The Ultimate Guide to AI Intake Triage for In‑House Legal Teams

Across mid-market and enterprise teams, our benchmarks show 35–55% of legal requests arrive missing basic context—contract counterparties, jurisdictions, or urgency—adding days of avoidable delay. Intake isn’t just a form problem; it’s a knowledge problem. The fix isn’t more fields. It’s an AI agent that understands your playbooks, routes with precision, and learns from every decision.

Why Intake Breaks at Scale

When demand grows faster than headcount, intake becomes a noisy switchboard. Common failure modes:

- Ambiguous request types ("need legal to review") without the right metadata

- Serial follow-ups to gather facts, creating hidden queues and context switching

- Misrouted work (privacy vs. commercial vs. employment) and unclear SLAs

- Fragmented knowledge—positions live in email threads, not in a shared brain

At scale, the cost isn’t just slower cycle times; it’s trust erosion. Sales codes around legal, procurement piles up risk debt, and leadership sees legal as a bottleneck. The antidote is a triage layer that’s consistent, explainable, and aligned to how your business actually operates.

What an AI Intake Agent Actually Does

An effective intake agent isn’t generic chat—it's your institutional knowledge operationalized. On a platform like Sandstone, it:

- Gathers the right facts: Dynamically asks only the questions needed based on the request type, using your matter taxonomy.

- Classifies and routes: Maps requests to the correct queue (e.g., NDA vs. DPA vs. MSA amendment) with confidence scoring and fallback rules.

- Applies playbooks: Pulls positions and clause guidance from your knowledge layer to propose next steps or auto-resolve simple matters.

- Enforces SLAs: Acknowledges receipt, sets expectations, and escalates if thresholds are breached.

- Logs decisions: Captures the who/what/why for auditability and learning.

This is strength through layers: layered data (request metadata), layered decisions (triage → routing → suggested action), and layered knowledge (playbooks that get sharper with each outcome). Crafted precision comes from encoding positions at the right level of abstraction—what you will and won’t accept—so the agent can act without guessing.

Playbooks You Can Automate on Day One

Start with high-volume, low-variance work where rules are clear and risk is bounded:

- NDAs: Auto-classify unilateral vs. mutual, propose your standard template or redlines, and route only exceptions (e.g., governing law changes) to counsel.

- Vendor Intake: Collect data processing facts, trigger DPA/TRA workflows, and pre-check security artifacts before legal review.

- Marketing Review: Detect claims vs. puffery, flag required disclosures, and route regulated content (finance/health) to the right reviewer.

- Low-risk Amendments: Identify non-material changes (e.g., contact info) and provide a self-serve path with guardrails.

Each playbook pairs decision logic with the source of truth: clause libraries, fallback positions, approval matrices, and jurisdictional quirks. In Sandstone, these live as modular steps the agent composes at run time—natural integration with how your team already works.

KPIs and Guardrails That Matter

Measure what the business feels and what the board expects:

- Time to First Touch: Instant acknowledgement with accurate triage target: <5 minutes.

- Cycle Time by Request Type: Trend lines pre/post automation; aim for 30–60% reduction on target playbooks.

- Auto-Resolution Rate: Percentage closed without attorney time; start at 20–30% for NDAs.

- Reopen/Exception Rate: Quality barometer—keep <5% for automated matters.

- SLA Adherence: By queue and by region; surface risks before they breach.

Guardrails keep speed aligned with risk:

- Confidence thresholds and routed fallbacks

- Mandatory human review on sensitive categories (employment termination, IP assignment, regulatory notices)

- Versioned playbooks with audit trails and approval lineage

- Data governance: least-privilege access and redaction on ingestion

A 30‑Day Pilot Plan

Week 1: Baseline and Scope

- Pull 90 days of intake data. Identify two request types with high volume and clear rules (e.g., NDAs, marketing review).

- Extract current positions, clause fallbacks, and SLAs. Define “auto-resolve” criteria.

Week 2: Configure and Test

- Build dynamic intake forms and routing rules. Connect your template library and approval matrix.

- Run the agent in shadow mode. Compare classifications, proposed actions, and outcomes.

Week 3: Limited Launch

- Roll out to one business unit. Publish SLAs and a simple escalation path.

- Track KPIs daily; capture exceptions to refine playbooks.

Week 4: Expand and Institutionalize

- Broaden coverage; add one more playbook (vendor intake or low-risk amendments).

- Formalize change governance: who edits playbooks, how approvals are recorded, and how learnings feed back.

Actionable takeaway: Stand up a shadow-mode intake agent for NDAs in two weeks. If auto-resolution is <20% or exception rate >10%, your playbook isn’t specific enough—tighten positions before scaling.

The Foundation: Turning Intake into Institutional Knowledge

Great intake isn’t a form; it’s a memory system. Every request teaches your operating system what “good” looks like—what to ask, how to route, where risk hides. Sandstone turns those lessons into reusable layers: modular workflows, living playbooks, and decisions that compound instead of disappearing. That’s how legal stops being a checkpoint and becomes the connective tissue of the business—faster cycles, cleaner risk posture, and durable trust at scale.