Article 4.4 · The Costs Judge Meets the Machine

T3PS Legal Dynamics · Series 4: Practical AI for Legal Operations · Fact-verified June 2026

Every costs consultant/costs lawyer knows the file that arrives like an archaeological dig. The matter settled six months ago. The partner who ran it has moved on. The time narratives read “review file” and “consider documents,” the attendances do not obviously match the procedural events, and somewhere in there is a counsel fee note nobody can quite account for. The job is no longer to assess the costs. It is to reconstruct them — to impose, after the event, a structure on a matter that was never managed structurally in the first place.

There is a strange discontinuity in civil litigation. During the life of the case, everyone talks budgets, proportionality, phases, incurred and estimated costs, assumptions, contingencies, cost management. Then judgment lands, or settlement is reached, or the costs order is made — and the process falls backwards into that dig. Time entries reviewed. Narratives rewritten. Documents matched to phases. Counsel’s fees traced. Duplications defended. The bill becomes a retrospective attempt to make a managed-looking story out of an unmanaged matter. It is inefficient. More to the point, it is now unnecessary.

If AI can build litigation budgets, track live phase spend, flag variance and generate client reporting during the case, it can also transform what happens after: bill preparation, points of dispute, replies, negotiation, provisional and detailed assessment. The profession has treated costs assessment as a postscript for years. AI makes that impossible. Costs are not the afterlife of litigation. They are the audit trail of how the litigation was run.

The post-judgment blind spot

Article 4.2 argued that Precedent H should be treated not as a spreadsheet but as a project-management framework. That argument is at its strongest at detailed assessment. The costs judge is not interested in the romance of the case. The court cares about recoverability, reasonableness, proportionality, necessity, hourly rates, phase allocation, duplication, incurred versus budgeted spend, and the relationship between the work done and the result achieved. The paying party hunts for attack points. The receiving party defends the bill. The client, often, simply wants to know why the argument about costs is still running.

Too much of this rests on imperfect data. Time recording was not phase-aligned. Narratives were written for billing, not recovery. Work was spread across fee earners without clean task allocation. Internal duplication was never captured contemporaneously. Counsel’s involvement was not mapped to assumptions. The budget was set at one stage, the matter evolved at another, and the bill at the end tries to reconcile all of it after the fact. That is the blind spot: the costs phase exposes whether the litigation was managed properly from the beginning. AI does not fix that by magic. It fixes it by making the structure visible earlier, while it can still change behaviour.

The bill is a dataset

The electronic bill of costs was an important step, but it did not go far enough — it made the bill more structured without necessarily making the underlying matter more intelligible. A bill is not just a document. It is a dataset. Every entry tells a story: who did the work, when, how long it took, which phase it belonged to, which task it advanced, what rate applied, what event it related to, and why it should be recoverable. The trouble is that most firms still assemble that dataset far too late.

AI changes the economics of doing it properly. During the matter, AI can help classify time entries by phase, task and activity; flag vague narratives before they harden into unrecoverable problems; identify duplication between fee earners; match work to procedural milestones; track counsel and expert spend against assumptions; compare incurred against budgeted cost; alert the team as a phase nears budget exhaustion; generate variance explanations while the reason is still fresh; and build a matter chronology linked to costs events. None of that removes the need for costs expertise. It makes costs expertise available earlier — when it can influence the matter rather than merely repair the record. That is the central shift: AI moves costs from reconstruction to management.

Points of dispute, as pattern recognition

The paying party’s task in detailed assessment is largely pattern recognition. Where are the excessive attendances? The duplications? Which hourly rates are exposed? Which entries are vague? Where has work been pushed into the wrong phase? Where does the bill look disproportionate to the issues or the result? Where do counsel’s fees sit out of line? Where is the disconnect between budget assumption and claimed spend? This is exactly the work AI assists. A well-designed costs tool can review a bill and produce a first-pass issue map — rate challenges by grade and period, phase-level overspend, repeated internal communications, excessive attendance patterns, unexplained partner involvement, multiple fee earners at the same event, vague entries, work apparently outside the approved budget, counsel and expert anomalies, time spikes around procedural events, entries needing documentary support.

The costs consultants/costs lawyers still decides what matters. The solicitor still instructs. The paying party still has to plead proper points of dispute. But the first pass is faster, more consistent and more complete. And the same logic runs in reverse: a receiving party can have AI identify the likely points of attack before the bill is served — where narratives need clarifying, where evidence should be gathered, where proportionality needs explaining, where a concession may be commercially sensible. This is not automating advocacy before a costs judge. It is making the preparation better.

Replies and negotiation: separating signal from ritual

Costs negotiation has a ritual quality. The bill is served. Points of dispute arrive. Replies are prepared. Positions harden. Some objections are serious, some tactical, some boilerplate, some plainly copied from the last case. AI helps separate the real dispute from the noise. A system working from the bill, budget, order, assumptions, costs-management history, correspondence and points of dispute can triage the objections:

Objection type	AI-assisted analysis	Human decision
Hourly rate challenge	Compare claimed rates to guideline, approved or agreed rates	Concede, defend or evidence
Duplication	Identify overlapping attendance entries	Was the duplication justified?
Proportionality	Compare claimed spend to value, complexity and outcome	Frame the proportionality argument
Phase overspend	Map spend to approved budget and assumptions	Show good reason, or concede
Vague entries	Highlight weak narratives	Amend, evidence or abandon
Counsel fees	Compare to scope, brief, hearing length and budget	Justify or negotiate
Expert fees	Link invoices to reports, meetings and procedural need	Evidence reasonableness

That table is not a substitute for expertise. It is a triage framework. The benefit is speed: the costs lawyer spends less time finding the dispute and more time judging its strength — and detailed assessment is a negotiation exercise before it is a judicial one. The party that understands the bill better, and earlier, holds the advantage.

