Legal & Compliance

The law is in the documents. The tracking is in Excel

Every diligence exercise, contract register, obligation matrix, and matter budget in the profession runs on a spreadsheet somebody maintains by hand. SheetAI works in that spreadsheet, abstracting contract terms into rows with page references, reconciling WIP and realisation, and flagging what needs a lawyer rather than pretending it does not.

Try it on your sheet See real workflows

Audit-ready trailNo data trainingReversible per action

94%

AI accuracy on NDA review, vs 85%

In a 2018 LawGeex study, software averaged 94% accuracy at spotting issues in five NDAs against 85% for 20 experienced US lawyers, taking 26 seconds against their 92 minutes.

Source: ABA Journal: AI software is more accurate, faster than attorneys when assessing NDAs (2018)

5 hrs

Weekly time AI could save

Professionals expect AI to save them 5 hours a week within a year, worth about $19,000 per person annually. Thomson Reuters surveyed 2,275 legal, tax, risk and compliance professionals in 2025.

Source: Thomson Reuters: Future of Professionals Report (2025)

8.6%

Value lost to poor contract management

Poor contract management erodes an average of 8.6% of a contract's value each year. From World Commerce & Contracting, the global contracting professional body.

Source: World Commerce & Contracting: Contract Management, an overlooked driver of business agility (2025)

Legal & Compliance: illustration for SheetAI's legal & compliance use case

Law firms bill only a fraction of the day they work.

Clio's Legal Trends research has consistently found lawyers record a small share of an eight-hour day as billable, and collect less than they record. The gap is administration, tracking, and the reporting that surrounds the work, most of which is a spreadsheet exercise.

Source: Clio: Legal Trends Report

This page is about the structured, repetitive work that surrounds legal practice rather than the practice itself: due diligence trackers, contract and obligation registers, matter budgeting and realisation analysis, e-billing review, entity and filing calendars, and the compliance matrices that regulated businesses maintain. SheetAI reads your documents into rows with page-level citations, builds the analysis in standard formulas, and flags ambiguity for a lawyer instead of resolving it. It does not give legal advice, it does not opine on whether a clause is enforceable, and it is not a substitute for professional judgement or supervision; the boundaries section says so in the terms your risk partner will want to see.

Where legal teams actually spend the day

Four numbers that explain why the profession's most-used software is still a spreadsheet.

Billable utilisation across the profession

~37%

Roughly 2.9 hours of an eight-hour day recorded as billable in Clio's research. Realisation and collection then reduce what is actually earned from that further.

Clio Legal Trends

AI performance on narrow contract issue-spotting

94%

The LawGeex NDA study remains the most-cited controlled comparison. It measured a specific, well-defined task against a defined rubric, not legal judgement, and the distinction matters.

LawGeex (2018)

Legal departments citing spreadsheets for matter tracking

Majority

Despite widespread matter management software, the actual tracking of diligence, obligations, and budgets overwhelmingly happens in Excel because every matter has a bespoke structure.

Legal operations surveys

Time in a diligence exercise spent on extraction rather than analysis

Most of it

Reading agreements to populate a tracker is the bulk of the hours in a transactional diligence review. The analysis that follows is the part clients actually pay for.

Practitioner experience

The pattern: the extraction is mechanical and the judgement is not, but both currently consume the same lawyer's evening. Moving the extraction into the sheet, with a page citation on every field so it can be verified in seconds, is the change that gives the judgement its time back.

Anatomy of a diligence exercise

Where the hours go on a mid-market transactional diligence review, before any of it is automated.

Phase 1: Scope and build the tracker

›Agree the issues list with the client
›Build the tracker columns and conventions
›Assign documents across the team

⚠ Every team builds the tracker slightly differently, and the conventions drift within a week of the exercise starting.

Phase 2: Read and abstract

›Term, renewal, and termination mechanics
›Assignment and change-of-control provisions
›Liability caps, indemnities, and exclusions

⚠ This is the overwhelming majority of the hours, it is done at speed under a deadline, and consistency across reviewers is a matter of hope.

Phase 3: Flag and escalate

›Identify provisions triggered by the transaction
›Escalate the genuinely problematic ones
›Chase the missing documents

⚠ The flag list is only as good as the least attentive read, and nobody records which reviewer looked at which agreement.

Phase 4: Summarise for the report

›Aggregate findings by category and severity
›Cross-reference back to the source documents
›Draft the issues narrative

⚠ Cross-referencing back to documents is the part that gets cut, which means the report's assertions are hard to verify later.

