Research the Carrier Won't Pay For

Summary judgment does not come in sizes.

Rule 56 has no small-firm edition. The movant must still show there is no genuine dispute of material fact. The nonmovant must still cite the record, because the court need consider only the cited materials. In California, the defendant must still present evidence rather than simply point to a gap. The standard does not ask who is paying for the brief. The standard does not ask how many hours were approved. The standard does not ask anything at all about the firm. It asks about the brief.

The standard is uniform. The budget is not.

This paper follows that asymmetry downstream, to see who ends up holding it.

The mechanism: research that needs permission

Insurance defense work runs on litigation management guidelines. Carriers issue them; panel firms agree to them as a condition of receiving files. Many cover rates, staffing, travel, and reporting. Many also reach legal research.

The language is not hidden. A North Carolina ethics opinion considered guidelines under which the carrier would not pay for research exceeding three hours per case without prior written approval. The Defense Research Institute's own recommended guidelines for insurers ask counsel to consult the carrier before any research project requiring more than three hours. Commentators on both sides of the tripartite relationship describe restrictions on legal research as a routine feature of these documents.

Ethics authorities have pushed back. The Montana Supreme Court held that requiring defense counsel to obtain the insurer's prior approval before undertaking research fundamentally interferes with counsel's independent judgment, and that lawyers who submit to it violate the Rules of Professional Conduct. ABA Formal Opinion 01-421 took the position that guidelines may not be allowed to interfere with the lawyer's professional judgment on behalf of the insured. Model Rules 1.8(f) and 5.4(c) say the same thing from two directions: the party who pays does not direct the judgment.

So the rule is clear. Now ask what the rule leaves standing.

Follow the discretion

"Discretion" sounds symmetrical until we ask who possesses it.

The carrier holds discretion over what gets paid. It receives predictability: a capped research line, an audited bill, a defense cost it can reserve against.

The defense firm holds the professional duty. It receives a choice. Write off the research the guideline would not have covered, or do less of it. On a panel rate, the first option is a margin problem. The second is a quality problem. Neither shows up on the invoice.

The insured holds the outcome. It receives whatever the firm could afford to do. The insured did not negotiate the guideline, rarely sees it, and is the party the independence rules exist to protect.

The court holds the standard. It receives a brief and applies the full standard to it.

Notice where each thing sits. The discretion sits with the party that does not bear the outcome. The duty sits with the party that does not control the budget. The outcome sits with the party that controls neither. And the standard sits with an institution that sees none of this, and should not have to.

A rule forbidding the carrier to direct the lawyer's judgment does not change that distribution. It relocates the pressure from a written instruction to an economic one. The carrier no longer says "do not research." It says "we will not pay for it," and lets arithmetic do the rest.

Enter the efficiency pitch

Legal AI arrives in this market with a promise: faster research, fewer hours.

Ask who captures the saving.

If research that took six hours now takes two, the carrier's natural response is not to buy better research with the four hours saved. It is to lower the approved number. Efficiency in a capped system flows up the chain, to whoever sets the cap. The firm is left doing the same work at a lower ceiling, and the pressure to cut the parts of research that are slow and invisible increases.

Which parts are those? Not the part the client can see. The part the client can see is a brief with citations in it. The parts that get cut are the ones that show up only when they fail:

• the check for controlling authority that nobody retrieved;
• the search for the adverse case that the opposition will cite;
• the test of whether a case was decided under the same standard, in the same forum, on the same kind of record;
• the element no one briefed.

A tool that makes the visible part faster, and leaves the invisible part to a budget that just shrank, does not help the small firm. It helps whoever sets the budget.

The five criteria

A small defense firm should not ask which AI tool is best. That question invites a ranking, and rankings are what vendors publish about themselves. It should ask whether a tool does the work that the budget squeezes out. Five criteria, not a leaderboard.

1. Posture filtering. Can the tool restrict research to the stage, standard, and forum of the motion, and does it tell you when a case was decided under a different one? A summary judgment case cited in a demurrer is a real case in the wrong room. The firm that cannot afford to check every citation by hand needs the tool to check it.

2. Adverse surfacing. Does the tool put the controlling case against you in front of you without being asked? A tool that returns only authority for your position is a tool that saves time by skipping the search your client most needs.

3. Completeness checks. Does the tool report, element by element, what it did not find? Silence is not a result. A tool that writes smoothly around a gap has hidden the gap from the one person who could fix it.

4. Traceable provenance. Does the tool keep a record of what was searched, what was retrieved, what was rejected, and why? That record matters to the court and to a malpractice carrier. It also matters to the firm in a way vendors rarely mention: it is evidence. A research log that shows why an additional three hours were necessary is the firm's best argument when the bill is audited. Provenance turns the firm's judgment into something the carrier has to answer.

5. A published error rate. Measured, per error class, including missed authority, by someone other than the vendor. A tool that will not publish its own error rate is asking the firm to extend the trust the firm cannot afford to extend.

Any vendor, Legawrite.AI included, should be held to all five.

Who this is really about

It would be easy to read this as a complaint about carriers. It is not. Carriers manage cost because cost is what they are accountable for. Guidelines are a rational response to that accountability.

The point is narrower and harder. When a system distributes discretion to one party, duty to another, and outcome to a third, every efficiency gain will be claimed by the party holding discretion unless someone designs against it. The small firm cannot redesign the insurance market. It can refuse to adopt tools that make its hardest-to-see work easier to cut, and it can insist on tools that make that work visible, documented, and defensible.

The carrier can decline to pay for the research.

It cannot decline the standard.

Sources: Fed. R. Civ. P. 56(a), 56(c)(3); Aguilar v. Atlantic Richfield Co., 25 Cal.4th 826, 854 (2001); In re Rules of Professional Conduct & Insurer Imposed Billing Rules & Procedures, 2 P.3d 806 (Mont. 2000); ABA Comm. on Ethics & Prof'l Responsibility, Formal Op. 01-421 (2001); N.C. State Bar, 98 Formal Ethics Op. 17; DRI, Recommended Case Handling Guidelines for Insurers (2000); Model Rules of Prof'l Conduct R. 1.8(f), 5.4(c).