POSNER, Circuit Judge.
Class actions differ from ordinary lawsuits in that the lawyers for the class, rather than the clients, have all the initiative and are close to being the real parties in interest. This fundamental departure from the traditional pattern in Anglo-American litigation generates a host of problems well illustrated by this appeal, which challenges the settlement in a class action.
In 1983, the Chicago law firm of Joyce and Kubasiak filed Tunney v. Continental Illinois National Bank in an Illinois state court. This was a class action on behalf of persons who had borrowed money from Continental at interest rates pegged to Continental’s prime rate. The complaint alleged that since 1973 Continental had defrauded (and broken its contracts with) these borrowers by failing to adhere to its agreement to charge an interest rate [679]*679pegged to the “prime rate,” which the loan agreements defined as the rate the bank charges “for 90-day unsecured commercial loans to large corporate customers of the highest credit standing.” The complaint alleged that throughout this period Continental had made loans to large corporate customers at rates well below the prime rate quoted to members of the class. In 1984 the state court judge certified the suit as a nationwide class action, appointed Joyce and Kubasiak to represent the class, and certified his certification for an interlocutory appeal. Because of the settlement negotiations described later in this opinion, the appeal was never taken; neither was notice to the class ever issued. Joyce and Kubasiak conducted little discovery, and did little other investigating, concerning the merits of the “prime rate” claim.
In 1985, Mars Steel Corporation, represented by Jerome Torshen, sued Continental in federal court in Chicago on behalf of a class defined identically to that in the Tunney suit. The only violation alleged in Mars was a violation of the RICO statute (Racketeer Influenced and Corrupt Organizations Act, 18 U.S.C. §§ 1961 et seq.), for a period beginning in 1973; treble damages were sought. Although Mars was filed after Tunney, Torshen, unlike Joyce and Kubasiak, pursued discovery on the merits. To comply with his document demands Continental developed a computer program that enabled it to discover which 90-day unsecured loans made between January 1980 and September 1982 might have been made below its prime rate. The computer search turned up 140 questionable loans, but further investigation showed that none of these was below prime within the meaning of the loan agreements with the members of the class. Some of the loans were for fewer than 90 days, some for more than 90 days; some were at fixed interest rates rather than rates pegged to the prime rate; some had not been made to corporate customers; some imposed compensating-balance requirements that had the effect of jacking up the real interest rate; and in some the interest rate was misstated because of clerical error. There is no suggestion that in making loans to large corporate customers for periods different from 90 days or at fixed interest rates rather than rates pegged to a fluctuating prime rate Continental was attempting to circumvent the terms of the loan agreements with the members of the class.
Discovery set the stage for settlement negotiations, which Continental conducted separately with Joyce and Kubasiak and with Torshen. Early in 1986 Joyce and Kubasiak offered a settlement whereby Continental would agree not to oppose a request for an award of $2 million in attorney’s fees (later reduced to $1.25 million) and the class members would be given an opportunity to take out new loans from Continental at below-market rates. Continental refused the offer, and shortly afterward settled with Torshen. Under the terms of the settlement Continental would not oppose Torshen’s request for $305,000 in fees, while the members of the class, defined as all corporate borrowers from Continental since 1973 at rates tied to the prime rate, would be entitled to take out new loans from Continental of up to $100,-000 for one year at an interest rate roughly one-half of one percent below the borrower’s previous interest rate. If (a big if) interest rates had not changed, Continental would be giving a $500 interest credit to every member of the class who wanted to borrow and met the bank’s standards of creditworthiness. If interest rates had risen, the class members would do better than this; if rates had fallen, they would do worse. Since there are 23,000 class members, the maximum value of the settlement to them if interest rates have not changed is $11.5 million.
The proposal by Joyce and Kubasiak would have entitled most class members to borrow up to $200,000 for up to one year at an interest rate one percent below the average interest rate that Continental had charged the borrower during the complaint period (i.e., since 1973). Class members who had borrowed more than $200,000 from Continental during that period would be allowed to borrow up to an amount equal to one half of their largest loan; this one half might be more or less than $200,-[680]*680000. How favorable this offer is to the class, compared to Torshen’s, depends primarily on the relationship between current interest rates and those prevailing during the complaint period (not just, as with Torshen’s offer, on the relationship between current interest rates and the interest rates paid by members of the class most recently). As the record does not contain the necessary data for estimating the actual value to the class of either settlement, there is no reason to suppose either that Joyce and Kubasiak’s offer is “sweeter” than Torshen’s or that the reverse is true.