Phase 5: Track to completion

›Monitor consents and waivers obtained
›Update the tracker as documents arrive
›Reconcile against the disclosure schedule

⚠ The tracker is stale the moment the data room updates, and nobody knows which version the report was written from.

Cautionary tale

When the spreadsheet is the control

In 2008, Barclays agreed to acquire assets from the collapsing Lehman Brothers. The contract schedule listing which trades were included was prepared under extreme time pressure in a spreadsheet, and rows that had been hidden rather than deleted were converted to PDF and included in the filed agreement. The result was that roughly 179 trades Barclays had not intended to acquire were captured by the executed document. The firm sought relief on the basis of clerical error. Nobody misunderstood the law; a hidden row in a spreadsheet made it into a signed agreement.

The lesson: In legal work the spreadsheet frequently is the control: the tracker is what determines what is disclosed, what is consented to, and what is acquired. That makes provenance and change history operational necessities rather than good practice. SheetAI records the source document and page behind every extracted term, logs every change it makes with a reversible diff, and flags ambiguity for a human instead of resolving it silently.

Source: Reuters: Barclays, Lehman and the hidden spreadsheet rows

Who it's for

If your matter is tracked in a workbook somebody updates by hand, start here.

Transactional Associates

Populate diligence trackers from hundreds of agreements under a closing deadline.

In-House Counsel

Maintain a contract register and obligation calendar with no dedicated system and no team.

Legal Operations

Analyse outside counsel spend and matter economics across e-billing exports.

Compliance Officers

Keep an obligations matrix current across jurisdictions and regulatory change.

Practice & Finance Managers

Report WIP, lockup, realisation, and matter profitability every month from raw exports.

Real legal workflows

The prompt, what the agent does, and what a supervising lawyer receives.

legal_workbook.xlsx - SheetAI

You ask

“I've attached 22 supplier agreements. Populate Tracker!A1 with: counterparty, effective date, term, renewal mechanism, termination rights, assignment provision, change-of-control provision, liability cap, and governing law. Cite the clause number and page for every field. Where drafting is ambiguous, write REVIEW and state what is unclear; do not interpret it.”

SheetAI does

✓Reads each agreement and locates the relevant provisions.
✓Writes one row per agreement with dates as dates and caps as numbers.
✓Cites clause number and page against every extracted field.
✓Flags ambiguous or unusual drafting for a lawyer rather than resolving it.

Formula written

=COUNTIF(H2:H23,"REVIEW")

Result

22 agreements abstracted with clause-level citations, and 17 fields marked REVIEW across 9 documents. The review list was worked by an associate in an hour instead of the whole set being re-read.

Everything legal teams need, in the tracker they already keep

Clause-level citations on every extracted term, ambiguity flagged rather than resolved, and matter economics computed from your raw exports in formulas anyone can audit.

✓Contract abstraction with clause and page citations

✓REVIEW flags on ambiguous drafting

✓Obligation and notice-period calendars

✓Realisation, WIP, and lockup analysis

✓Outside counsel guideline compliance checks

✓Full change log with reversible diffs

Plays well with your stack

Data-room document exports
Practice management exports (Clio, Aderant, Elite)
E-billing and LEDES exports
Contract lifecycle management exports
Companies House and registry downloads
Word and PDF agreement attachments

What one diligence exercise looks like

A representative mid-market transactional diligence review across 22 material agreements, before and after the extraction moves into the sheet.

Before SheetAI

~44 hours

Mon
Scope the issues list; build the tracker; allocate documents
~6h
Tue
Read and abstract the first tranche of agreements
~11h
Wed
Second tranche; conventions have already drifted
~11h
Thu
Flag triggered provisions; chase missing documents
~9h
Fri
Summarise for the report; re-open documents to cross-reference
~7h

✓

With SheetAI

~12 hours

Mon
Tracker built; agreements abstracted with clause citations
~3h
Tue
REVIEW list worked by an associate; ambiguities resolved
~3h
Wed
Change-of-control exposure grouped and value-ranked
~2h
Thu
Consent workstream started with the report drafted from cited rows
~4h

~73% reduction on a representative 22-agreement diligence exercise.

What SheetAI will not do

In a regulated profession the limits are not caveats, they are the terms on which the tool can be used at all.

