Cycle time is the metric every legal operations team reports, and intake is where most of it is lost. A request that arrives as a forwarded email with an attachment and the words can you look at this will sit in a queue, get reassigned twice and lose a week before anyone opens the document. The drafting was never the bottleneck.
Intake is also where the contract record is born. Whatever is captured at that moment, counterparty entity, value, term, governing law, becomes the metadata the organization relies on for reporting, renewals and obligation tracking. Get it wrong at intake and every downstream process inherits the error, usually without anyone noticing until a portfolio question cannot be answered.
A working contract intake process does three things. It collects enough structured information to route the request correctly, it applies a triage rule that separates the routine from the negotiated, and it starts the right workflow automatically. None of that requires artificial intelligence, though AI makes each step faster and cheaper to run at volume.
What a broken intake process looks like
Teams rarely describe intake as their problem. They describe the symptoms, and the symptoms are consistent enough across organizations to be diagnostic on their own.
- Requests arrive through many doors, email, chat, a shared inbox and hallway conversations all feed the same queue, so there is no single record of what was asked or when it was asked.
- The request is incomplete, the requester does not know which fields matter, so the first exchange is spent asking for the counterparty entity, the term and the commercial value.
- Everything is triaged by a human, a senior reviewer reads every request to decide whether it is a standard NDA or a negotiated master agreement, which is expensive triage for a routine decision.
- Priority follows the loudest requester, urgency is set by how hard someone chases, so a low-value renewal outranks a strategic agreement that nobody is escalating.
- Nothing is measured, with no timestamped intake record the team cannot say where time goes, so improvement efforts target drafting instead of the queue that precedes it.
The common thread is that intake is treated as an inbox rather than as a process with defined inputs, rules and outputs. Once it is modeled as a process, most of the delay becomes addressable without adding a single reviewer to the team.
Fields captured at intake are the fields that populate the repository, drive renewal alerts and feed obligation tracking. Treating intake as a form to get past, rather than as the first and cheapest opportunity to capture clean data, is what produces repositories that cannot answer basic portfolio questions two years later.
Designing the intake form and the triage rules
The form should be short enough that requesters complete it and structured enough that the answers can drive routing without interpretation. Conditional logic does most of the work, revealing procurement or privacy questions only when they are relevant.
- Ask for the entity, not the name. Counterparty legal entity, jurisdiction and any existing relationship, so duplicate suppliers and live master agreements surface before drafting starts rather than after.
- Capture commercial shape early. Annual value, term length and renewal type determine approval thresholds, and they are the fields most often missing when a request first reaches a reviewer.
- Separate the paper source. Whether the agreement sits on your template or the counterparty's changes the review path completely, and it is a one-click question for the requester to answer.
- Route by risk, not by document type. A low-value NDA on your template can self-serve; the same NDA with an unusual term or a regulated counterparty needs review, and the rule should say so explicitly.
- Give every request a service target. A published turnaround per request type sets expectations and gives the team a measurable commitment instead of an implicit and permanently disappointing one.
A five-step roadmap to rebuild intake
Step 1, close the side doors
Publish one intake channel and route everything else to it, including a mailbox rule that replies automatically with the link. Until requests arrive in one place, no measurement or automation is possible. Expect resistance for a few weeks and hold the line through it.
Step 2, model the request types
List the ten request types that make up most of your volume, then define for each one the fields required, the approvals needed and the target turnaround. This list, rather than the software selection, is the real design work and it belongs to the team, not the vendor.
Step 3, write the triage rules
Convert the routing decisions your senior reviewers make by instinct into explicit conditions on the intake fields. Anything that can be decided from structured answers should not require a human to open the document at all.
Step 4, automate the self-serve path
Route routine requests straight to template generation and signature with no review step. This is where cycle time drops most visibly, and it releases reviewer capacity for the agreements where judgment genuinely adds value.
Step 5, instrument and iterate
Timestamp every state change and review the queue monthly. The first month of data usually shows the largest delay sitting between submission and first touch, which is a routing and staffing question rather than a drafting one.
See intake, triage and generation in one workflow
Aavenir routes requests from a structured intake form to self-service generation or guided review based on rules you define and control.
Where AI belongs in intake
The useful applications of AI at intake are narrow and worth doing. Extraction can read attached third-party paper and pre-populate the intake fields, which removes the most common cause of incomplete requests. Classification can propose a request type and a risk tier from the document itself, giving the triage rule a starting point rather than a blank form. Retrieval can surface an existing master agreement with the same counterparty, preventing the duplicate agreements that clutter most repositories. What AI should not do is make the final routing decision unreviewed on high-value or regulated agreements, because the cost of a misroute there is measured in weeks. Treat the model as a fast first pass and keep the rule engine authoritative. The same logic applies to AI contract review further down the process.
Triage tiers and what each one needs
Most portfolios sort cleanly into four or five tiers. Defining them explicitly is what allows the majority of requests to bypass review entirely without anyone feeling that control has been lost.
| Tier | Typical request | Path | Target turnaround |
|---|---|---|---|
| Self-serve | Standard NDA on own template | Automated generation and signature | Same day |
| Light review | Low-value order form, own paper | Checklist review by contract manager | Two business days |
| Standard review | Third-party paper within policy | Playbook-guided review | Five business days |
| Negotiated | Master agreement or regulated counterparty | Assigned counsel with approvals | Scoped per matter |
| Escalated | Non-standard indemnity or liability cap | Counsel plus risk sign-off | Scoped per matter |
Intake is unglamorous and it is where the leverage sits. A team that publishes one channel, defines ten request types and automates the routine tier will usually cut its median cycle time before touching how anything is drafted or negotiated. It also produces something the organization has probably never had, a timestamped record of what was asked, by whom and how long each step took, which becomes the evidence base for every subsequent improvement argument and the fastest way to settle an internal debate about where capacity is actually going. Pair it with a repository that keeps the captured data usable; the difference between a repository and a CLM is worth settling before you buy anything.
For related guidance, see Multilingual contract review that legal ops can trust, The legal ops guide to legacy contract migration and Contract retention policy: a guide for legal ops.