Without holding an evidentiary hearing the district judge gave preliminary approval to the Mars settlement, at the same time certifying the suit as a class action for settlement purposes only. This mode of class certification, not expressly provided for in Rule 23, is perhaps best interpreted as tentative certification. See In re Beef Industry Antitrust Litigation, 607 F.2d 167, 177-78 (5th Cir.1979). The judge also ordered that notice of the class action and of the settlement proposal be disseminated to the class by mail and publication. Only two members of the class objected to the settlement — one being Tunney, of course. Wiesbrook, another member of the class and a client of Joyce and Kubasiak’s, opted out of the settlement in order to keep the state court suit alive.
The district court then held a “fairness” hearing and decided that the settlement was fair and approved it, thus extinguishing by operation of res judicata the claims of all class members who had not opted out. Only 1.5 percent of the class members had opted out, a surprisingly small fraction if the settlement is as bad as Joyce and Kubasiak argues. Cf. Wellman v. Dickinson, 497 F.Supp. 824, 833 (S.D.N.Y. 1980). Although not all members of the class are corporations, all (we surmise) are business borrowers rather than consumers and all must have known — the notice clearly told them — that, if the settlement went through, it would extinguish their legal rights against Continental.
Tunney appeals, charging that the class should have been certified long before the settlement was given even preliminary approval, that the settlement is unfair, that the class notice was inaccurate and misleading, and that he (realistically, Joyce and Kubasiak) was denied an opportunity to present evidence of the settlement's unfairness either before preliminary approval of the settlement or at the fairness hearing.
Rule 23(c)(1) of the Federal Rules of Civil Procedure provides that, “As soon as practicable after the commencement of an action brought as a class action, the court shall determine by order whether it is to be so maintained.” Mars, filed in 1985, was not certified as a class action until 1986 (and then only for purposes of settlement), when the district judge, before notifying the members of the class but after considering objections to the proposed settlement, gave the settlement his preliminary approval. It is hard to see why the propriety of maintaining the suit as a class action could not “practicably” have been determined much earlier. And, common though the practice of deferring class certification while settlement negotiations are going on is, it not only jostles uneasily with the language of Rule 23(c)(1) but also creates practical problems. Settlement negotiations are made more complicated when the parties don’t know whether they are trying to settle a class action or an action limited to the named plaintiffs, don’t know whether the named plaintiffs would be deemed adequate representatives of the class if the case proceeded to trial, and don't know the composition and size of the class. The danger of a premature, even a collusive, settlement is increased when as in this case the status of the action as a class action is not determined until a settlement has been negotiated, with all the momentum that a settlement agreement generates — if its class status is ever determined, for Judge Aspen never did make a formal determination that this class action could be maintained consistently with the criteria of Rule 23. And where notice of the class action is, again as in this case, sent simultaneously with the notice of the settlement itself, the [681]*681class members are presented with what looks like a fait accompli. Moreover, the delay may make opting out of the class action less feasible than if members of the class had received prompt notice, not because the statute of limitations will have run (the filing of the suit will have stopped it from running) but because evidence may be growing stale.
Mainly for these reasons the practice of deferring class certification until a settlement has been negotiated has been criticized, see, e.g., McDonald v. Chicago Milwaukee Corp., 565 F.2d 416, 420 (7th Cir. 1977); Manual of Complex Litigation 59-61 (5th ed. 1982); cf. Premier Electrical Construction Co. v. National Electrical Contractors Ass’n, Inc., 814 F.2d 358, 363 (7th Cir.1987); Watkins v. Blinzinger, 789 F.2d 474, 475 n. 3 (7th Cir.1986), yet every court that has addressed the question, including our own, has declined to hold that the procedure is a per se violation of Rule 23 and therefore requires that the settlement automatically be disapproved. See, e.g., Simer v. Rios, 661 F.2d 655, 664-66 (7th Cir. 1981); McDonald v. Chicago Milwaukee Corp., supra, 565 F.2d at 420; Glidden v. Chromalloy American Corp., 808 F.2d 621, 627 (7th Cir.1986); Weinberger v. Kendrick, 698 F.2d 61, 72-73 (2d Cir.1982) (Friendly, J.); In re Beef Industry Antitrust Litigation, supra, 607 F.2d at 173-78. Two considerations explain this result. The first, stressed in Simer, is that since a suit begun as a class action may often and quite properly be settled as an individual action, that is, without preclusive effect on other members of the class, it would be a waste of time and money if before such a settlement could be negotiated a class-certification proceeding, possibly entailing the issuance of notice to the members of the class, had to be conducted in every case. (Of course, notice was sent in this case, and this case was settled as a class action.) Against this it can be argued that plaintiffs will be tempted to add class claims in order to intimidate the defendant, then delete them by way of compromise. But second, the procedure creates only a possibility, not a certainty, of abuse, a possibility held in check by the requirement that the judge determine the fairness of the settlement before he can approve it.
