Shoosmiths reviewed 15 contract-AI suppliers, then built its own
UK law firm Shoosmiths is piloting a contract reviewer built on its own playbooks and estimates 3 to 5 hours saved a document. No cost or error rate is given.
At a glance
- Company
- Shoosmiths · Professional services · Size not stated
- Team and use case
- Legal and compliance · Contract review
- Tools
- Microsoft Azure, Microsoft 365 Copilot, Microsoft Copilot Studio, SharePoint
- Stage
- Pilot, since Pilot under way, June 2026
- Results
- Review time saved per document (estimate from earlier AI work): 3 to 5 hours
- Documents a year handled by the pilot team: About 3,000
- Suppliers reviewed before building: About 15
- Partners in an internal poll wanting early access: 63% (respondents not counted)
- Cost
- Not disclosed. No build cost, build time, running cost or licence figures.
- People
- The chief executive says the tools are not built to cut jobs; no headcount figures. A trainee describes Apollo as a learning tool. Who signs off a reviewed contract: Not disclosed.
- Evidence
- Vendor-reported
Published by the company selling the product, usually with the customer. Wins are selected; cost is usually missing.
What this leaves out: Published by Microsoft, which sells the cloud service used. The hours figure is an executive's estimate from earlier AI work, not a pilot measurement. No baseline, error rate, sample or cost.
Shoosmiths, a UK law firm, is piloting a contract review system called Apollo that checks draft contracts against the firm's own rulebooks, and its director of client technology says this kind of review can save three to five hours on a single document. The account is in a [Microsoft UK Stories feature dated 24 June 2026](https://ukstories.microsoft.com/features/how-shoosmiths-is-turning-legal-expertise-into-ai-powered-advantage/). Microsoft sells the cloud service Apollo runs on, so the page is vendor-published, and the hours figure is an estimate, not a pilot result.
Why the firm built instead of buying
Tony Randle, director of client technology and service improvement, says the firm reviewed around 15 suppliers and could not find a product that met its standards. It had built a contract tool before: an earlier system for commercial supply agreements "took us two years … to train" on one contract type, he says.
Apollo works from playbooks: the firm's written rules for a contract type, saying what the contract should contain, why it matters and what wording fixes a problem. Randle says a playbook can hold hundreds of rules. Built on Microsoft Azure, Apollo compares a draft with the playbook, flags each place a rule is not met, and shows the lawyer which rule it applied and why. Microsoft introduced the firm to development partners; the page does not name them or the AI models used.
The number, and what it is not
Randle's three to five hours a document is described as "drawing on earlier AI work". The page gives no measured result from the Apollo pilot. It says the pilot team handles about 3,000 documents a year. Multiplied out, that is 9,000 to 15,000 hours a year (3,000 × 3 and 3,000 × 5), but only if every document saved the full estimate; Microsoft does not make that claim and neither should a business case.
The page is silent on what Apollo cost to build and run, how long the build took, how often a flag is wrong or a breach is missed, who signs off a reviewed contract, and whether clients are billed differently for the work.
Reading this at work? Get one story like it explained every morning.
A five-minute read every morning at 7am ET: one story explained, two in brief, what else happened, one number. Unsubscribe in one click.
The people side
Chief executive David Jackson says the firm is "not building these tools to cut jobs"; the aim he gives is to help staff learn. No job or hiring figures are given. A trainee solicitor quoted on the page says Apollo flags issues she might have missed and explains why. Randle says 63% of partners who answered an internal poll want to be early adopters; the page does not say how many answered.
Staff also have Microsoft 365 Copilot and Copilot Studio, Microsoft's tool for building agents without code. Joanne Bevan, who joined 28 years ago as a secretary and has never written code, built an agent that answers plain-English questions from a SharePoint library of procedures for a retail client with hundreds of leased properties. She calls it "just a simple retrieval agent". No time saving is reported for it.
What to do if you run legal operations
- Pick your highest-volume contract type (NDAs, supplier terms, leases) and write its playbook as numbered rules with approved fallback wording. Shoosmiths' account makes the rules, not the software, the asset, and you need them to test any product.
- In vendor demos this month, ask for the rule and the reasoning behind every flag, on your documents. A tool that cannot show them cannot be checked by the lawyer who signs.
- Before a pilot, time a manual review on 20 documents, then count wrong and missed flags as well as hours. Shoosmiths' three to five hours is an estimate from earlier work.
Take this to work
Forward this
UK law firm Shoosmiths reviewed about 15 contract-review suppliers, built its own tool on its internal playbooks, and estimates three to five hours saved a document, per a Microsoft UK Stories feature of 24 June 2026; no cost or error rate is given. Before we buy a contract review tool, we should write our own playbook for one contract type and test vendors against it.
Questions to ask
- For each flag, can the tool show which of our playbook rules it applied and the reasoning, in a form the signing lawyer can check?
- On a sample of our own contracts, how many flags were wrong and how many breaches were missed, and who counted?
- What does it cost at our document volume, including the work of writing and maintaining the playbooks?
- Who signs off a reviewed contract, and who is accountable when a missed clause reaches a client or counterparty?
- If we stop using the tool, do we keep the playbooks in a format another product can read?
The arithmetic
Hours freed a year = documents reviewed a year × hours saved per document × share of documents the tool can take. Net value = hours freed × hourly cost of the reviewer − cost to build, license and maintain.
Shoosmiths' pilot team: about 3,000 documents × 3 to 5 hours = 9,000 to 15,000 hours a year if every document saved the full estimate. The source gives neither the share of documents Apollo can take, the hourly cost nor the build and running cost, so net value cannot be worked out.
Questions this article answers
What is Shoosmiths' Apollo?
Apollo is a contract review system the UK law firm Shoosmiths built on Microsoft Azure. It compares draft contracts with the firm's own playbooks and shows the lawyer which rule each flag comes from, according to a Microsoft UK Stories feature dated 24 June 2026, which described it as still in pilot.
How much time does AI contract review save at Shoosmiths?
The firm's director of client technology says this kind of review can save three to five hours on a single document, drawing on earlier AI work. The Microsoft page gives no measured result from the Apollo pilot and no error rate.
Why did Shoosmiths build its own contract review tool?
The Microsoft page says the firm reviewed around 15 suppliers and could not find an off-the-shelf product that met its standards. It chose to encode its own playbooks, which can run to hundreds of rules, and have the system apply them.
Did Shoosmiths cut jobs because of AI?
The page reports none. The chief executive says the tools are not being built to cut jobs. No headcount or hiring figures are given.
BNY gives 140 AI agents a supervisor, a scorecard and a 16-step approval
BNY runs nearly 140 AI agents like staff: each has a named supervisor and a scorecard, and a new one passes 16 review steps. No cost is given.
Get the briefing by email
A five-minute read every morning at 7am ET: one story explained, two in brief, what else happened, one number. Unsubscribe in one click.