Intake to Decision: Building an AI‑Ready Legal Workflow…

Knowledge workers spend nearly a day each week just searching for information. For in-house legal, that drag shows up as intake ping-pong, missing context, and slow decisions. The result: legal becomes a bottleneck precisely when the business needs clarity and speed.

What if every request captured the right context, routed to the right path, and returned a decision with the why attached—automatically? That’s the shift from ad hoc service desk to an AI-ready operating system for legal.

Why This Matters Now

Budgets aren’t scaling with demand, but expectations are. Business partners want answers in hours, not weeks. Regulators are raising the bar across privacy, AI, and procurement. And gen AI is now mature enough to help—if you give it well-structured playbooks and guardrails.

The goal isn’t a chat bot. It’s a durable path from intake to decision: standardized inputs, executable positions, automated triage, human-on-the-loop approvals, and an audit trail. Every cycle should make the next faster and smarter.

Design the Golden Path: Standardize Intake and Triage

Start where work begins. Map your top three request types—e.g., NDAs, vendor contracts, and marketing reviews—and define the minimum data required to act.

Required fields: counterparty, use case, data exchanged, contract value, due date, jurisdiction, and system/tool names.
Channel-native intake: meet the business in Slack/Teams and email; don’t force a new portal. Use lightweight forms that feel native.
Routing rules: route by matter type, risk flags, and business unit. Default to self-serve when the policy is clear; escalate on exceptions.
Service levels: set time-to-first-response and resolution targets per request type so legal and business are aligned.

On Sandstone, this becomes a modular intake: structured fields, auto-classification of free text, and triage rules that send simple work to self-serve and complex work to counsel with context attached.

Make Playbooks Executable, Not Just Readable

Most teams have policy PDFs. AI needs decisions, not prose. Convert playbooks into positions with conditions and fallbacks.

Example (NDA): If counterparty is customer and sharing is mutual, use Standard Mutual NDA v3; if unilateral and counterparty is vendor, use Vendor Unilateral NDA v2 with fallback clauses A/B; if data is sensitive (PII/health), require DPA link and privacy review.
Encode redlines: Clause-level rules for what to accept, counter, or escalate—plus the rationale.
Attach evidence: Link positions to prior approved deals and risk notes so every answer has a source.

In Sandstone’s knowledge layer, these positions become living rules an AI agent can apply, cite, and learn from, with every approval strengthening the next decision.

Put AI Agents in the Loop—With Guardrails

AI agents shine when they operate inside a well-defined workflow.

Classify: Detect request type and risk signals from unstructured messages.
Enrich: Extract entities (counterparty, systems, data types) and fill missing fields by asking the requester concise follow-ups.
Retrieve: Pull the relevant position, template, or clause from the knowledge layer.
Propose: Generate the next step—approve, template, or redline—with an explanation and links to sources.
Escalate: Hand off edge cases with a clean summary, suggested options, and a confidence score.

Every action should be logged. Humans approve, the system learns. The outcome is consistent decisions delivered faster, with a transparent “why” that builds trust.

Measure What Scales

Track a small set of metrics that reflect business value, not just activity.

Time to First Response (TFR): From intake to acknowledgement.
Request-to-Decision Time: End-to-end resolution per request type.
Self-Serve Rate: Percentage resolved without lawyer intervention.
Deflection Quality: Reopened or escalated after self-serve.
Playbook Coverage: Share of volume mapped to executable positions.
Confidence and Citation Use: Agent confidence vs. human overrides; percentage of decisions with sources.

These metrics show whether knowledge is compounding and where policy gaps remain.

A One-Week Pilot: NDA Intake to Decision

You don’t need a year-long program to see value. Stand up a focused pilot.

Day 1–2: Choose NDA as your high-volume workflow. Define required fields and map the golden path. Agree on SLAs.

Day 3–4: Convert your NDA guidance into executable positions with fallbacks. Attach templates, clauses, and prior approvals as evidence.

Day 5: Configure intake in Slack/Teams and email. Enable the AI agent to classify, enrich, and propose the right NDA path with rationale.

Day 6–7: Launch to one business unit. Monitor TFR, Request-to-Decision, and Self-Serve Rate. Capture edge cases to refine positions.

On Sandstone, this becomes a reusable module. Add vendor DPAs next, then low-risk procurement terms, reusing the same layered data and triage patterns.

Actionable Next Step

Pick one workflow with high volume and low variance—NDAs, low-risk MSAs, or marketing reviews. In one hour, write five if/then positions with clear fallbacks and attach the right templates. Turn that into a channel-native intake and set a two-day SLA. Measure three metrics for two weeks: TFR, Request-to-Decision, and Self-Serve Rate. Iterate.

The Bedrock of Speed, Alignment, and Trust

Legal shouldn’t be a ticket queue; it should be the connective tissue of the business. When intake is structured, positions are executable, and AI agents operate with guardrails, every request strengthens your legal foundation. That’s the Sandstone approach: strength through layers, crafted precision, and natural integration with how teams already work. The payoff is scalable, streamlined operations that move the business faster—with clarity, confidence, and an audit-ready trail of decisions that compound over time.