Simer and Weinberger emphasize, consistently with the last point, that when class certification is deferred, a more careful scrutiny of the fairness of the settlement is required. We agree, and that brings us to the second issue presented by the appeal, the fairness of the settlement. The fairness of a settlement of a legal dispute is like the adequacy of the consideration supporting a contractual promise: a matter best left to negotiation between the parties. A settlement is a contract, and normally the test for the fairness of a contract is strictly procedural: were the parties competent adults duly apprised of the basic facts relating to their transaction? The problem in the class-action setting, and the reason that judicial approval of the settlement of such an action is required, see Fed.R.Civ.P. 23(e), is that the negotiator on the plaintiffs’ side, that is, the lawyer for the class, is potentially an unreliable agent of his principals. See, e.g., Dam, Class Actions: Efficiency, Compensation, Deterrence, and Conflict of Interest, 4 J. Legal Stud. 47 (1975); Rosenfield, An Empirical Test of Class-Action settlement, 5 J. Legal Stud. 113 (1976). Ordinarily the named plaintiffs are nominees, indeed pawns, of the lawyer, and ordinarily the unnamed class members have individually too little at stake to spend time monitoring the lawyer — and their only coordination is through him. One solution, illustrated by the events narrated above, is competition by other lawyers to represent the class. It is only a partial solution, because a lawyer may be able to obtain a large fee not just by outbidding another class lawyer but, alternatively, by “selling out” the class, a danger discussed in In re General Motors Corp. Engine Interchange Litigation, 594 F.2d 1106, 1125 (7th Cir.1979). Joyce and Kubasiak hints that Torshen sold out the class to Continental (which was eager to buy) in exchange for a $305,000 mess of potage. The danger of collusive settlements — vividly described many years ago by Justice Jackson in Cohen v. Beneficial Industrial Loan Corp., [682]*682337 U.S. 541, 549-50, 69 S.Ct. 1221, 1227, 93 L.Ed. 1528 (1949), and rendered much greater than in the ordinary litigation by the tenuousness of the control exerted by the client (principal) over the lawyer (agent) —makes it imperative that the district judge conduct a careful inquiry into the fairness of a settlement to the class members before allowing it to go into effect and extinguish, by the operation of res judicata, the claims of the class members who do not opt out of the settlement. See generally Shelton v. Pargo, Inc., 582 F.2d 1298 (4th Cir.1978). The inquiry must as we have said be especially careful and penetrating in a case such as this where class certification is deferred to the settlement stage. But since so many imponderables enter into the evaluation of a settlement, we shall continue to review the district court’s decision to approve a settlement in such a case under the deferential “abuse of discretion” standard. See, e.g., EEOC v. Hiram Walker & Sons, Inc., 768 F.2d 884, 890 (7th Cir.1985); Curtiss-Wright Corp. v. Helfand, 687 F.2d 171, 175 (7th Cir.1982).
A settlement is fair to the plaintiffs in a substantive sense (we examine the procedural fairness of the settlement later) if it gives them the expected value of their claim if it went to trial, net of the costs of trial (minus the costs of settlement, but we can disregard that detail). In re General Motors Corp. Engine Interchange Litigation, supra, 594 F.2d at 1132 n. 44. Suppose the claim of the class in this matter is worth $750 million, but the probability that the class would prevail if the case were tried is only one percent and if it did prevail it would have to pay a contingent fee of $10 million. Then assuming risk neutrality (meaning indifference between a sum certain and its uncertain equivalent — e.g., between $2 and a 10 percent chance of $20), the class would be better off settling for any amount greater than $7.4 million than taking its chances on a trial. And if the members of the class were risk averse they might consider themselves better off even if the settlement were much smaller.
