Routine NDA review
88 to 96% lowerAbout an hour of review, at in-house to outside-lawyer rates. Sources: ACC, Clio.
Turn $15/hr generalists into verified domain experts. Without $350/hr billable dependencies.
Outsource your most complex, high-liability work, from contracts and tax filings to SOX audits and clinical prior auth, to screened operators backed by AI.
Request a 50-case pilot- A person checks every case
- A licensed reviewer takes every exception
- Billed per completed task
Try the operator’s view
Simulated case, no real data
- Request
- Policy rule
- Confidentiality may survive termination for up to 3 years.
- What the AI found in the contract
AI guidance
Pick a verdict. You can press 1 or 2 on your keyboard.
Case 1 of 2
Every task follows the same three steps.
We give each step to whoever is most likely to get it right.
-
Break down the situation
AI first, a person confirmsAI reads every page. A person spots what is missing.
Vendor NDA: three-year term, Delaware law. -
Apply the right knowledge
AI leads, a person spot-checksAI searches tax codes, contracts and precedent without tiring.
Survival clause runs 5 years. Policy allows 3. -
Make the judgement call
A person decidesPeople weigh ambiguity and own the outcome.
Accept 5 years for a critical vendor?
We route each step by measured accuracy. On hidden control cases we score AI alone, a person alone, and both together, then use whichever is right most often. Low confidence goes to a licensed reviewer.
In one published NDA study, AI caught 94% of issues and lawyers 85% (LawGeex, vendor-run).
Résumés claim. Cases prove.
We look for people who read carefully, reason well under ambiguity, and keep going until the case is right.
Top consulting firms don’t hire on résumés alone. They run structured cases that show how a candidate breaks down a problem, reasons through ambiguity and keeps going when it gets hard. Our screening follows that recipe, adapted to the work our operators do. Every candidate gets the same cases, scored on the same written rubric.
Work-sample case
A real-style case, scored against a known answer.
Can they do the job?Problem-solving test
A structured exercise that tests how they break a situation down.
Do they think in structure?Persistence task
A deliberately ambiguous case. We measure whether they dig in or give up.
Do they keep going?Paid trial
Live work on control cases before any customer case.
Do they hold up on real work?
Structured, job-like tests predict performance best: Sackett et al., 2022, Journal of Applied Psychology.
Quality is measured on every case, not promised.
One system follows each case from the operator’s first exam to your weekly report. Each stage checks the one before it.
-
1
Qualify
Entry examOper8s pass a scored exam before they touch a customer case.
-
2
Intake
Hidden control casesCases with known answers are mixed into the queue. Oper8s can’t tell which.
-
3
Decide
Time-on-evidence check, confidence scoreA minimum inspection time stops one-click approvals. Every task gets a confidence score.
-
4
Verify
Two-round consensus, licensed reviewerTwo operators decide independently, and disagreements get a third look. Low confidence goes to a licensed reviewer.
-
5
Audit
QA sampling, weekly reportA reviewer audits a sample of approved cases every week, and you see the results.
One audit trailA time-stamped log follows each case through every stage.
Feedback loopControl-case and QA results update each operator’s score and routing. Weak results trigger retraining.
You see this every week
Control-case accuracy, escalation rate, and rework rate. Our targets: control-case accuracy of 98% or higher, escalation rate of 15% or lower, rework rate of 2% or lower.
Up to 90% lower cost than specialist billing. Billed per completed task.
A flat fee per finished case, with a credit if accuracy falls below the agreed floor. Savings are largest where specialists bill by the hour.
Prior authorization
About 38% lowerManual cost per request for providers. Source: CAQH Index.
Billing review before submission
Pays off at 1 prevented denial per 14 to 67 claimsAbout 1 in 8 claims is denied on first submission. Sources: Experian, MGMA, HFMA.
Every price is confirmed on your own cases in the pilot before you commit. If control accuracy falls below 95% in any week, that week’s fees on the affected case type are credited in full.
Also available on request: internal audit three-way match and cross-border tax exemption, priced in your pilot. Oper8 prices are targets confirmed in the pilot. Sources: Clio, ACC, CAQH Index, MGMA, HFMA, Experian.
What your security and compliance leads will ask.
Your data
Sensitive identifiers are masked before an operator sees a record. Each customer is isolated in its own tenant. Documents are deleted within 24 hours of filing. Your data is never used to train models. Verified in your pilot.
Who is licensed
Oper8s are not licensed professionals. Where the law requires a licensed person, a licensed reviewer signs the case. Reviewer credentials: REPLACE. Oper8 location, HIPAA training and background checks: REPLACE
Who is liable
Liability, insurance and indemnity terms are set in the contract before the pilot starts: REPLACE
SOC 2: not yet audited, Type I planned REPLACE date
HIPAA BAA: available on request
Intake: API, SFTP or ERP webhook
Effort on your side: one intake channel
Prove it on your own cases before you commit.
A 50-case pilot shows the accuracy, the cost and the savings on work you already know the answers to.
Send 50 past cases
Pick them from your own history, under NDA.
We run them blind
Our answers are produced without seeing yours.
You compare
Accuracy, cost and turnaround, side by side with your own records.
You decide
Keep the full report. Continue only if the numbers hold.
No commitment before the numbers.If control accuracy falls below 95% in any week, that week’s fees are credited.
Questions we expect.
Is this legal?
For contract review, operators prepare and check the work and a licensed reviewer takes every exception. Rules on who may prepare or sign legal, tax and healthcare work vary by state and task, so we confirm them with you before any pilot.
Why not just use AI alone?
AI alone has no accountable reviewer and no measured accuracy on your cases. We use AI where it is stronger, people where judgement matters, and we measure both.
Who is liable if a case is wrong?
The contract sets it out before the pilot starts. A credit applies automatically if control accuracy falls below the agreed floor.
Where does our data go?
Into an isolated tenant, with identifiers masked before operators see it, and deleted within 24 hours of filing. It never trains a model.
What if two operators disagree?
The case goes to a third reviewer. The disagreement is logged and shows up in your weekly escalation rate.
What does the pilot involve?
You send 50 historical cases. We run them, you compare our answers and turnaround to your own records, and you get a report with measured accuracy and projected savings. With 50 cases the report shows a confidence range, not a single number. Pilot fee: REPLACE. Results within REPLACE weeks of receiving your cases.
Test us on 50 of your own cases.
Send historical cases, compare our answers with yours, and decide with measured results. Tell us what you’d like to test and we’ll reply to scope it.