Article 1.4 · Funding, Risk and the AI-Compressed Cost Base

T3PS Legal Dynamics · Series 1: The Billable Hour Under Siege ·

Written by Will Whawell. Human intelligence throughout; AI assisted with the drafting.

Fact-verified June 2026

Picture the room. Four people, four spreadsheets, four versions of the same case. The client partner has the pricing. The litigation team has the funding. The costs lawyer has the budget. The broker has the ATE. And the funder — the one writing the largest cheque — is somewhere outside the door, reading the papers after everyone has already agreed how expensive the case is going to be.

I have sat in that room more times than I can count. Everyone speaks a slightly different dialect of the same language, and nobody quite owns the number that matters. It worked, more or less, while the cost base was stable. In an AI-enabled litigation market, it stops working. The four spreadsheets stop agreeing with each other, and the disagreement is the story.

The separation was always artificial

Litigation funding, CFAs, DBAs and ATE insurance all rest on the same four-legged stool: a view of risk, cost, duration and likely recovery. AI does not touch all four legs equally. It goes straight for cost. It compresses research, document review, chronology building, disclosure analysis, first-draft correspondence, pleadings and costs reconstruction. Once those tasks stop demanding the same volume of human time, every funding model built on the old cost base starts to wobble.

This is not a narrow point about law-firm rate cards. It goes to the commercial architecture of funded litigation. Colin Chapman, who built Lotus, had a single instruction for his engineers: simplify, then add lightness. Strip out everything the car does not need, and what remains goes faster. The AI-era funder is asking the legal market to do exactly that to its cost base — and discovering that a great deal of inherited weight has nothing to do with winning the case.

The funding stack

Most substantial disputes are no longer funded by one clean mechanism. They are funded through a stack. A claimant might hold a CFA with its solicitors, a discounted CFA with counsel, ATE cover for adverse-costs exposure, and a third-party litigation funding agreement covering disbursements and a slice of solicitor fees. Sometimes a DBA. Sometimes portfolio funding. In group actions, something far more elaborate again — claimant acquisition, book-building, layered insurance, court approval and a set of commercial interests that do not always point the same way.

Each layer runs its own economics. The solicitor watches WIP, cash flow and success-fee recovery. Counsel watches brief fees, deferred fees and uplift. The ATE insurer watches merits, adverse-costs exposure and premium adequacy. The funder watches capital deployment, case duration, budget discipline, likely damages and return multiple. The client watches one number only: what is left of the recovery once everyone else has been paid.

AI moves through all of them at once. If the work to progress the claim can be done faster, the solicitor’s cost base changes. If the cost base changes, the funding requirement changes. If the funding requirement changes, the funder’s return may need recalibrating. If projected legal spend falls, the proportionality of the ATE premium and the adverse-costs exposure comes back into question. And if the client can see AI taking friction out of the process, the client will ask why the economics of the stack still look as though nothing has changed.

That is the uncomfortable question now arriving in the market. The Wolters Kluwer 2026 Future Ready Lawyer Survey found that 62% of lawyers report weekly time savings of between 6% and 20% from AI — averaging close to a tenth of the working week — and that 54% expect firms to use that efficiency either to serve more clients or to price more competitively. Read that second figure twice. More than half the profession expects the gain to show up in price. The funding stack is where that expectation gets tested first.

CFAs and the time-cost problem

Conditional fee agreements have always sat awkwardly between hourly billing and genuine risk pricing. A CFA is not really a value-based fee. It is usually a deferred or contingent version of the hourly model, with a success fee bolted on to compensate the lawyer for the risk of non-payment. That logic holds while the base costs are a fair proxy for the work actually required. It gets harder to defend when AI compresses the work but the pricing still assumes the pre-AI time profile.

Take a disclosure-heavy claim. Historically the team needed hundreds of hours to review documents, build issue-based chronologies, find the key correspondence and map the evidential gaps. Carrying that work justified the success fee. But if AI-assisted workflows cut the human effort materially, the economics shift. The firm is still carrying risk — just a different risk: less labour exposure, more technology reliance, more quality assurance, more verification, more supervision.

The success fee may still be justified. The point is not that AI removes risk; it does not. The point is that the risk being priced is no longer the same risk. And the professional answer can no longer be “we would have spent 300 hours before AI, so the uplift stands.” The better answer is that the matter has been priced by reference to risk, complexity, capital lock-up, the value of the outcome, the verification burden, and the cost of running the technology and governance that make efficient delivery possible. That is a more sophisticated conversation. It is also a more honest one.

DBAs and the recovery question

Damages-based agreements make the issue sharper still. A DBA links the representative’s fee to the damages recovered rather than the time spent. They have been permitted for contentious work in England and Wales since 1 April 2013, under the Damages-Based Agreements Regulations 2013, with the payment in non-employment civil litigation generally capped at 50% of sums recovered, inclusive of VAT and counsel’s fees but excluding other disbursements.

