004); Torrisi v. Tucson Elec. Power Co., 8 F.3d 1370, 1375 (9th Cir.1993).
The district court considered each of these factors and found that several of them favored settlement approval: (1) plaintiffs’ case was not particularly strong in light of defendants’ significant defenses; (2) further litigation would be time-consuming, complex, and expensive for both sides; (3) the settlement provided injunctive and cy pres awards, which did not do the class any harm and which was more than plaintiffs might achieve at trial given the litigation risks; (4) the parties had already consulted experts and exchanged significant discovery permitting an informed decision about settlement; and (5) the settlement was negotiated over an extended period of time by experienced counsel on both sides, and was mediated and approved by a retired judge.
But where, as here, a settlement agreement is negotiated prior to formal class certification, consideration of these eight Churchill factors alone is not enough to survive appellate review.
Prior to formal class certification, there is an even greater potential for a breach of fiduciary duty owed the class during settlement. Accordingly, such agreements must withstand an even higher level of scrutiny for evidence of collusion or other conflicts of interest than is ordinarily required under Rule 23(e) before securing the court’s approval as fair. Hanlon, 150 F.3d at 1026; accord In re Gen. Motors, 55 F.3d at 805 (courts must be “even more scrupulous than usual in approving settlements where [947]*947no class has yet been formally certified”); Mars Steel Corp. v. Continental III. Nat’l Bank & Trust Co. of Chicago, 834 F.2d 677, 681 (7th Cir.1987) (Posner, J.) (“[W]hen class certification is deferred, a more careful scrutiny of the fairness of the settlement is required.”); Weinberger v. Kendrick, 698 F.2d 61, 73 (2d Cir.1982) (Friendly, J.) (reviewing courts must employ “even more than the usual care”); see also Manual for Complex Litig. § 21.612 (4th ed.2004). The district court’s approval order must show not only that “it has explored [the Churchill ] factors comprehensively,” but also that the settlement is “not[ ] the product of collusion among the negotiating parties.” In re Mego Fin. Corp., 213 F.3d at 458.
Collusion may not always be evident on the face of a settlement, and courts therefore must be particularly vigilant not only for explicit collusion, but also for more subtle signs that class counsel have allowed pursuit of their own self-interests and that of certain class members to infect the negotiations. Staton, 327 F.3d at 960; see also Third Circuit Task Force, 108 F.R.D. at 266. A few such signs are:
(1) “when counsel receive a disproportionate distribution of the settlement, or when the class receives no monetary distribution but class counsel are amply rewarded,” Hanlon, 150 F.3d at 1021; see Murray v. GMAC Mortg. Corp., 434 F.3d 948, 952 (7th Cir.2006); Crawford v. Equifax Payment Servs., Inc., 201 F.3d 877, 882 (7th Cir.2000);
(2) when the parties negotiate a “clear sailing” arrangement providing for the payment of attorneys’ fees separate and apart from class funds, which carries “the potential of enabling a defendant to pay class counsel excessive fees and costs in exchange for counsel accepting an unfair settlement on behalf of the class,” Lobatz, 222 F.3d at 1148; see Weinberger v. Great N. Nekoosa Corp., 925 F.2d 518, 524 (1st Cir.1991) (“lawyers might urge a class settlement at a low figure or on a less-than-optimal basis in exchange for red-carpet treatment on fees.”); and
(3)when the parties arrange for fees not awarded to revert to defendants rather than be added to the class fund, see Mirfasihi v. Fleet Mortg. Corp., 356 F.3d 781, 785 (7th Cir.2004) (Posner, J.).
B. Approval of the Settlement Agreement in the Present Case
Here, the pre-certification settlement agreement included all three of these warning signs. As discussed earlier, the settlement’s provision for attorneys’ fees is apparently disproportionate to the class reward, which includes no monetary distribution. The settlement included a “clear sailing agreement” in which defendants agreed not to object to an award of attorneys’ fees up to eight times the monetary cy pres relief afforded the class. Moreover, the settlement also contained a “kicker”: all fees not awarded would revert to defendants rather than be added to the cy pres fund or otherwise benefit the class. Confronted with these multiple indicia of possible implicit collusion, the district court had a special “obligation] to assure itself that the fees awarded in the agreement were not unreasonably high,” Staton, 327 F.3d at 965, for if they were, “the likelihood is that the defendant obtained an economically beneficial concession with regard to the merits provisions, in the form of lower monetary payments to class members or less injunctive relief for the class than could otherwise have been obtained,” id. at 964.
The Approval Order, however, does not provide adequate assurance. Rather than inquire further into why the parties had negotiated such a disproportionate distribution between fees and relief, the district court did not scrutinize the clear sailing [948]*948attorneys’ fee provision because (1) the parties claimed to negotiate the “core terms” of the settlement agreement with a neutral mediator before turning to fees, (2) the attorneys’ fee provision was severable from the agreement, and (3) the fees were to come from a separate fund and thus would have no bearing on the amount of class recovery.
But these factors did not obviate the need to examine the fee provision in light of the rest of the agreement. First, the mere presence of a neutral mediator, though a factor weighing in favor of a finding of non-collusiveness, is not on its own dispositive of whether the end product is a fair, adequate, and reasonable settlement agreement. While the Rule 23(a) adequacy of representation inquiry is designed to foreclose class certification in the face of “actual fraud, overreaching or collusion,” the Rule 23(e) reasonableness inquiry is designed precisely to capture instances of unfairness not apparent on the face of the negotiations. Staton, 327 F.3d at 960 (emphasis in original). In a Third Circuit class action settlement where class counsel provided verbal assurances that “attorneys’ fees were negotiated separately, after we agreed on everything else,” the Third Circuit refused “to place such dis-positive weight on the parties’ self-serving remarks.” In re Gen. Motors, 55 F.3d at 804; see generally Third Circuit Task Force, 108 F.R.D. at 266-70 (suggesting guidelines for mitigating the potential for collusion while still facilitating successful settlement). Similarly here, the district court should have pressed the parties to substantiate their bald assertions with corroborating evidence.
