Why we wrote this

Over the past couple of years, I have talked with hundreds of claims leaders, managing partners, defense lawyers, and adjusters. Nearly all of them see that AI could help. Very few have worked out how to make it pay for everyone involved.

That isn't for lack of effort. Costs keep climbing, good people are hard to keep, and defense billing was built for a time when hours were the only practical measure of care. Adopting a technology that muddies your own economics is a lot to ask.

We built this playbook, and a set of tools to run it, as our attempt to help. It isn't the definitive answer. It is a starting point, drawn from those conversations and from what has worked in other industries, and we expect it to be argued with and improved. This version already has been: a managing partner and a head of litigation at a carrier read the first draft closely, and several of the changes below are theirs.

The short version

AI helps on a litigated file: faster drafts, more consistent summaries, fewer things missed. Under the hourly bill, though, it is hard to see who benefits. Saved time might become capacity for new matters, deeper work on hard files, or a smaller invoice. Neither firm nor carrier can easily tell which, so adoption moves slowly if it happens at all.

We propose a framework that we believe will help. Put routine work that AI speeds up on fixed fees, priced so the carrier and the firm share the time AI saves on terms they agree openly. Keep the more variable work that AI has little impact on, such as depositions, hearings, mediation, and trial, hourly. Start with one fixed-fee lane per line of business and a few outs for files that don't fit, and add routing rules only where the routine work itself grows with the case. Track a short list of measures for speed and quality, so everyone stays focused on more than the size of the legal bill. The carrier gets a smaller, steadier bill; the firm keeps enough of the saved time to earn more.

The larger prize is potentially improved indemnity, about 85 cents of every litigated claim dollar. A 3.5 percent cut in settlements saves as much as a 20 percent cut in legal fees. A pilot of 50 to 100 files can't prove that, because settlements vary too much. So the pilot promises what it can show in 90 days, which is faster evaluations, better reporting, and legal spend no higher than today, and the settlement effect is measured across the portfolio over the following year or two.

The playbook is for heads of claims and managing partners together, because neither can do this alone. It will take boldness on both sides and real partnership between client and counsel. Change like this is hard, but there are proven frameworks for it, and we used them throughout to create this playbook and our tools.

If you are already persuaded and just need to agree on terms, skip to the term sheet near the end. It covers the decisions in two pages, and the deal builder in our tool turns them into a draft both sides can mark up.

Why the hourly bill gets in the way

The defense side is short of money and people, and behind the plaintiff bar on technology. AI could help on all three, but the hourly bill makes its value hard to see.

US commercial casualty losses reached $143bn in 2023, and social inflation ran at 7 percent, its highest in two decades, according to the Swiss Re Institute. Corporate nuclear verdicts hit a record $31.3bn the following year (Marathon Strategies). The plaintiff bar, well funded and often backed by litigation finance, has been quicker to pick up new tools than the defense.

A 2025 CLM survey found that 60 percent of insurance defense and litigation firms are turning cases away for lack of capacity, and nine in ten say recruiting is harder than it was five years ago (CLM). Carriers are struggling to find experienced adjusters too, and retirements will make that worse. AI isn't replacing anyone here; there is more work than people.

Roughly 85 cents of every litigated claim dollar goes to indemnity and 15 cents to legal and adjusting expenses. Plaintiff counsel take about 40 percent of indemnity, or 34 cents of every claim dollar, more than twice the 15 cents the defense spends on its own side. Low rates and aggressive bill review have squeezed defense margins until hiring is hard.

Without a clear economic benefit to the firm, AI adoption will undoubtedly be slower than it could be. The hourly model is the main obstacle, and finding a simple, clean way to pay for this work makes the whole process much easier.

Where the money really is

Because settlements are about 85 percent of cost, a 1 percent change in settlements is worth nearly six times a 1 percent change in legal fees.

Settlements come down when files are handled well and on time, and when someone spots early that a case is going sideways. That is what AI does well. No one has fully quantified the effect at scale yet, but the anecdotal evidence is consistent: early resolution and well-handled files help reduce ultimate settlements. As adjusters say, "files don't get better with age."

A different way to bill: the weighted hybrid

Routine work that AI speeds up moves to fixed fees, priced so carrier and firm share the time saved. Work whose volume nobody can predict stays hourly. Some firms already carve out depositions, trials, hearings and mediation as hourly inside value-based arrangements, and that list is a sensible place to start. Either way, both sides know how a file will be billed before work starts. The fixed portion is large enough that a firm earns more by resolving files than by keeping them open. How to Price Legal Work in the Age of AI sets out the tiers in more detail.

