Section A · First 90 days

The 90-Day Plan

Week-by-week milestones for a contract-intelligence deployment. Four phases, concrete outputs per week, the moves that win or lose each phase.

The four phases

Phase	Days	Goal	Sign of success
Discovery	1–14	Understand the customer's data, stakeholders, and decision context	Signed one-page working agreement
First slice	15–30	Ship the smallest useful end-to-end artifact	Customer can act on the output
Scale to SLA	31–60	Expand coverage, hit accuracy targets, integrate with the customer's stack	SLA holds for 30 consecutive days
Mature & plan handoff	61–90+	Runbooks, training, customer-team enablement, sunset of FDE involvement	Customer's team can resolve the top-5 issues without you

The numbers shift per deployment. A small, well-scoped customer might compress to 60 days total. A large, multi-business-unit customer might run 6+ months. The phase shapes are the same.

Days 1–14: Discovery

Goals

Map the customer's stack, stakeholders, and data reality.
Identify the highest-leverage use case for the first slice.
Ship a signed one-page working agreement.

Week 1

Day 1: kickoff call with the customer's executive sponsor and data lead. Set the meeting cadence. Get access requests started (identity, sample documents, warehouse).
Day 2: file identity / IT tickets — Okta SSO, SCIM, sample-data access. These have the longest lead time; start now.
Day 3–4: interview the customer's data analysts, procurement / legal analysts, the executive sponsor (often three separate conversations). Use the question banks in 02-discovery-template.
Day 5: read 20+ representative documents from the customer's corpus. Take notes on layout variance, completeness, age range.

Week 2

Day 6–7: map their existing stack — ERP, CLM, warehouse, BI, identity. Draw a diagram. Verify with their data team.
Day 8: run an OCR-quality preflight on a 50-document sample. Eyeball the output.
Day 9: synthesize findings into a draft working agreement. One page. Objectives, scope, success criteria, owners, kill criteria. Use the template in 03-working-agreement.
Day 10–12: review draft with the customer's data lead, then the executive sponsor. Iterate.
Day 13–14: sign the working agreement. This is the milestone that closes discovery.

Don't write code in discovery

The temptation is real. Resist it. Code written in week one is almost always rework — the schema you'd build before talking to stakeholders is the schema you'll throw away by week three. Discovery is conversation and reading.

What you should know by end of week 2

The five most important document classes in the customer's corpus.
What manual workflow the platform is replacing, and how many hours per week it costs them today.
Who at the customer signs off on the data being right.
Where the output needs to land for the customer to act (BI dashboard? ERP? procurement suite? all three?).
What freezes / change windows constrain when you can deploy.

Days 15–30: First slice

Goals

Stand up ingestion for one document source.
Run extraction against one document class with HITL.
Land output in the customer's staging environment.
Customer can see and act on a first useful result by end of week 4.

Week 3

Build ingestion connector for the customer's chosen source (drop bucket, CLM API, file share).
Configure extraction for one document class.
Wire HITL queue with the customer's analyst capacity in mind.
Begin landing data into a staging schema in the customer's warehouse.

Week 4

Run end-to-end on 100–500 documents.
Build one starter dashboard in the customer's BI tool showing extracted output.
Walk the customer through the dashboard live. Capture corrections.
Ship Friday status doc with "this is what you can see today" framing.

The "smallest useful slice" principle

Customers care about outcomes, not coverage. A working pipeline for one document class with 60 dashboards on the data beats a half-built pipeline for ten classes. Resist the urge to ingest everything; ship something the customer can act on.

Days 31–60: Scale to SLA

Goals

Expand to all document classes the working agreement scoped.
Hit the contracted accuracy SLA per class.
Wire output into the customer's downstream systems (warehouse marts, BI, possibly ERP write-back).
Begin training the customer's data team.

Weeks 5–6

Onboard remaining document classes one at a time.
Calibrate HITL thresholds per class against measured accuracy.
Start building the customer-facing SLA dashboard (extraction quality by class, HITL queue health, refresh cadence).

Weeks 7–8

Build the mart layer in the customer's warehouse using dbt — renewal pipeline, off-contract spend, supplier rollups.
Hand the customer's analysts the semantic layer; pair-session through how to extend it.
Validate against the SLA on a labeled audit sample. If any class is below, contain and remediate.

Sign of phase completion

The SLA dashboard shows accuracy above contracted thresholds for all in-scope classes for at least 30 consecutive days. The customer's data team has accessed the marts independently and built at least one of their own queries on top.

Days 61–90: Mature & plan handoff

Goals

Runbooks live, tested, and owned by the customer's team.
On-call posture shifted from FDE-primary to joint.
Customer success picks up the relationship for ongoing maintenance.
Plan written for the FDE's exit and future expansion conversations.

Weeks 9–10

Write the runbooks (see 06-runbook-templates): pipeline failure, quality alert, schema drift, backfill, new doc class, HITL queue overflow.
Pair-test the runbooks with the customer's data team. They follow; you observe.
Document the customer-specific extension configuration (their doc-class taxonomy, their threshold tunings, their mart definitions).

Weeks 11–12

Transition to joint on-call. They handle first response with your runbooks; you escalate only when they can't resolve.
Hand off the relationship to customer success. Joint introduction call.
Write the expansion proposal — additional business units, additional document classes, deeper integration — for the AE to take into renewal.
Ship the deployment summary doc to internal leadership: what was built, what was learned, reusable artifacts for the next deployment.

Sign of phase completion

The customer's data team has resolved at least one incident from the runbooks without paging you. Customer success has the relationship. You have a clear next-deployment.

Milestone checklist

The hard milestones, in order. If you slip past one, surface to internal leadership before slipping past the next.

☐ Day 14: signed one-page working agreement
☐ Day 30: customer can see useful output from the first slice in their staging environment
☐ Day 45: all in-scope document classes onboarded
☐ Day 60: SLA holds for 30 consecutive days
☐ Day 75: runbooks written, tested with customer's team
☐ Day 90: joint on-call active; customer success has the relationship; expansion proposal drafted

Variations per deployment

Smaller deployments (one team, one document class)

Compress to 45–60 days. Skip the multi-class onboarding phase. Heavier focus on the customer-team handoff because there's no "expansion to other BUs" to fill the time.

Larger deployments (multi-business-unit, multi-region)

Expand to 6+ months. The discovery phase often runs three weeks per BU. Stagger the first slices — don't try to ship for all BUs simultaneously. Pick a lighthouse BU, prove the pattern, then replicate.

High-stakes / regulated deployments

Add a security-and-compliance phase between discovery and first slice. SOC 2 attestation reviews, BAA / DPA execution, customer-side security questionnaires. Plan an extra two weeks; some of it lives in parallel with other phases but the calendar slip is real.

Reseller / partner-driven deployments

The partner (often a Big Four consultancy) is the customer-of-record interface. Working agreement scope changes — your customer is the partner, but the SLA holds against the end-customer's data. Trickier to navigate; the discovery phase needs to include the partner's project lead as a primary stakeholder.