How to run a low-risk outsourcing pilot: 1–2 developers, 1–2 months
The pilot is the cheapest way to test an offshore partner: how to scope it, what to measure, the contract terms to insist on, and how to read the results.
The cheapest way to find out whether an offshore partner is any good is to buy a small amount of their normal work: one or two developers, one or two months, a real but bounded task, under the same contract terms you would accept at scale. A pilot converts sales claims into observed behaviour - code quality, communication, estimate honesty - for the price of a few weeks of development. Every remote-partner failure mode is visible within a month if you look at the right things. Here is how to scope, contract and judge one.
Why a pilot beats every other evaluation method
References are curated. Portfolios show the successes. Sales calls feature the best communicators, who may never touch your account again. A pilot samples the thing you are actually buying - the vendor’s routine working behaviour - under production conditions. It also tests you: whether your requirements, review process and availability are ready for a remote team. Finding your own gaps on a two-person pilot is a bargain; finding them on a ten-person program is not.
What makes a good pilot project?
Four properties, in tension you must balance:
- Real. Production code your team will review and maintain - not a toy exercise. Toy projects measure toy behaviour.
- Bounded. A natural finish line 4–8 weeks out: an integration between two systems, a self-contained module, a test-automation suite, a well-scoped data migration.
- Representative. In your stack, against your standards, through your review process - otherwise the result doesn’t transfer.
- Off the critical path. Meaningful enough to take seriously, not so central that a disappointing outcome hurts the roadmap.
The classic mistake is piloting with the least important work you can find - stale requirements, no internal owner, nobody reviewing. The vendor’s questions go unanswered for a week and you conclude “communication was poor”. The pilot measured you.
What terms should govern a pilot?
The same ones you’d want at scale - that is half the test. Insist on:
- Full contract fundamentals: IP assignment, NDA, and exit terms exactly as they would be in the long-term agreement. A vendor who wants “lighter” terms for the pilot is telling you the real terms will be a fight.
- Normal rates. Free or heavily discounted pilots select for desperation and get staffed accordingly. You are buying information; pay the market price for it.
- Named people. The developers on the pilot should be people who could stay - and the conversion path (pilot team becomes core of the dedicated team) agreed in advance.
- Your infrastructure. Code in your repositories, tickets in your tracker, access revocable by you - from day one, so scaling up changes nothing.
What should you measure?
Write the scorecard before the start. Ours, from the buyer’s side:
| Dimension | What good looks like | What failure looks like |
|---|---|---|
| Code quality | Passes your review standards without special pleading | ”It works” but reviewers wince |
| Communication | Questions early, written status, bad news volunteered | Silence, then surprises |
| Estimates | Assumptions stated; misses explained and re-planned | Optimism, then quiet slippage |
| Problem handling | Issues surfaced with options | Issues hidden until discovered |
| Your own side | Requirements answered within a day | Vendor blocked on you for days |
That last row is not a courtesy - it is the most common cause of failed pilots, and it is fixable before you scale.
How do you read the results?
- Clear pass - quality and communication both solid: convert the same people into a standing team; don’t restart with strangers and throw the context away.
- Clear fail - either dimension poor under fair conditions: stop, pay, part politely. The pilot did its job at minimum cost; changing vendors after a pilot costs nothing compared with after a year.
- Mixed - usually strong work, rough process (or the reverse). Diagnose which side owns each gap, fix what’s yours, and run a second month with explicit expectations. One iteration is legitimate; a third month of “mixed” is a fail being rationalized.
The discipline that makes all three readings possible: decide the thresholds before the pilot starts. Afterwards, sunk cost and optimism will negotiate on the vendor’s behalf.
In short
- A pilot buys observed behaviour - the only evaluation data that predicts an engagement.
- Scope it real, bounded, representative, off the critical path; 1–2 developers, 4–8 weeks.
- Run it under full contract terms and normal rates - light terms and free work invalidate the test.
- Score code, communication, estimates, problem handling - and your own responsiveness - against criteria written in advance.
- Pass → scale the same people. Fail → cheapest exit you’ll ever get. Mixed → fix, iterate once, then decide.
AD System Asia proposes a pilot to most new clients before any large commitment - same contract, same rates, named people who stay if you scale. Tell us what a fair test would look like on your side and we’ll scope one.
Frequently asked questions
What is an outsourcing pilot?
A deliberately small first engagement - typically one or two developers for one or two months on a real, bounded piece of work - run under the same contract terms you would use at scale. Its purpose is to test code quality, communication and management fit at a cost that doesn't need a steering committee.
What is a good pilot project for testing an offshore vendor?
Real work with clear boundaries: an integration, a self-contained module, a test-automation suite, a well-scoped migration. It should touch your review process and standards but not sit on the critical path - meaningful enough to reveal quality, small enough to cancel without pain.
How do I evaluate the results of an outsourcing pilot?
Against criteria written down before the start: code quality under your review standards, communication (questions asked early, honest status), estimate accuracy, and how problems were handled when they appeared. Decide the thresholds first, or the sunk cost will decide for you.
Should a pilot be free or discounted?
Neither. A free pilot selects for vendors desperate enough to work unpaid and tells you nothing about the real working relationship. Pay normal rates under normal terms - the information a genuine pilot produces is worth far more than the discount.
What happens after a successful pilot?
Scale the same people, don't restart: convert the pilot developers into the core of a dedicated team so the context they built carries forward. The conversion path - and the contract terms staying identical - should be agreed before the pilot begins.