Second, the district court should not have ignored the clear sailing fee provision simply because approval of the award was not dependent on the approval of fees. “[T]he very existence of a clear sailing provision increases the likelihood that class counsel will have bargained away something of value to the class.” Weinberger, 925 F.2d at 525. Therefore, when confronted with a clear sailing provision, the district court has a heightened duty to peer into the provision and scrutinize closely the relationship between attorneys’ fees and benefit to the class, being careful to avoid awarding “unreasonably high” fees simply because they are uncontested. Staton, 327 F.3d at 954.
Furthermore, that a provision is severable does not render it irrelevant to the overall reasonableness of the agreement, for “[i]t is the settlement taken as a whole, rather than the individual component parts, that must be examined for overall fairness.... The settlement must stand or fall in its entirety.” Hanlon, 150 F.3d at 1026 (citing Officers for Justice v. Civil Serv. Comm’n of S.F., 688 F.2d 615, 628 (9th Cir.1982)) (emphasis added). By disregarding the contents of the clear sailing fee provision here, including both the disproportionate amounts negotiated and the reversionary kicker arrangement, the district court effectively “delete[d]” it from the settlement — an approach that is beyond the scope of the court’s discretion. Officers for Justice, 688 F.2d at 630 (noting that, while “the district court may suggest modifications,” it “may not delete, modify, or substitute certain provisions” of the settlement agreement but rather “must consider the proposal as a whole and as submitted”); accord Hanlon, 150 F.3d at 1026. Simply put, a severable clause simply may not be severed from the court’s Rule 23(e) analysis.
Finally, “[t]hat the defendant in form agrees to pay the fees independently of any monetary award or injunctive relief provided to the class in the agreement does not detract from the need carefully to scrutinize the fee award.” Staton, 327 F.3d at 964. Even when technically funded separately, the class recovery and the [949]*949agreement on attorneys’ fees should be viewed as a “package deal.” Johnston, 83 F.3d at 245-46. Although we do not go so far as to hold that the district court must treat the package as a constructive common fund for purposes of analyzing the reasonableness of the fee award, assessment of the settlement’s overall reasonableness must take into account the defendant’s overall willingness to pay. Ordinarily, “a defendant is interested only in disposing of the total claim asserted against it,” and “ ‘the allocation between the class payment and the attorneys’ fees is of little or no interest to the defense.’ ” Staton, 327 F.3d at 964 (quoting In re Gen. Motors, 55 F.3d at 819-20). A district court therefore must ensure that both the amount and mode of payment of attorneys’ fees are fair, regardless of “whether the attorneys’ fees come from a common fund or are otherwise paid.” Zucker, 192 F.3d at 1328 & n. 20.
For this same reason, a kicker arrangement reverting unpaid attorneys’ fees to the defendant rather than to the class amplifies the danger of collusion already suggested by a clear sailing provision. If the defendant is willing to pay a certain sum in attorneys’ fees as part of the settlement package, but the full fee award would be unreasonable, there is no apparent reason the class should not benefit from the excess allotted for fees. The clear sailing provision reveals the defendant’s willingness to pay, but the kicker deprives the class of that full potential benefit if class counsel negotiates too much for its fees. Although the district court here reasoned that class counsel’s possibility of recovering nothing in fees rendered the clear sailing provision fair, awarding class counsel nothing would not cure an otherwise unfair settlement if those funds should have been negotiated to revert to the class rather than to the “putative wrongdoer[s].” See Mirfasihi, 356 F.3d at 785. Unless the district court is able to conclude that in this particular case, a kicker provision is in the class’ best interest as part of the settlement package, the kicker makes it less likely that the settlement can be approved if the district court determines the clear sailing provision authorizes unreasonably high attorneys’ fees.
Although clear sailing provisions are not prohibited, they “by [their] nature deprivet ] the court of the advantages of the adversary process” in resolving fee determinations and are therefore disfavored. Weinberger, 925 F.2d at 525; see Malchman v. Davis, 761 F.2d 893, 907-08 (2d Cir.1985) (Newman, J., concurring), abrogated on other grounds by Amchem. Prods., Inc. v. Windsor, 521 U.S. 591, 117 S.Ct. 2231, 138 L.Ed.2d 689 (1997). Given the questionable features of the fee provision here, the court was required to examine the negotiation process with even greater scrutiny than is ordinarily demanded, and approval of the settlement had to be supported by a clear explanation of why the disproportionate fee is justified and does not betray the class’s interests. Because the district court did not provide such explanation, we must vacate the Approval Order and remand for further consideration.9
CONCLUSION
On remand, the district court may reach any number of conclusions: it may find the $800,000 attorneys’ fee award reasonable in light, of the hours reasonably expended [950]*950and the results achieved, and re-approve both orders; it may determine the fee request is excessive but find no further evidence that class counsel betrayed class interests for their own benefit, and thus uphold the agreement while lowering the fee award; or it may find the fee request excessive and conclude that class counsel therefore negotiated an unreasonable settlement and direct the parties back to the negotiating table.
In vacating the Approval and Fee Orders, we express no opinion on the ultimate fairness of what the parties have negotiated, for we have no business “substituting] our notions of fairness for those of the district judge.” Officers for Justice, 688 F.2d at 626 (internal citations omitted). Rather, we vacate and remand to allow the district court to properly exercise its discretion in accordance with the principles discussed here.
VACATED and REMANDED. Each party shall bear its own costs on appeal.