Why This Matters Now

Design the Golden Path: Standardize Intake and Triage

Start where work begins. Map your top three request types—e.g., NDAs, vendor contracts, and marketing reviews—and define the minimum data required to act.

Required fields: counterparty, use case, data exchanged, contract value, due date, jurisdiction, and system/tool names.
Channel-native intake: meet the business in Slack/Teams and email; don’t force a new portal. Use lightweight forms that feel native.
Routing rules: route by matter type, risk flags, and business unit. Default to self-serve when the policy is clear; escalate on exceptions.
Service levels: set time-to-first-response and resolution targets per request type so legal and business are aligned.

Make Playbooks Executable, Not Just Readable

Most teams have policy PDFs. AI needs decisions, not prose. Convert playbooks into positions with conditions and fallbacks.

Example (NDA): If counterparty is customer and sharing is mutual, use Standard Mutual NDA v3; if unilateral and counterparty is vendor, use Vendor Unilateral NDA v2 with fallback clauses A/B; if data is sensitive (PII/health), require DPA link and privacy review.
Encode redlines: Clause-level rules for what to accept, counter, or escalate—plus the rationale.
Attach evidence: Link positions to prior approved deals and risk notes so every answer has a source.

In Sandstone’s knowledge layer, these positions become living rules an AI agent can apply, cite, and learn from, with every approval strengthening the next decision.

Put AI Agents in the Loop—With Guardrails

AI agents shine when they operate inside a well-defined workflow.

Classify: Detect request type and risk signals from unstructured messages.
Enrich: Extract entities (counterparty, systems, data types) and fill missing fields by asking the requester concise follow-ups.
Retrieve: Pull the relevant position, template, or clause from the knowledge layer.
Propose: Generate the next step—approve, template, or redline—with an explanation and links to sources.
Escalate: Hand off edge cases with a clean summary, suggested options, and a confidence score.

Every action should be logged. Humans approve, the system learns. The outcome is consistent decisions delivered faster, with a transparent “why” that builds trust.

Measure What Scales

Track a small set of metrics that reflect business value, not just activity.

Time to First Response (TFR): From intake to acknowledgement.
Request-to-Decision Time: End-to-end resolution per request type.
Self-Serve Rate: Percentage resolved without lawyer intervention.
Deflection Quality: Reopened or escalated after self-serve.
Playbook Coverage: Share of volume mapped to executable positions.
Confidence and Citation Use: Agent confidence vs. human overrides; percentage of decisions with sources.

These metrics show whether knowledge is compounding and where policy gaps remain.

A One-Week Pilot: NDA Intake to Decision

You don’t need a year-long program to see value. Stand up a focused pilot.

Day 1–2: Choose NDA as your high-volume workflow. Define required fields and map the golden path. Agree on SLAs.

Day 3–4: Convert your NDA guidance into executable positions with fallbacks. Attach templates, clauses, and prior approvals as evidence.

Day 5: Configure intake in Slack/Teams and email. Enable the AI agent to classify, enrich, and propose the right NDA path with rationale.

Day 6–7: Launch to one business unit. Monitor TFR, Request-to-Decision, and Self-Serve Rate. Capture edge cases to refine positions.

On Sandstone, this becomes a reusable module. Add vendor DPAs next, then low-risk procurement terms, reusing the same layered data and triage patterns.

Actionable Next Step

Pick one workflow with high volume and low variance—NDAs, low-risk MSAs, or marketing reviews. In one hour, write five if/then positions with clear fallbacks and attach the right templates. Turn that into a channel-native intake and set a two-day SLA. Measure three metrics for two weeks: TFR, Request-to-Decision, and Self-Serve Rate. Iterate.

The Bedrock of Speed, Alignment, and Trust

Intake to Decision: Building an AI‑Ready Legal Workflow That Scales

Why This Matters Now

Design the Golden Path: Standardize Intake and Triage

Measure What Scales

A One-Week Pilot: NDA Intake to Decision

Day 1–2: Choose NDA as your high-volume workflow. Define required fields and map the golden path. Agree on SLAs.

Day 3–4: Convert your NDA guidance into executable positions with fallbacks. Attach templates, clauses, and prior approvals as evidence.

Day 5: Configure intake in Slack/Teams and email. Enable the AI agent to classify, enrich, and propose the right NDA path with rationale.

Day 6–7: Launch to one business unit. Monitor TFR, Request-to-Decision, and Self-Serve Rate. Capture edge cases to refine positions.

Actionable Next Step

About Jarryd Strydom

Intake to Decision: Building an AI‑Ready Legal Workflow That Scales

Why This Matters Now

Design the Golden Path: Standardize Intake and Triage

Measure What Scales

A One-Week Pilot: NDA Intake to Decision

Day 1–2: Choose NDA as your high-volume workflow. Define required fields and map the golden path. Agree on SLAs.

Day 3–4: Convert your NDA guidance into executable positions with fallbacks. Attach templates, clauses, and prior approvals as evidence.

Day 5: Configure intake in Slack/Teams and email. Enable the AI agent to classify, enrich, and propose the right NDA path with rationale.

Day 6–7: Launch to one business unit. Monitor TFR, Request-to-Decision, and Self-Serve Rate. Capture edge cases to refine positions.

Actionable Next Step

About Jarryd Strydom