Charlie Munger put the test simply: "Show me the incentive, and I will show you the outcome." Both carrier and firm need an easier working life and more money, or the program stalls.

How it fits together

Do you need routing?

A head of litigation at a carrier asked a fair question. If the fixed fee only covers routine work that AI speeds up, and the hard parts of a file are already hourly, why does routing matter at all? For many lines, you don't. One fixed-fee lane per line, plus a set number of outs, does the job.

Routing earns its place where the routine work itself grows with the case. A construction defect file with ten parties has far more records, discovery, and document chasing than a single-defendant auto claim, even if each document is simpler to handle with AI. Those are the files firms are most afraid of being burned on, and a lane for them, or a generous number of outs, makes firms more willing to take part while they get used to fixed fees.

Where routing is needed, base it on facts known at assignment: claim type, exposure band, number of parties, and venue. As an illustration, a single-defendant auto claim with modest exposure in a familiar venue might go to the standard lane, a premises claim with several parties or a difficult venue to the complex lane, and a catastrophic injury claim straight to hourly. The real rules should come from the carrier's and firm's own history, by checking which of these facts best predicted the lane each closed matter's final cost put it in.

Routing pairs naturally with an early strategy call. Within an agreed number of days of assignment, the attorney uses AI-assisted analysis to provide the adjuster with a case evaluation, a recommended path to resolution, a valuation range, and the required settlement authority. The call confirms the lane or moves the file. Once both sides know which phases the matter is likely to need, the fixed fee is easier to trust. The attorney owns the recommendation; the software helps them get there faster and supports it.

Routing won't be perfect, and it doesn't need to be. The triggers and outs below handle file changes, and each quarter both sides review how many matters stayed in their lane. When the rule keeps missing, fix the rule and leave the fee alone.

Who keeps the savings? A worked example

Take the AI-impacted work on one matter type: 100 hours at $175. Assume AI removes 40 percent of the hours and the fixed fee is 17.5 percent below the old cost. The numbers are illustrative.

The middle column looks best for the carrier, but it assumes every saved hour comes off the bill, which leaves the firm no reason to invest. The fixed fee splits the gain instead. AI frees $7,000 of time on this matter. At a fee of $14,438, the carrier keeps about $3,060 and the firm about $3,940, a split of roughly 44% to 56%. A fee 20 percent below the old cost splits it exactly in half.

Setting the firm's 38 percent higher rate per hour against the carrier's 17.5 percent saving makes the deal look lopsided, but one is a rate and the other a cost. The fair comparison is the dollars saved. Where the split lands is a negotiation. The case for giving the firm somewhat more at first is that it carries the variance of the fixed fee and does the hard work of changing how files are handled, and a bigger prize speeds adoption. The case for an even split is that the carrier still carries the indemnity risk and everything outside the lanes. Both are reasonable. What matters is agreeing it openly.

A fixed discount also has a weakness. It splits the savings fairly only when AI delivers what everyone assumed.

So we now suggest setting the fee from an agreed split of measured savings rather than a fixed discount. The pilot measures the time actually saved, and the fee follows. At the annual review, the split stays put; fees move when the work in a lane changes or measured savings differ, and they aren't reopened just to win a bigger cut.

Two more costs belong in the price. AI tools cost money, often per user or per use, so the fee should cover the committed software cost per matter at the agreed volume. And the firm's gain depends on what it does with the freed time. If staffing changes so fewer paid hours go into routine work, profit per matter rises even though revenue per matter falls. If the same people simply have less to do, the firm only wins by taking on more matters. Model both before signing.

It also starts a flywheel. Adopting firms deliver better, faster work. Carriers send more of it to the firms that show speed and quality; other firms follow, and as the gains on indemnity become clearer, that opens options for paying firms more for what they deliver on that front. The end state many managing partners describe is legal fees flat or a little lower, firm margins higher, and indemnity meaningfully lower.

How to get the incentives right and limit the risk to both sides

A fixed fee is a way of managing variance: each party should carry the variance it can control, and none that it can't. Some practical rules:

How to set prices and understand the impact

Fixed fees should come from a carrier or a firm's own invoices and costs, not from a negotiation. We built a free, secure tool that runs this analysis. Firms and carriers can use it themselves, and our team can help. Here is what it needs and what it does.

What you need to provide

1. Invoices with UTBMS task and activity codes. Two years is ideal; six months is the minimum. Firm-wide bills help, since pilot staff also work other matters.

2. Basic firm financials: a loaded cost per hour for each timekeeper level (salary, bonus, and a share of overhead), not individual salaries. The analysis runs in your browser, and your rates, costs, and margins are never sent to OraClaim or to any carrier.