Give legal advice

It extracts, organises, and computes. It does not advise on whether a clause is enforceable, what a provision means in a given jurisdiction, or what course of action a client should take. Every output is work product for a qualified lawyer to review and adopt, and it should be treated exactly as you would treat a junior's first draft: useful, and not yet advice.

Resolve ambiguous drafting

Where a provision could reasonably bear two readings, it says so and flags it. This is deliberate: an extraction tool that quietly picks the more likely interpretation produces a tracker that looks complete and is silently wrong. The REVIEW column is the most valuable output on the sheet, and it is usually short.

Replace supervision or conflict checks

Professional conduct obligations (supervision of work product, conflicts, client confidentiality, and privilege) are yours and do not transfer to a tool. Nothing here substitutes for the supervising solicitor's or attorney's review, and no output should reach a client without it.

Guarantee privilege is preserved

Processing privileged material through any third-party service raises questions your firm's risk function needs to answer, not your software vendor. Check your professional indemnity terms, your client engagement letters, and your jurisdiction's guidance on cloud processing before putting privileged documents through this or any comparable tool.

“

What changed was not the speed, though that was real. It was the citations. Every line in the tracker points at a clause and a page, so when the partner challenges an entry it takes ten seconds instead of reopening the document set. The REVIEW column is the first thing I read now; it is where the actual work is.

Senior Associate, corporate team (mid-market transactional practice)

Frequently asked

The questions risk partners and general counsel ask first.

Does this give legal advice?

›

No, and the distinction is not a formality. It extracts terms into structured rows, computes metrics from your data, and organises what is in your documents. It does not advise on enforceability, interpretation, or strategy. Everything it produces is work product requiring review and adoption by a qualified lawyer, the same standard you would apply to a trainee's first pass, which is roughly the right mental model for what it is doing.

How accurate is contract abstraction?

›

For commercially drafted agreements with a text layer, extraction of well-defined terms (parties, dates, caps, governing law, notice periods) is reliable, and every field carries a clause and page citation so verification takes seconds rather than a re-read. The LawGeex NDA study is the useful reference point: on narrow, well-specified issue-spotting, automated systems performed comparably to or better than experienced lawyers. That study measured a defined task against a rubric, not legal judgement, and the same limitation applies here.

What happens with ambiguous drafting?

›

It writes REVIEW and states what is unclear rather than choosing an interpretation. This is the single most important behaviour in the workflow. A tracker where every cell is filled looks finished; a tracker with seventeen honest REVIEW flags is finished, because you know exactly where the judgement is required. Resolving ambiguity is a lawyer's job and it stays one.

Can I use it on privileged or confidential material?

›

That is a question for your firm's risk function rather than for us, and you should treat it as one. Your professional indemnity terms, your client engagement letters, and your jurisdiction's guidance on third-party cloud processing all bear on it. Your workbook stays in your account and is not published anywhere, but the obligations attached to privileged material do not change because the file is open in a different application.

Will it work with our existing tracker format?

›

Yes. It populates the columns you have rather than imposing a schema, because every diligence exercise and every client has conventions. Tell it your column headings and your conventions for dates, currencies, and flags, and it follows them.

Can it help with matter profitability and WIP?

›

Yes, and this is often where the fastest return sits. Realisation, write-off, WIP ageing, and lockup all come out of time and billing exports that firms already produce and rarely analyse past a headline. It computes them in standard formulas per matter, fee earner, and practice area, and trends them across quarters so you can see movement rather than a single reading.

Does it handle e-billing review?

›

Yes. Supply your outside counsel guidelines and it applies each one as an explicit test across the line items: block billing, rate-card breaches, administrative work at fee-earner rates, excessive single entries. It produces the exception list with amounts by firm. The decision on what to challenge remains a commercial judgement.

Is there an audit trail?

›

Yes. Every action is logged with the prompt, the affected range, and a reversible diff, and every extracted field carries its source citation. For a tracker that will determine what is disclosed or consented to, that record is the point, as the Barclays–Lehman hidden-rows episode illustrates, in legal work the spreadsheet frequently is the control.

Sources & further reading

Every benchmark and statistic on this page is drawn from publicly available research. We cite our sources because we read theirs.

Get started

Stop retyping clauses.
Start reviewing them.

Attach a set of agreements, name the columns your tracker uses, and get abstracted rows with a clause citation on every field, and an honest list of what needs a lawyer.

All use cases

Pay as you go, no subscription required