Joyce and Kubasiak argues that a trial might result in a judgment for the class of anywhere from $750 million to $1.5 billion, but it has not established the realism of this projection and the judge was not required to take it at face value. Continental’s computer study, though admittedly limited to only a portion of the complaint period, reveals no loans below Continental’s prime rate. If the results of that study can be generalized to the rest of the complaint period, not only was there no fraud or breach of contract, there were no damages. Furthermore, although many “prime rate” cases have been brought against banks in recent years, none has resulted in a victory at trial for the plaintiffs and apparently none in a settlement significantly (if at all) more favorable to the plaintiffs than the Mars settlement. See, e.g., NCNB National Bank v. Tiller, 814 F.2d 931 (4th Cir. 1987); Haroco, Inc. v. American National Bank & Trust Co., 662 F.Supp. 590 (N.D.Ill.1987), vacated upon grant of motion of reconsideration; Kleiner v. First National Bank, 581 F.Supp. 955 (N.D.Ga.1984). As Kleiner notes, a prime, rate is merely a bank’s forecast of what it would charge its most creditworthy corporate customers for a 90-day unsecured loan. It is not an actual transaction price, because the computation of such a price — requiring, as it would, averaging interest rates across numerous loans made at different times on different terms (e.g., compensating balances) — would be infeasible. See id. at 958-60. Reasonable pretrial discovery conducted by Tor-shen brought to light no evidence that the forecasts that Continental used in deciding how much interest to charge members of the class were not good-faith estimates of what Continental would charge its most creditworthy customers for a 90-day unsecured loan.
Hence if this case went to trial (assuming it could surmount a summary judgment motion by the bank) the prospects for the plaintiffs would be very dim. In these circumstances a settlement possibly worth as much as $11.5 million to the plaintiffs (conceivably even more), net of attorneys’ fees and other costs of suit and without the delay of litigation, seems generous — and certainly adequate, as the district judge [683]*683found. True, the settlement extinguished the claims of the Tunney class; but almost no members of that class objected, for the excellent reason that the settlement gave the class more than it could realistically have hoped to obtain either by litigation or by separate settlement negotiations — given the underlying weakness of the claim, the lack of vigorous prosecution of Tunney, and Joyce and Kubasiak’s preoccupation with attorneys’ fees. The fact that Mars charged only a RICO violation would be relevant if the omission of other theories had caused Torshen to scale down his settlement demand; the settlement would not be fair if it yielded the plaintiffs less money than they could reasonably have hoped to obtain in the parallel state court suit, where other theories had been advanced. But the amount of the settlement negotiated by Torshen was reasonable not merely in relation to the RICO claim but in relation to any theory of Continental’s liability, for any such theory depended on proof of violations of the prime-rate loan agreements, and it appears there were no such violations.
We turn to the procedural challenges to the settlement, beginning with the adequacy of the class notice. If as Joyce and Kubasiak charges the notice was misleading, the paucity of opt-outs may have misled the judge concerning the strength of the support in the class for the settlement. The focus of the charge is paragraph 7 of the notice, which describes the Tunney suit in unflattering terms, stating that (1) Joyce and Kubasiak had “made a non-negotiable demand for $1,250,000 in attorneys’ fees, more than four times the maximum amount potentially to be paid to counsel representing Mars Steel and the class in this case,” (2) “a purported class in the Tunney case has no members since that class has never been given notice or an opportunity to opt-out or to object to the adequacy of representation,” (3) “the attorneys representing the plaintiffs in Tunney do not adequately represent the class,” and (4) “should you choose to opt-out of the Class in this case and reject the settlement benefits, and if the Tunney case proceeds as a class action and those plaintiffs prevail, you may at some unknown future time be able to receive benefits from the Tunney case.” In objecting to statement (1), Joyce and Kubasiak points out that no demand for an award of attorneys’ fees in a class action is “non-negotiable,” since the judge has to approve the award. However, it is a fair inference from the negotiations that Joyce and Kubasiak would not agree to a settlement with Continental that provided for attorneys’ fees of less than $1.25 million. Thus the demand really was a stumbling block to negotiations and this was a relevant point to make in the notice, since if the Tunney suit was never settled the prospects that the class members would ever get anything were remote. Statement (2) is potentially misleading. The fact that notice had not been sent in Tunney didn’t mean that the Tunney class had no members. Nevertheless a careful reader of the statement would understand it to be saying that the class had no members in the sense that no one had been given notice of the class action and an opportunity — which most or all might take —to opt out of it. And this was true; no one had been given notice and an opportunity to opt out. Statement (3) is a fair inference from the failure of Joyce and Kubasiak to conduct discovery on the merits in Tunney, as well as from the firm’s dogged insistence on negotiating attorneys’ fees über alies. Statement (4) is unquestionably correct.
