Two recent legal tech developments highlight a practical lesson for law and accounting firm leaders: the biggest gains in AI often come from how work is structured before the model ever sees it. One story focuses on lowering token costs by changing how a contract is sliced up before AI review. Another shows demand for transaction workflows that keep secure document sharing, permissions, version control, and audit trails in one place.
Lower AI costs start with better document structure
A recent report on Syntheia says token savings came from changing the starting conditions of document review, not from switching models or routing work to a cheaper LLM. The company said the amount of text fed into an LLM varied by up to 30x across methods, while the final answer length barely changed.
For firm leaders, the practical point is straightforward: if you are exploring custom AI for contract review, due diligence, or matter triage, the workflow design matters as much as the model choice. Structure-aware indexing and retrieval can reduce the amount of unnecessary text sent into the model and may help control cost at scale.
Agentic workflows will matter more when tools choose what to read
The Syntheia findings also matter because AI agents increasingly decide for themselves what to retrieve and read. That means firms need systems that are built for selective retrieval, not just full-document input.
This is relevant for law firms handling transactional work and for accounting firms building review workflows around contracts, client documents, or supporting records. A good first step is to map which parts of a file actually need model attention and which parts can stay outside the prompt until needed.
Secure transaction workflows are replacing disconnected point tools
Another recent development comes from Legatics, which launched Data Rooms as part of a wider deal platform. The company framed the product as secure document sharing for due diligence and other controlled-access situations, with granular permissions, version control, and audit trails built into the same workflow used for transaction management.
That direction reflects a broader market preference for integrated workflows instead of separate tools stitched together after the fact. For professional-services firms, the lesson is to look for automation that supports the whole process: intake, document control, review, approvals, and reporting.
What this means for law and accounting firm leaders
The common thread across these announcements is not just AI capability, but operational design. Firms get more value when AI sits inside a controlled workflow with clear permissions, well-defined document boundaries, and a reason to retrieve only what is needed.
If your firm is planning custom AI, think in terms of specific workflows rather than general-purpose chat. Transaction review, due diligence, client intake, and document-heavy advisory work are all better candidates for automation when the process can be structured and audited.
- Focus on workflow design first: the way documents are sliced, indexed, and retrieved can affect cost and usefulness more than model choice.
- Use AI where the process is structured enough to support selective retrieval, permissions, and audit trails.
- Prefer integrated systems that combine document control and matter or deal management rather than adding another standalone tool.
- For custom AI projects, start with one high-value document workflow and define exactly what the model should and should not see.
Sources watched
- Syntheia Slashes Token Costs With Novel Approach (Artificial Lawyer)
- Legatics Data Rooms Launches as VDR Alternative (Artificial Lawyer)
