Tech Spending Grew 9.7%. Demand Grew 1.9%. The 2026 Legal Tech ROI Gap Is an Accounting Problem Before It Is an AI Problem

📊 The Two Numbers That Do Not Belong Together

Read the 2026 law firm financial reporting released through August and one pairing keeps surfacing: technology and knowledge-management spending grew about 9.7%, the fastest pace tracked in recent years, while overall demand - the actual volume of legal work - grew roughly 1.9%. Profits are at records, but the record profits are being driven by rate increases, not by doing meaningfully more work.

That combination is unusual. In most industries, a five-fold gap between input spending growth and output volume growth would trigger an immediate ROI review. In legal, it has mostly triggered another round of procurement.

🔍 Why "Did the AI Work?" Is Currently Unanswerable at Most Firms

Ask a managing partner whether their AI drafting tool paid for itself and you will usually get one of three answers: an anecdote about a associate who saved a weekend, a vendor-supplied time-savings estimate, or an honest shrug. None of those is a measurement. The reason is structural, and it has three layers.

1️⃣ The cost sits in the GL as an overhead line

Software subscriptions are typically booked to a single technology expense account, allocated across the firm as overhead, and never touched again. Nothing in that entry says which practice group used the license, which matters it touched, or what it replaced. From the ledger's point of view, an AI research platform and the office coffee service are the same kind of number: fixed, allocated, unquestioned.

2️⃣ The benefit shows up as an absence

When technology works, hours do not get recorded. A brief that used to take nine hours takes five. In a firm that measures success by hours billed, the four hours that vanished look like a productivity decline, not a win. Unless the firm is tracking cost-to-complete by matter type, the benefit is literally invisible - and on an hourly matter, it can register as revenue loss.

3️⃣ The two datasets never meet

This is the one that matters. Technology cost lives in the accounting system. Matter effort, realization, and write-downs live in the practice management or billing system. Connecting them requires an export, a spreadsheet, and someone willing to build a mapping by hand. Most firms attempt it once, during budget season, and never repeat it - which means the analysis is always retrospective and always stale.

⚖️ The Bubble Risk Is Real, But It Is Not the Technology

The "AI bubble" framing appearing in 2026 commentary is worth taking seriously, though not for the reason it is usually stated. The risk is not that legal AI fails to work. By most accounts it works well on document review, research retrieval, drafting first passes, and summarization. The risk is a pricing mismatch: firms are investing capital to compress the hours they sell, while continuing to bill primarily by the hour.

That only resolves in one of two ways. Either firms move a larger share of work to fixed and outcome-based pricing - in which case cost-to-serve becomes the critical number - or clients capture the efficiency through lower bills, in which case margin compresses and the spend needs to be justified on cost, not revenue. Both paths end at the same requirement: the firm has to know what each matter type actually costs to run.

🏗️ What Attribution Actually Requires

Measuring technology ROI at a law firm does not require a data science team. It requires four things to be recorded in the same place.

🔗 A cost that can be tagged beyond the GL account

Vendor bills need to carry more than an amount and a date. Practice group, deployment scope, and replaced-system reference should be attributes on the payable itself, so the expense can be sliced later without a reconstruction project.

⏱️ Effort captured by matter type, not just by timekeeper

The unit of analysis is the matter template - "uncontested naturalization," "auto liability under $50K," "commercial lease review" - not the individual attorney. Firms that run matters from templates already have this structure; firms that open every matter as a blank record do not, and cannot compare like with like.

💰 Realization and write-downs coded by reason

A write-down absorbed silently at pre-bill teaches leadership nothing. A write-down coded as "over standard hours for matter type" is a measurement. When AI compresses drafting time, this is the field that proves it.

📈 One ledger where all three can be queried together

This is the part that cannot be solved with process discipline alone. If the general ledger and the matter ledger are different systems from different vendors, every cross-cutting question requires a reconciliation first.

🧭 How CaseQube and LawAccounting Change the Question

CaseQube runs practice management and legal accounting on a single Salesforce-powered ledger. Vendor bills, expense allocations, time entries, billing, write-downs, trust activity, and the general ledger are records in one system, related to matters and matter templates rather than reconciled to them after the fact.

Practically, that means a firm can ask - and answer, without an export - questions like: what did the "immigration - employment-based petition" matter type cost to run this quarter versus last, how did median hours move, what share of billed value was written down and for what coded reason, and what is the fully loaded technology allocation against that practice group's contribution margin. The reporting engine treats those as ordinary queries because the underlying data was never split apart.