In theory the DBA is well suited to the AI age, because it is already outcome-linked. The client pays by reference to recovery, not units of time. If AI lets the lawyer reach the result more efficiently, the lawyer earns a better margin and the client avoids hourly exposure. But that only holds if the percentage is commercially defensible.

Where AI sharply reduces the work needed to run a category of claim, a DBA pitched at the top of the permissible range comes under pressure — even where it remains technically compliant. Clients increasingly understand that not all claims carry the same labour intensity. A high-value but document-light claim is not the same animal as a sprawling group action with millions of records, even where the damages figures look alike. AI pushes DBA pricing towards granularity. Percentage-of-recovery survives, but it has to be calibrated against more than the damages number:

—   evidential volume

—   merits complexity

—   number of parties

—   procedural uncertainty and likely interlocutory burden

—   enforcement risk

—   how suitable the matter is for AI-assisted review

—   the human specialist input genuinely required

—   expected duration, and the capital and opportunity cost of the lock-up

The crude percentage will not vanish. It will simply become easier to challenge wherever it bears no visible relationship to the work, the risk or the value.

PACCAR, and the uncertainty that will not settle

PACCAR exposed how fragile parts of the funding market’s legal architecture had become. In 2023 the Supreme Court held that a litigation funding agreement calculating the funder’s return as a percentage of damages could fall within the statutory definition of a DBA — with serious enforceability consequences if it did not comply with the DBA regime. The Civil Justice Council’s June 2025 review recommended reversing PACCAR and bringing in light-touch statutory regulation of third-party funding: capital adequacy, limits on funder control, conflict provisions, and added protection in consumer, group and collective proceedings.

Here is where the draft this article grew from needs updating, because the position has moved again. On 17 December 2025 the Government confirmed it would legislate to reverse PACCAR and put beyond doubt that LFAs are not DBAs — but with prospective effect only, and with no firm timetable beyond the familiar formula that legislation will come “when parliamentary time allows.” Then the 2026 King’s Speech, delivered on 13 May 2026, came and went without a litigation funding bill in it. The commitment stands; the legislative slot does not. PACCAR remains live law.

For pricing, that limbo matters more than a clean reversal would. Uncertainty raises scrutiny. A sophisticated funder is no longer only asking whether a claim has merit. It is asking whether the cost base is credible, whether the budget reflects how work is actually delivered in 2026, whether the team has the operational discipline to hold spend, and whether the projected return survives a tougher set of efficiency assumptions. A post-PACCAR market — if and when it arrives — will not be a return to the pre-PACCAR world. It will be more regulated, more transparent and more operationally demanding. AI accelerates every part of that shift.

The funder’s real question

The question a serious funder now asks is not “what will this litigation cost if it is run traditionally?” It is “what shouldthis litigation cost if it is run intelligently?” Those are very different questions, and the gap between them is exactly the weight Chapman would have told you to remove.

A firm seeking funding for a major claim now has to justify its budget by reference to current capability, not historic habit. If the firm has AI-assisted disclosure tools, why does the disclosure phase still assume the junior fee-earner hours of a 2019 budget? If it uses AI-assisted research, why is the research estimate unchanged? If chronology building, issue tagging and privilege review can be accelerated, where does that show up in the phase budget?

“AI output needs human review” is true, but it is the beginning of the analysis, not the end of it. The real questions are how much review, by whom, at what grade, under what protocol, and with what audit trail. Funders are commercially rational. They will not begrudge a firm a strong margin where value is delivered. They will object to funding inefficiency dressed up as professional caution.

This is where litigation project management earns its place. The funded firm needs to show more than a Precedent H. It needs a delivery model: task allocation, AI use cases, supervision points, quality gates, variance reporting, change control, and client and funder reporting. In BREW terms, the funder’s diligence lives in the B and R pillars — Business Intelligence and Revenue & Returns — and cannot be answered without the E and W pillars, Execution & Orchestration and Workflow Excellence, underneath it. Without that operational spine, a budget is just a spreadsheet wearing confidence formatting.

ATE and adverse costs

ATE insurers sit slightly apart in the structure, but the same pressure reaches them. If AI improves early case assessment, document review and merits analysis, insurers should in principle price risk more accurately — sharper premiums, fewer surprises, more disciplined monitoring as the case runs.

There is a second-order effect, though. If AI cuts the claimant’s own legal spend, it shifts the proportional relationship between own-side costs, adverse-costs exposure, premium and likely recovery. In marginal or lower-value claims, that can make viable what previously was not. In larger claims, it changes how staged and deferred premiums are justified.

The insurer’s concern will not be whether AI was used. It will be whether AI was used competently. Poor AI use manufactures adverse-costs risk of its own: defective pleadings, hallucinated authorities, missed issues, privilege errors, disclosure failures, overconfident merits. The courts have already shown, in the line of cases following the AI-citation scandals, that they will treat unverified AI-generated material as a serious professional and procedural failing. So AI reduces risk and creates risk at the same time, and the line between the two is governance.

