Case study · AI-native · Conversational CLM
Create and manage contracts by conversation, with known details pre-filled and guardrails that scale with risk.
Client branding is removed from all screens. People, accounts and contracts shown are demo data.
Aime is a contracting agent inside CLM. Users create and manage contracts through conversation instead of navigating forms and menus. Aime gathers details, pre-fills what the system already knows, applies the right template and approvals, and moves the contract through its lifecycle.
Organizations stay safe through risk-based guardrails, an audit trail after every state change, and a fast path only for low-risk, unchanged standard templates.
Creating a contract meant hunting through forms, re-entering known data and guessing which template and approval path applied. Many chat assistants make it worse by asking redundant questions, which kills trust fast.
Business users need a contract quickly without legal knowledge, but today they re-enter known data and guess at templates and approvals, and a chat agent that asks the obvious loses their trust.
Use what the system already knows, so Aime never asks the obvious?
Show where every pre-filled value came from, so users can trust it?
Speed up low-risk contracts while keeping guardrails on everything else?
The plan covered the people who ask for contracts, the people who approve them, and the system rules that decide templates and review paths.
| Method | Who | What we wanted to learn |
|---|---|---|
| Stakeholder interviews | Product leadership, PM, spec owner | Agent scope, skills, guardrail rules |
| User interviews | Account executives | How contracts are requested and where they stall |
| Reviewer interviews | Legal ops, contract managers | What must always be reviewed, and what never needs it |
| Pattern review | AI assistants and CLM tools | Slot-filling, confirmations, source attribution |
| Prototype sessions | Account executives | Trust in pre-filled values and assertions |
We synthesized interviews with account executives, legal ops, contract managers and product leadership into five themes.
"If it already knows the account, why is it asking me?"
Account executive · Illustrative"Half my queue is standard NDAs nobody changed."
Legal ops · Illustrative"I'll trust it when it tells me where the number came from."
Contract manager · IllustrativeWe compared conversational patterns across categories of tools: how they collect details, recommend options, confirm actions and attribute sources.
| Pattern | General AI assistants | CLM wizards | Enterprise chat bots | Our opportunity |
|---|---|---|---|---|
| Uses page context to skip questions | ||||
| Groups required fields in one step | ||||
| Asserts the best option, then confirms | ||||
| Shows the source of each value | ||||
| Confirms before irreversible actions | ||||
| Logs every state change |
One named persona carried through every flow and prototype, which kept discussions concrete. The legal reviewer is a secondary role who appears only when a contract needs review.
Requesting a contract was easy. Choosing the right template and waiting for review, even for standard paper, were where deals stalled.
These principles turned debates about wording into quick decisions: does this turn follow them or not?
Each entry point starts from what the page already knows. All three converge on the same NDA flow, where an unchanged standard template skips legal review.
Recommendation menus look helpful, but each list adds a turn and makes the user do the choosing. Stating the most-used option, with its reason and an easy way to change it, kept users in control with fewer steps.
Reuse carries over attributes and the contract team from the best precedent, trimmed to two actions. Required details arrive in one grouped card, with confirmed values separated from the one choice still needed.
Open the assistant from Home, Accounts or Contracts and take an NDA from request to activation.
Password protected. Email me at ramprasadp1304@gmail.com for access. Best on a laptop or desktop screen.
1234
1234Each lifecycle stage allows a specific set of actions. Guardrails scale with risk: unchanged standard templates take the fast path, while edits or higher-risk terms route to legal review.
| Stage | Generate | Route for review | Send for signature | Update | Activate | Guardrail |
|---|---|---|---|---|---|---|
| Draft | ✓ | ✓ | — | ✓ | — | Irreversible warning before generating |
| In review | — | — | — | ✓ | — | Skipped for standard, unchanged templates |
| Approved | — | — | ✓ | — | — | Signers from rules engine and precedent |
| In signatures | — | — | — | — | ✓ | Activate only when both parties sign |
| Executed | — | — | — | — | — | Read-only; amendments start a new flow |
12Each piece of feedback became a decision and a concrete change in the prototype.
| Feedback | Decision | Change |
|---|---|---|
| Counterparty asserted too early on Home | Ask first where there's no context | Guessed account→"Who is the counterparty?" |
| Creation was blocked from Contracts | Allow it, just not as the default prompt | Blocked→Allowed, actions first |
| Reuse card had too many actions | Trim to the two decisions that matter | 4 buttons→Confirm reuse · Start fresh |
| Duplicate warning contradicted reuse | Remove it; reuse already covers it | "Duplicate found"→Reuse proposal only |
| Persona changed between flows | One named persona throughout | Mixed users→Nico Reyes everywhere |
Six participants completed three tasks: create an NDA from Home, create one from Accounts, and reuse an existing contract.
| Finding | Severity | Change made |
|---|---|---|
| Users hesitated to accept values without knowing their origin | High | Source label on every pre-filled value |
| Some expected to pick the template themselves | Medium | "Use different template" next to the primary action |
| Inactive team members carried over silently | High | Flagged as "Inactive — will be removed" |
| Unclear why legal review was skipped | Medium | "Standard template — no legal review required" message |
| Hard to tell what had already happened | Low | Logged line after every state change |
The prototype moved from v0 to Claude Design and was prepared for grooming and skill-spec review with annotated flows.
These are targets, not measured results. They'll be replaced with live data after launch.
To create and send a standard NDA
Fewer conversation turns
Of standard NDAs skip legal review