Proportionality becomes visible

Proportionality is where the AI-enabled costs environment gets genuinely interesting. Historically the argument is blunt. The paying party says the total is too high. The receiving party says the work was necessary. The court weighs the sum claimed, the sums in issue, complexity, conduct and the wider circumstances — everyone arguing from a mixture of principle, instinct and selective detail. AI makes it granular. It can show where the cost accumulated — disclosure, witness evidence, expert evidence, interlocutory applications, trial prep, settlement activity. It can distinguish unavoidable complexity from inefficient process, and show whether a disproportionate total came from one exceptional event or from persistent low-level inefficiency throughout. That matters because proportionality should not be a blunt postscript. It should be a live management discipline. If a phase is turning disproportionate while the case is running, the client should know, the funder should know, the solicitor should know, and the court may need to. Waiting until detailed assessment to discover it is the precise opposite of litigation project management.

The receiving party who builds the record

The biggest near-term advantage belongs to receiving parties who structure their matters from day one. A firm that records time accurately, tags work by phase, maintains budget assumptions, captures reasons for variance, stores counsel fee notes cleanly, links documents to procedural events and records client decisions contemporaneously will be in a far stronger position at assessment — and AI makes that discipline both easier to maintain and easier to evidence. The firm that does not do this faces a different future. Picture a paying party able to identify, within hours, every duplicated attendance, every vague entry, every phase mismatch, every unexplained spike, every internal-communication pattern that smells of over-lawyering — while the receiving party takes weeks to reconstruct why those entries exist. That imbalance will be commercially painful. The answer is not to fear AI-assisted challenge. It is to build an AI-ready costs record.

This is not the death of the costs lawyer

Quite the opposite. AI increases the value of genuine costs expertise because it increases the volume and quality of analysable material. Someone still has to know what matters — CPR, the practice directions, guideline rates, the budgeting case law, proportionality, solicitor-client retainers, inter partes recovery, the indemnity principle, counsel fee treatment, VAT, disbursements, Part 36 consequences, and the plain psychology of costs negotiation. What changes is the labour mix. The low-value work of extracting, sorting, matching and first-pass identifying becomes more automated. The high-value work — strategy, judgement, negotiation, advocacy preparation, risk assessment — becomes more prominent. That is good news for serious costs professionals. It moves them closer to the centre of litigation strategy, where costs advice shapes the event rather than arriving after it.

Risks, limits and the governance line

There are obvious dangers. An AI tool can misclassify work, misread context, miss a procedural nuance, overstate duplication, understate strategic necessity, or treat two superficially similar entries as the same when they are not. It can also create false confidence — a beautifully formatted analysis is not necessarily a correct one. So AI-assisted costs work is governed by the principle that runs across all legal AI: human responsibility is non-delegable. The SRA’s position is that existing professional duties apply in full to AI-assisted work, and accountability cannot be handed to an IT team, a vendor or an outsourced provider. The Civil Justice Council’s consultation on AI in court documents, which closed on 14 April 2026, points the same way — AI use is not prohibited, but named responsibility, verification and evidential integrity remain central. The line of cases following the AI-citation scandals, from Ayinde onward, shows the courts will treat unverified AI-generated material as a serious professional and procedural failing; a fresh example surfaced as recently as June 2026, so this is a live and lengthening list, not a settled one.

Costs work also carries acute confidentiality and data concerns. Bills, narratives, retainers, privileged strategy material, settlement discussions, internal correspondence — all sensitive. A firm cannot feed that into a public AI system and hope for the best. Vendor due diligence, data-processing terms, access control, deletion rights, audit logs and confidentiality safeguards matter as much here as in disclosure or drafting. Arguably more: the costs file often holds the hidden operational story of the whole litigation.

What should change now

Do not wait for a mature “AI costs platform” before improving costs operations. The essential changes are process first, technology second. Every matter should have phase-aligned time recording from the outset. Every budget assumption should be stored so it can be retrieved and compared against what actually happened. Every material variance should be explained contemporaneously. Counsel and expert costs should be linked to phases and events. Client instructions on scope, compromise and escalation should be recorded clearly. Proportionality concerns should be flagged while the matter is live. AI can assist with all of it — but it cannot conjure discipline from chaos. The operating model is simple to state:

—   Budget the matter structurally.

—   Record work structurally.

—   Monitor variance structurally.

—   Explain change structurally.

—   Prepare the bill from structured data.

—   Assess the bill using structured analysis.

—   Negotiate from structured insight.

That is the end of retrospective cost guesswork.

Three questions to take back to your desk

—   If your firm had to prepare a bill of costs tomorrow, how much of the work would be extraction and reconstruction rather than analysis?

—   Are your time narratives written for billing convenience, or for future recoverability under scrutiny?

—   If the paying party used AI to analyse your next bill, would it understand your file better than you do?

If that last question made you wince, you are not alone — and it is a better problem to confront over a coffee now than across a costs negotiation later. A Fika is a sensible place to look at what an AI-ready costs record would mean for your matters.