An agency pilot exists because hiring an agency is otherwise a decision made with almost no information. References are selected, case studies are survivors, and the pitch measures presenting. The only reliable evidence is the work itself — which argues for buying a small, defined piece of it before committing to the rest.
Done badly, a pilot is a thirty-day trial that proves nothing and sours both sides. Done well, it is the cheapest due diligence available.
Why ninety days, not thirty
The first fortnight goes on access, tracking and structure. Bidding then needs a volume of conversions before it stabilises, and that takes weeks rather than days in most accounts. A thirty-day pilot therefore measures onboarding, which is worth knowing but is not the thing you are buying. The realistic timeline is the constraint here, and it is arithmetic rather than opinion.
Ninety days gets you through setup, through learning, and into one genuine read of what works. If your sales cycle is long, the pilot measures lead quality signals rather than closed revenue, and you should agree that up front.
Narrow the scope deliberately
A pilot that tries to cover every channel tests nothing properly. Pick one of these shapes:
- One channel, one market. The cleanest. Usually your largest or most broken channel.
- One segment. A product line, a city, a customer type — anything that can be isolated and measured.
- One problem. Tracking rebuild, feed overhaul, account restructure. Appropriate when you already know what is wrong.
Leave everything else running as it is. That gives you something to compare against, which is the part most pilots skip.
Agree the success criteria before it starts
Write down, in one paragraph, what a good outcome looks like and how it will be measured. Be specific about which number: platform-reported, blended, or a finance figure. The three diverge, and the divergence is where most end-of-pilot arguments live — the gap between platform ROAS and blended MER explains why.
Include a quality criterion as well as a volume one. "Forty leads at or below this cost, of which at least this share are reachable and in territory" is a target that cannot be met by buying cheap rubbish. For ecommerce, use contribution rather than revenue; the contribution margin calculator turns it into a single number both sides can agree on.
A pilot without a written definition of success becomes a negotiation about whose interpretation of the data is more generous.
Set the budget so the test can pass
This is the most common way pilots fail. A budget too small to gather conversions guarantees an inconclusive result, and then both parties waste ninety days discovering what a calculator could have told them in a minute. Work out the conversions the platform needs, divide by your conversion rate, multiply by cost per click, and you have the floor. The learning phase calculator does it directly.
If your available budget is below that floor, a pilot is not the right instrument. Either fix the conversion rate first or accept a longer, slower read.
What both sides owe each other
The agency owes you: full access handed back at the end, written documentation of what was changed, and a mid-point check-in rather than a surprise at day ninety. You owe them: the data they asked for, decisions inside a working day or two, and sign-off on creative that does not take three weeks. Agencies are slowed down by client response time more often than anything else, and a pilot magnifies it.
Agree the exit in advance
Two sentences. If the pilot does not continue, access is returned and accounts stay with you — assuming they were yours to begin with, which is worth checking. And if it does continue, the terms are already agreed, so you are not negotiating a retainer from a position where stopping would waste the three months you just spent.
What an agency pilot genuinely tells you
Less about the final numbers than you would expect, and a great deal about everything else: whether they found things you did not know about, whether the reporting contained decisions or dashboards, whether they told you something unwelcome, and whether the person who pitched was the person in the account. Those four signals predict the next twelve months better than the cost per lead at day ninety.
We work month to month with no lock-in, which makes every engagement function as a pilot by default — the pricing page and the process both start from that assumption. Before anything begins, a free audit usually shows whether a pilot is even the right next step.
Questions people ask before hiring
How long should an agency trial period be?
Ninety days is the shortest window that covers setup, the learning phase and one genuine read of performance. Thirty days tests onboarding, which is useful but is not the thing being bought.
Should a pilot cover all my channels?
No. One channel, one market or one clearly defined problem gives a cleaner read and leaves the rest of the account as a comparison. Broad pilots produce broad, unfalsifiable conclusions.
What should the success criteria be?
A volume target, a cost target and a quality target, each tied to a named metric and a stated data source. Without the quality criterion, the easiest way to pass is to buy cheaper and worse.
What if the pilot budget is too small?
Then the pilot cannot produce a conclusive answer, because the account will not gather enough conversions for bidding to stabilise. Work out the floor first and either meet it or choose a different approach.
Should I pay full rate during a pilot?
Usually yes. A discounted pilot tends to get discounted attention, and the fee is small relative to the media and to the cost of choosing the wrong agency for a year.