3. Your AI and software costs, per user or per matter, at the volume you expect.

What the tool does with it

1. Breaks spend down by task and by matter type.

2. Estimates profitability by line of business, by biller and by matter.

3. Flags likely bottlenecks, such as partner review, associate capacity, or one slow task, and models what removing them would do to revenue and profit.

4. Groups matters into lanes using the average cost and its spread within each matter type, separating the tightly clustered work that suits a fixed fee from the variable work that should stay hourly.

5. Models a hybrid fee structure and estimates what AI does to revenue, profit and capacity under each option.

6. Builds a draft term sheet from the choices both sides make: tasks in and out of each lane, the savings split, outs, pilot scope, and billing terms.

The output frames the decisions people still have to make: what each fee covers and costs, how many lanes, how matters are routed, and which lines stay hourly. The answers vary by firm and carrier, which is why the analysis starts from each firm's own data.

Each line of business needs its own lanes. Workers' compensation, general liability, and construction defect behave differently, and in Florida, construction defect pre-suit work under Chapter 558 is a natural lane of its own. The 85/15 split between indemnity and legal spend is a liability figure and doesn't carry over to comp, so price comp from its own data. TPAs and captives have different incentives from carriers, and a program built for one may need adjusting for the others.

The same history can test the routing rules. Looking back at closed matters, check whether the facts known at assignment would have placed each one in the lane its final cost suggests. If they would have, the rules are ready to use. If not, look for the fact that would have flagged the difference earlier, such as venue or the number of parties.

For heads of claims

Ideally, carriers lead. They fund the system and set its rules, so a carrier-led program gives firms the confidence to invest.

Carriers want files handled well, on time, consistently, within guidelines and efficiently. AI helps with all of these, and a program framed around all of them gets far more buy-in from panel counsel than one about legal spend alone.

1. Frame the program around better results across the board, with lower legal spend as one goal among several.

2. Choose a few lines of business with steady, repeatable work and invite around five firms. That is enough variety to learn from and small enough to manage. Keep the firms concentrated in those lines so each line has several firms doing comparable work. That makes the comparison fair, and it lets pricing benchmarks be pooled without exposing any one firm's rates. Have counsel review how benchmarks are pooled before anything is shared, since pricing information shared among competing firms raises antitrust questions and the federal agencies withdrew their old safe-harbor guidance on such exchanges in 2023.

3. Define the pilot cohort up front: which claim types, exposure bands, jurisdictions, and litigation stages are in; a mix of new assignments and open files; and roughly 10 to 20 files per firm. Set aside comparable files handled the usual way as a control group, so results are judged against like-for-like matters as well as history.

4. Agree any routing rules with the pilot firms, and ask for an early strategy call on every pilot file so your adjusters see a supported recommendation, a valuation range and an authority request within the first few weeks.

5. Commit volume. Fixed fees only work across a portfolio, so guarantee pilot firms a defined share of assignments for at least a year. It is the hardest thing you'll be asked for, and it carries more weight spread across more than one line.

6. Share outcome data: closed-file indemnity, reserve history and cycle times, so the baseline belongs to both sides.

7. Adapt guidelines for fixed work: no line-by-line auditing inside a lane, a clear statement of the AI use you expect and how to disclose it to you and to the insured, and confidentiality and security standards firms can meet without building their own models (see our Defense Lawyer's Cybersecurity Guide to AI).

8. Review results each quarter against each firm's baseline, using the scorecard later in this playbook.

9. Work out with your billing team in advance how fixed-fee invoices will clear your e-billing rules, so the first invoices don't get kicked back.

10. Route more work to the firms that deliver, and after a year of data consider an outcome incentive with a collar, paid for resolving files at fair value.

Two things help. Share some of the early savings with firms instead of banking all of it as a rate cut. And bring your adjusters along: if the program saves them time on invoice review and gives them earlier recommendations, it sticks (see "The Rider, the Elephant, and the Path").

For managing partners

For a firm, this is a pricing and capacity decision. One way to sequence it:

1. Use our tool to run the pricing analysis on your own data first, so you understand profitability by line, biller, and matter before any carrier conversation.

2. Pick a carrier relationship with steady volume and a claims leader open to trying something new, and a small team (a few partners, two or three associates, and a paralegal) who want to do it.

3. Price from an agreed split of measured savings rather than a fixed discount, and build in your AI tool costs. Check that your margin still improves if AI saves only half the time you expect. Unpredictable work stays hourly.

4. Lead with what you can show in 90 days: time to initial evaluation, report quality, and legal spend no higher than today. Don't open by offering a discount up front, or by warning the carrier that you'll lose money while you learn.

