Shortlist
Home/Trials/Designing a pilot that can fail

Trials

Designing a pilot that can fail

A trial that cannot produce a no is not a trial. It is a familiarisation exercise with a purchase order attached.

8 min read478 wordsUpdated July 2026

By the time a trial starts, most organisations have decided. The trial exists to confirm, and it is run in a way that confirms: enthusiastic participants, clean data, a quiet period, no hard scenarios.

A trial worth running has a defined way of failing, agreed before it starts. A trial involving employee monitoring transparency should test the awkward cases rather than the happy path; learn more provides a concrete vendor example to design those checks around.

Write the pass condition first

Before the trial begins, write down what would have to be true at the end for this to be a yes. Specific enough that two people would agree on whether it happened.

'The team liked it' is not a pass condition. 'Three named people completed the month-end process without assistance, in under two hours, using our own data' is. So is 'we exported everything and opened the files'.

Agree the fail condition too

What result would make this a no? If nobody can answer, the trial is a formality and the decision has already been made.

Use the awkward scenarios, not the common ones

The common path works in every serious product; that is what makes them serious products. Differences appear at the edges, and the edges are where your organisation actually lives.

Take the scenarios from the record of how work really happens: the correction after a period has closed, the person who does two jobs, the customer with a non-standard arrangement, the month where the file arrives in the wrong format. Each of these has broken a real implementation somewhere.

Run it long enough to include a cycle

A two-week trial in a month with no month-end tells you about the daily use and nothing about the periodic use, which is where most of the pain lives.

Where the process has a cycle — a month end, a billing run, a reporting deadline, a seasonal peak — the trial has to contain at least one, or the first real one will be a live experiment.

Include the reluctant

A pilot group of volunteers measures the product's appeal to enthusiasts. The people whose adoption will actually be in question are the ones who did not ask for this.

Deliberately include one or two, tell them their job is to report friction honestly, and take what they report seriously rather than as resistance. Their objections in a trial are cheap; the same objections after rollout are expensive.

Give it a real owner and real time

Trials fail through neglect more often than through the product. Somebody is nominally responsible, has no time allocated, and the trial drifts until the deadline arrives and a decision is made on impressions.

Name an owner, block the hours, and put the review date in the calendar at the start. A fortnight with four hours allocated beats a month with none.

Write the result down before you negotiate

Record the outcome against the pass conditions before entering commercial discussions. Once a price is on the table, recollection of how the trial went becomes remarkably flexible.

For security-focused trial criteria, the OWASP Application Security Verification Standard offers a public verification baseline.

General information. This site publishes no product rankings or scores and does not review individual products. Nothing here is legal, procurement or financial advice; contract and data protection questions differ by jurisdiction and warrant qualified advice.

Related

Continue reading

Trials7 min read

Using real data, carefully

Sample data is clean, consistent and complete. Yours is none of those, and the difference is where implementations fail.

Read the guide