Capability area 01 / 04
AI that works fromyour firm’s playbooks
I structure your playbooks, templates and case records so AI can use them, and I assemble the context a reviewer needs. Every draft shows its rationale and sources, and a lawyer decides.
Delivered in client workAll 4 capabilities on this page.
30-minute intro call on Calendly
01 / Who it’s for
Is your law firm’s knowledge ready for AI?
This work suits firms whose best guidance is hard to find and harder to apply the same way twice.
Playbooks and review criteria sit in documents, inboxes and a few senior lawyers’ heads.
Reviewers gather case and client context by hand before every decision.
You want AI drafts, but each one has to show its sources and reasoning.
Your product must apply each customer’s own guidance without mixing it with another’s.
02 / What’s included
The 4 capabilities in this area
Structure firm guidance and reusable playbooks, assemble relevant case and client context, and use them in drafting and recommendations. Reviewers can see the information and rules relevant to a proposed next step.
- 01
Organizational knowledge and firm playbooks
I turn your playbooks, review criteria, templates and drafting preferences into structured knowledge that staff and AI workflows can both use. Your firm’s way of doing the work stops depending on who is in the room, and each rule stays tied to its source.
How it works
- Collect
Gather the playbooks, templates, precedents and review criteria your team already relies on, and note who owns each one.
- Structure
Break guidance into rules, clauses and preferences, each with its source, its owner and the document types it applies to.
- Retrieve
Set up retrieval so a task finds the right rule and shows where that rule came from.
- Maintain
Add a review path so owners can correct or retire guidance as the firm’s practice changes.
- What it needs
- Access to the playbooks, templates and precedents you want used, and a named person who can say which version is current.
- Limits
- I structure and retrieve your guidance, and your lawyers decide the firm’s position and approve every change to it.
- Collect
- 02
Case and client intelligence
I assemble the records, correspondence, events and workflow state that matter to the task in front of a reviewer. Attorneys and case managers see the relevant case and client context in one place, limited to what their permissions allow.
How it works
- Connect
Link the case-management records, email and documents the task needs, through the interfaces and permissions those systems provide.
- Structure
Turn records, events and correspondence into structured case and client metadata the system can query.
- Select
For each task, select only the records that bear on it, within the requesting user’s permissions.
- Present
Show the assembled context with links back to the originals, so a reviewer can check it quickly.
- What it needs
- Read access to the case, email and document systems involved, and agreement on which records count as relevant for each task.
- Limits
- The system gathers and organizes context, and the responsible attorney or case manager decides what it means for the matter.
- Connect
- 03
Contextual drafting and explained recommendations
I build drafting and recommendation steps that apply your playbook and the case context, and show their reasoning. AI contract review against your own playbook is one example. A reviewer sees the draft, the rule applied and the sources behind it.
How it works
- Gather
Pull the case context and the playbook rules that apply to this document or task.
- Draft
Prepare the draft, markup or recommendation from that context, using structured outputs so each point can be checked.
- Explain
Attach a rationale to each point, citing the rule applied and the source passages it relied on.
- Hand over
Present the result to the reviewer for edits and a decision, and record what they changed.
- What it needs
- Playbook rules and templates in a usable form, the case context for the task, and a reviewer who owns the final text.
- Limits
- Drafts and recommendations are proposals, and a lawyer reviews, edits and decides before anything is relied on, sent or filed.
- Gather
- 04
Governed AI decisions and actions
I define what an agent may read, recommend, prepare and execute, and I enforce those lines in the system. Consequential actions run only after the responsible person approves, through controlled tools that log each step.
How it works
- Map
List every action the agent could take, and sort each one into read, recommend, prepare or execute.
- Permit
Tie access to the requesting user’s role, so the agent sees only what that person may see.
- Approve
Route consequential actions to the named approver, with the recommendation and its sources beside the decision.
- Record
Run approved actions through controlled tools and write an action-level audit entry for each one.
- What it needs
- A clear list of the actions in scope, the roles that may approve them, and the systems an approved action would touch.
- Limits
- The system enforces the boundaries you set, and your firm decides which actions need approval and who approves them.
- Map
03 / How it fits together
How does AI use a law firm’s knowledge?
Firm knowledge moves through five layers, from your sources to a reviewer’s decision. Each layer keeps the link back to where the information came from.
Firm sources
Playbooks, templates, precedents, case records and correspondence are the inputs. Each has a named owner, so someone is accountable for keeping it current.