That should make AI governance part of underwriting. Does the firm have an AI policy? Are citations verified? Are disclosure outputs sampled? Are privilege calls reviewed by qualified lawyers? Are the tools approved, secure and auditable? Has the COLP signed off the framework? Those questions are moving to sit alongside prospects of success and opponent solvency — and the direction of regulatory travel supports them. The Civil Justice Council’s consultation on the use of AI in preparing court documents closed on 14 April 2026, with a final report awaited. Its provisional shape already tells underwriters where the standard is heading: professional responsibility carried by the named legal representative for statements of case, a declaration that AI has not generated the content of trial witness statements, and an amendment to PD35 requiring experts to identify any substantive use of AI in their reports.

The client’s net recovery

The most important number in funded litigation is not the headline damages figure. It is the client’s net recovery once legal fees, funder return, premium, disbursements and irrecoverable costs have been taken out. AI makes that number impossible to look away from.

If AI compresses cost but the stack quietly absorbs the benefit, clients will notice. A claimant who recovers £10 million and watches a large share distributed between lawyers, funders and insurers may reasonably ask whether AI improved their position or merely improved everyone else’s margin. That does not mean every efficiency must pass straight to the client as a discount. Firms and funders invest in technology, training, governance and risk, and they are entitled to price for value. But they have to be able to explain how the benefit was shared. A mature AI-enabled funding proposal should show:

—   the expected traditional cost profile

—   the AI-enabled delivery model

—   the revised budget, phase by phase

—   the residual human review and supervision cost

—   the risk premium being charged

—   the projected client net recovery across realistic outcome scenarios

—   the sensitivity of that recovery to delay, partial success and adverse costs

That is where AI should take this market: not simply cheaper litigation, but more intelligible litigation economics.

The new pricing architecture

The future is not one model replacing another. It is hybrid. Hourly rates survive for genuinely unpredictable work. Phase-fixed fees grow where the process can be scoped by stage. CFAs continue where firms will share risk. DBAs hold where outcome and recovery are the truest measure of value. LFAs keep funding meritorious claims clients cannot or will not self-fund. ATE stays essential in costs-shifting litigation. What changes is that the models are forced to interact in the open. A sensible AI-era architecture looks something like this:

Litigation component	Pricing approach that fits	Why AI changes it
Early case assessment	Fixed diagnostic or capped phase fee	AI accelerates document review, chronology and merits triage
Pleadings	Phase fee with assumptions and change control	Drafting is faster; judgement and strategy still carry the work
Disclosure	Volume-based, with a complexity adjustment	AI breaks the old link between document count and human hours
Witness evidence	Fixed or phased, human-led review	AI may structure material, but the evidence must stay authentically the witness's
Expert evidence	Budgeted phase fee plus disbursement control	AI can assist analysis; independence and disclosure of its use remain critical
Trial preparation	Hybrid fee with defined scope	Advocacy strategy dominates; bundles, chronologies and issue lists can be accelerated
Funder return	Outcome and risk-based	A lower legal spend changes capital deployment and return modelling
ATE premium	Staged and risk-sensitive	Better early data should allow more dynamic underwriting

This is not theoretical. It is the direction the market is already moving, because the old single-model approach no longer explains the work.

What firms should do monthly.

Five things, in order.

—   Map where AI genuinely changes the cost base. Not where it might help one day — where it already cuts time, improves consistency, or changes the staffing assumption.

—   Rebuild litigation budgets from the task up. Precedent H should stop being a ritual document prepared for the court and become the pricing spine of the matter.

—   Write an AI-use narrative for funders and insurers. If AI is in the delivery model, explain how. If it is not, explain why not. Silence is starting to read as operational immaturity.

—   Revisit CFA and DBA assumptions. Success fees and percentages need to be defensible by reference to risk, value, complexity and capital exposure — not inherited market habit.

—   Report client net recovery plainly. In the AI era the client will not only ask what the claim is worth. They will ask who captured the efficiency gain.

That is the real pricing revolution in funded litigation. AI does not abolish risk. It does not make litigation easy. It does not remove the need for expert lawyers, costs specialists, funders, insurers or counsel. It does something more disruptive than any of that. It makes the old cost assumptions visible. And once an assumption is visible, somebody has to justify it.

Three questions to take back to your own desk

—   If a funder asked you to justify every phase of a litigation budget against your firm’s current AI capability, which of your assumptions would survive the scrutiny?

—   Are your CFA and DBA models pricing genuine litigation risk — or quietly preserving margins built on a pre-AI labour model?

—   In a funded claim, who should capture the benefit of AI efficiency: the client, the firm, the funder, the insurer — or all of them, in a structure transparent enough to show the split?

If your funding stack and your budget assumptions no longer quite agree with each other, that disagreement is worth an unhurried conversation before a funder forces it. That is what a Fika is for — an espresso, an hour, and an honest look at where the weight is.