Taken as a whole the notice is not seriously misleading (though standing alone statement (2) might well be thought so) and does not invalidate the approval of the settlement. In so concluding we do not recede from our emphasis in General Motors on the importance of procedural regularity in the management of class actions. The agency problems which these actions create require that the district judge be vigilant in protecting the procedural rights of class members.
Last, a variety of procedural rulings are challenged, including the district judge’s refusal to hold an evidentiary hearing before giving preliminary approval to the settlement. These rulings are said to [684]*684have had the cumulative effect of preventing Joyce and Kubasiak from conducting discovery of Continental’s settlement negotiations with Torshen — discovery that Joyce and Kubasiak argues would have shown that the settlement lacked merit— and from introducing evidence that the settlement was disadvantageous to the class. However, the challenged rulings were within the district judge’s broad discretion in managing a class action. Discovery of settlement negotiations in ongoing litigation is unusual because it would give a party information about an opponent’s strategy, and it was not required in the circumstances of this case. Suppose Joyce and Kubas-iak, allowed to discover the details of Continental’s negotiations with Torshen, had found out that Continental had acknowledged certain weaknesses in its defense; Joyce and Kubasiak could have used that information to drive a harder bargain with Continental or, if settlement negotiations had broken down, to undermine Continental’s defense at trial. Such discovery is proper only where the party seeking it lays a foundation by adducing from other sources evidence indicating that the settlement may be collusive, as in the General Motors case, where negotiations with one class counsel were carried out in violation of the district court’s order. See 594 F.2d at 1126. There is no indication of such hanky-pank here. Nothing in the terms or timing or other circumstances of the Mars settlement — a settlement highly favorable to the class, as we have said — suggests that Torshen was selling out the class in an effort to beat Joyce and Kubasiak to the attorney’s fee trough. Indeed, Torshen’s fee seems modest in relation to the benefits to the class in a lawsuit so thin that probably it should never have been brought, while Joyce and Kubasiak’s equity in requesting a discovery order is weak in light of its failure to pursue discovery in the state court proceeding. Cf. Weinberger v. Kendrick, supra, 698 F.2d at 79.
The temptation to convert a settlement hearing into a full trial on the merits must be resisted. Airline Stewards & Stewardesses Ass’n, Local 550 v. American Airlines, Inc., 573 F.2d 960, 963-64 (7th Cir. 1978) (per curiam). Yet that is essentially what Joyce and Kubasiak is seeking. It hopes to unearth evidence showing that, contrary to what the discovery conducted by Torshen revealed, Continental really did violate the prime-rate loan agreements. The time for that discovery was in 1983 when Tunney was filed.
It is true that the entire issue of discovery arises only because Continental negotiated separately with Torshen and with Joyce and Kubasiak, so that Joyce and Kubasiak was not privy to the negotiations with Torshen. But three-cornered negotiations are clumsy at best, especially when one of the corners (Joyce and Kubasiak) adopts an obdurate negotiating position. Rather than attempt to prescribe the modalities of negotiation, the district judge permissibly focused on the end result of the negotiation, which was more favorable to the class than any class member could reasonably have expected. The proof of the pudding was indeed in the eating.
The excluded evidence was conclusional testimony, of little value at best, that the settlement was inadequate, offered in the main by persons having financial or professional relationships with Joyce and Kubas-iak. The district judge did not abuse his discretion in excluding this evidence as having little or no probative value.
Given the momentum that, as we noted earlier, a settlement agreement acquires once it has been negotiated, the district judge might have been well advised to hold a brief evidentiary hearing before giving preliminary approval of the settlement and issuing notice of it to the class. But there is no ironclad requirement of such a hearing, see In re General Motors Corp. Engine Interchange Litigation, supra, 594 F.2d at 1133, and its omission in this case was not a reversible error. Such a hearing would not have brought out more facts regarding the settlement’s possible shortcomings than were brought out in the evidentiary hearing conducted before the settlement was finally approved, so the judge would surely have been persuaded to grant the preliminary approval. In this respect the present case resembles our oth[685]*685erwise unrelated decision in Lossman v. Pekarske, 707 F.2d 288, 290-91 (7th Cir. 1983). A different conclusion would be possible, however, if the case for final approval of the settlement were weaker than it was in this case and it appeared that the final approval was due to the momentum generated by a preliminary approval based on inadequate evidence.
The settlement was fair, both substantively and procedurally, and the order of the district court affirming it is therefore
Affirmed.