5. Bring the carrier a proposal: a short presentation, the fee schedule with what each lane includes and excludes, any routing rules, the results you will report, and a plain account of how you use and supervise AI. Our tool drafts each of these, and How to Get Client Buy-In for AI covers what to say and to whom.

6. On every new file, hold an early strategy call with the adjuster. Bring a case evaluation, a recommended path, a valuation range and the authority you need, built with AI-assisted analysis and checked by a lawyer. Track the time spent checking AI output, since it is part of the real cost of the work.

7. Put the freed-up time to work on additional matters. A firm that has been turning cases away can grow without hiring.

8. Replace hourly targets for staff with revenue targets paired with quality measures, give credit for fixed-fee matters resolved and for outcomes, and share the gains with the team that produced them.

9. Report results quarterly against the baseline; it is the best basis for more volume and better terms.

There is a human payoff too. Much of what wears people down in defense work is producing error-free work at volume on tight deadlines. Letting software handle routine checking and drafting, with lawyers supervising, makes the job more sustainable and the firm easier to staff (more in Why Your Best Associates Are Leaving).

The first firms and carriers to try this will make mistakes. They will also learn the most and build a competitive gap that will be hard for laggards to close as these models become more widely utilized.

Rollout: from pilot to scale

Begin with a few carriers and lines of business and a few firm teams, and widen the program only once the pilot has beaten its baseline at the three-month review. For the firm-side rollout, see How to Roll Out AI at a Law Firm in 90 Days.

Each phase ends with a checkpoint both sides must clear.

Three months is enough to judge speed, report quality and early billing. It is not enough to prove anything about settlements. Settlement values vary so much that a 3.5 percent shift in the average is invisible in 50 to 100 files; it would take thousands. So the pilot should promise the leading indicators and treat indemnity as a portfolio question. The way to measure it is file by file against the value expected at the first evaluation, which is what outcome-based metrics do. Comparing each file with its own expected value removes most of the noise that raw settlement averages carry.

One defense firm working with a carrier client has shown what those leading indicators can look like. In the first 90 days, the time to an initial evaluation fell from 28 days to 7, the carrier's quality reviewer raised its rating of the firm's reports from 2 to 4.5 out of 5, and legal spend per matter stayed in line with before. That is a far better pitch than a discount up front: faster, better work now at a similar cost, with lower settlements as the upside that follows.

Set the success thresholds before kickoff, for example a faster time to initial evaluation, better report quality, and cost per matter no higher than the control files. If they are met, widen the program in waves, with the pilot firms as references and the lane structure that worked as the default.

A short call every week or two with the carrier sponsor, the pilot firms and OraClaim keeps small problems from piling up. Use it to talk through what is working in the workflow, what isn't and what adjusters find most useful. Billing results and rates should go through OraClaim and be shared only in aggregate, so competing firms never see each other's numbers.

Run the pilot under an NDA and a short pilot agreement, with data-handling and privilege protections and a lawyer reviewing every piece of AI-assisted work.

What each side needs to launch

Keeping score

Carrier and firm review one short scorecard each quarter against the pricing baseline and the control files, and revisit fees once a year. Six measures are enough; any more and nobody reads it. During the pilot, OraClaim reports the running scorecard every week or two.

Recommendation accuracy is easy to game by setting wide ranges, which is why outcome-based metrics should reward narrow ones. AI-assisted evaluation helps here too, because the range is set at the start from comparable outcomes rather than adjusted by feel as the file develops.

Few carriers track quality and timeliness consistently, which is why their effect on indemnity is so hard to pin down. AI makes them cheap to collect.

Before launch, check that the program is simple, has clear goals, gives everyone a reason to want it to work, allows for how people react to change, and is easy to start. Fix anything that falls short.

A term sheet for those ready to start

For carriers and firms that are already on board, these are the terms to settle. Each one is a choice in the deal builder, which produces a draft both sides can mark up. It is a starting point for each side's counsel, not a finished agreement.

Where to go from here

You don't have to start from scratch. Over the past year we have built the pieces people kept asking for.

For the people side, the most practical guide we know is Chip and Dan Heath's Switch: give people a clear direction, a reason to care and an easier path. It underpins "The Rider, the Elephant, and the Path".

Tell us what works, what doesn't and what we've got wrong. No single firm or carrier will fix defense economics alone.

Further reading from Future of Claims

The longer arguments behind this playbook are in the Future of Claims newsletter.

On the OraClaim blog

Sources

• Swiss Re Institute, litigation costs and US liability claims

• Marathon Strategies, Corporate Verdicts Go Thermonuclear (2024)

• CLM, Overwhelmed and Understaffed (Law Firm Talent Survey, 2025)