Serious 2026 vendor guides keep recommending an 8 to 12 week pilot before multi-year commitment. Buyers who skip the pilot usually discover process gaps after the MSA is signed and switching costs are high.
This is a practical 90-day plan to pilot an outsourced engineering capability without turning the pilot into unpaid discovery theater.
Table of Contents
Days 0 to 15: design the pilot
- Write the model memo: staff aug, dedicated, or managed. Use the models article and playbook scorecard.
- Pick a thin vertical slice with real users or a real internal consumer. Avoid "research spikes" with no acceptance tests.
- Define exit criteria in writing: quality bar, communication SLAs, security baseline, and commercial next step.
- Lock access, AI-tool policy, and credential handoff paths before day one.
If the pilot touches production IP or customer data, treat AI as infrastructure, not a browser extension. AI architect guidance from Keith Associates on private AI systems inside a client boundary is a practical bar for the day-15 checklist: approved models, VPC or tenant limits, and an audit path before the first useful sprint.
Days 16 to 60: run it like production
- Same board, same Definition of Done, same CI gates as your internal teams where possible.
- Weekly tactical review; biweekly commercial/health review.
- Measure onshore management hours actually spent. If staff aug burns 25% of an EM, note it.
- Demand artifacts: PR quality samples, incident notes, test evidence.
Days 61 to 90: decide with evidence
At day 60, compare predicted TCO to actuals. At day 90, choose one: expand, convert model (for example aug to dedicated), or exit. Do not "extend the pilot" without new hypotheses. That is how temporary becomes accidental permanent.
Metrics that matter
| Metric | Why |
|---|---|
| Cycle time / lead time | Shows whether collaboration tax is real |
| Escaped defects | Quality bar honesty |
| Onshore management hours | Exposes fake savings |
| Response time to blockers | Timezone and process fit |
| Attrition or shadow churn | Continuity risk early |
Fail signals
- Vendor cannot name who owns outcomes
- Security exceptions "just for the pilot"
- No written backfill plan
- Demo theater without inspectable board
- Pressure to sign annual volume before day 60 evidence
Take the playbook ZIP into the kickoff meeting. Use the RFP bank and contract clauses before you expand.