Structured knowledge
Guidance becomes rules, examples and records, indexed for retrieval. Each item keeps a link to its source passage.
Task context
For each task, retrieval assembles the relevant playbook rules and case records, within the requesting user’s permissions.
Draft with rationale
The AI prepares a draft or recommendation and shows the rule it applied and the sources it used.
Review and record
A reviewer decides. Approved actions run through controlled tools, and each decision is written to the audit trail.
04 / Controls and limits
Who decides when AI drafts for your firm?
A lawyer decides. The AI retrieves and recommends within permissions, and anything with consequences waits for a responsible person’s approval.
Reads within permissions
The AI retrieves playbooks and case records only as the requesting user could. Firm and tenant boundaries stay in place.
Recommends with reasons
Drafts and suggestions arrive with the rule applied and the sources used, so a reviewer can check them.
Acts only when approved
Consequential actions run after the responsible person approves, through controlled tools, with an audit entry for each.
- Give legal advice or decide how a matter should come out.
- Send, file or change anything without a person’s approval.
- Read records outside the requesting user’s permissions.
- Treat a draft as final before a lawyer has reviewed it.
- ABA Formal Opinion 512 (July 2024)(opens in new tab)
It addresses competence, confidentiality, communication, supervision, candor and fees when lawyers use generative AI, which is why these workflows keep the review step explicit.
- NIST AI Risk Management Framework (Jan 2023)(opens in new tab)
This voluntary framework organizes AI risk work into govern, map, measure and manage, and it is useful background when you decide who owns what.
05 / Proof
What does this look like in practice?
I can describe two examples publicly. The rest of my knowledge work is under NDA.
Delivered with law firms, in-house legal tech teams and legal-tech founders over the past three years, mostly under NDA.
- Named client · Tech lead
Clausio
Clausio is an AI contract review tool for law firms. As tech lead, working directly with the co-founder, my work included tenant isolation security, new product features, third-party integrations including Stripe and a Microsoft Word add-in connected to firm knowledge. The add-in puts a firm’s knowledge and skills to work in contract review and day-to-day work.
- Public case study · US personal-injury firm
Human-governed AI intake
The intake system assembles case and client context for attorney or case-manager review. Documents are classified and key details extracted, case and client metadata give reviewers structured context, and access follows permissions. Suggested next actions stay separate from execution.
Read the case study - Under NDA · Other clients
Knowledge work for other firms
The knowledge and context capabilities on this page have also been delivered for other law firms, in-house legal tech teams and legal-tech founders, under NDA. I can’t name those clients.
07 / Next step
Tell me which knowledge your firm relies on
Describe the playbooks or records your team uses, the task they should support and where a lawyer must review. I’ll help you find a practical first scope.
30-minute intro call · or email hkhantanoli@gmail.com(opens in new tab)
What to share
- The playbooks, templates or records involved
- The systems that hold them today
- Where a lawyer must review
Please leave out passwords, access keys and client data.
Questions
What is a legal knowledge base for AI?
It is your playbooks, templates, precedents and case records, structured so software can find and apply the right piece. I break guidance into rules and examples, index them for retrieval with RAG and vector search, and keep a link from each item to its source. When the AI drafts or reviews a contract, the reviewer sees which rule it applied and where that rule came from.
Is our client data safe when AI reads our case files?
That depends on how access is built, so I design it before anything else. Retrieval runs within the requesting user’s permissions, each firm or tenant is isolated from the others, and every action is logged. ABA Formal Opinion 512 raises confidentiality as a duty when lawyers use generative AI, so I treat data handling as a design question from the first call. Which model provider sees which data, and under what terms, is a decision we make together.
Can this work with our existing document and case-management systems?
Often, yes. I first check the available APIs, permissions and data quality of the systems that hold your playbooks and case records, such as SharePoint, email through Microsoft Graph or your case-management system. Then we agree what can be connected and what stays as it is.
How is this different from an off-the-shelf legal AI tool?
An off-the-shelf tool brings its own general rules and workflow. This work builds around your playbooks, your case records and your approval steps, and connects to the systems you already use. A packaged tool can still be the right choice for a standard task, and I will say so if one fits.
How does a project start, and what does it cost?
It starts with a 30-minute intro call. You describe one workflow, the knowledge it depends on and where a lawyer must review. I then scope that workflow with clear deliverables and acceptance criteria. Projects are scoped per workflow, so cost follows the scope we agree.
Last updated · Written by Hamza Khan