Ninety days is the trial period almost every agency engagement runs on. Both sides know it. Almost nobody writes down what passing looks like, which means the review at the end gets settled by whoever is more persuasive in the room.
That is an expensive way to spend a quarter. Here is the version worth agreeing to in advance, ideally before anybody signs anything.
You cannot judge results in ninety days, and you should stop trying
If your sales cycle runs nine weeks, a lead generated in week six of the engagement closes somewhere around week fifteen. The first cohort that could fairly be credited to new work has barely finished converting by the time the ninety-day review happens. In categories where consideration runs months rather than weeks, the honest answer at day ninety is that the revenue question is still open.
That creates a trap. If results are not available yet and nobody agreed on anything else in advance, the review gets decided on sentiment: how responsive they are, how good the deck looks, whether the dashboard points up and to the right. All three are entirely compatible with an engagement that is going nowhere.
So judge the work. In ninety days you can establish with certainty whether the people you hired understand your business, built something durable, and left you able to measure them for the rest of the year. All three are knowable by day ninety, and together they predict the next nine months considerably better than an early return figure does.
Days 1 to 21: they should tell you something you did not know
The first three weeks are for reading the business, not improving it. What should come out the other end is a written diagnostic: a contribution-margin model net of returns, shipping, payment fees and the real cost of fulfilment. Cohort behaviour by acquisition channel rather than a blended average that hides the difference between them. A tracking audit that states plainly which numbers in your current reporting are wrong, and roughly by how much.
The test of that document is simple and slightly uncomfortable. It should contain at least one finding that embarrasses somebody. A channel credited with revenue it did not cause. A product line that loses money once returns are counted. A measurement setup that has been double counting for two years. If the first deliverable of a new engagement is entirely complimentary about everything that came before it, either nobody looked hard or you are being managed.
Around day 30: one page, not forty slides
By the end of the first month there should be a written position on what is actually holding the business back. One page. The constraint, the two or three moves that relieve it, what each is worth if it works, and what they expect to be wrong about.
The page limit is not a stylistic preference. A forty-slide strategy deck is usually evidence that nobody has decided anything: it contains every observation, ranked by nothing, with the hard tradeoff left for you to make in the meeting. If an agency cannot state the constraint and the plan in a page, they do not yet understand the business well enough to be spending its money.
The clause worth insisting on is the last one. A plan that names its own likely failure modes in advance is a plan somebody intends to be held to. A plan with no stated risks gets retrofitted, after the fact, into whatever happened to occur.
Days 30 to 90: things should be live, in accounts you own
The back two thirds of the quarter is build. Whatever the thesis called for: account architecture, measurement infrastructure, lifecycle flows, creative pipeline, the revenue path on the site. Shipped and running, not presented.
- Built in your accounts, under your billing and your administrative ownership, rather than inside an agency-owned container handed back later as a favour.
- Documented as it is built, in writing, somewhere you can read without booking a call.
- Instrumented, so the thing that was built reports on itself instead of being described to you monthly.
- Reversible. Anything that cannot be rolled back should have been tested against a holdout before it went everywhere.
That last point is where ninety-day engagements most often go quietly wrong. A full account restructure in week five feels like decisive action and destroys your ability to attribute anything that happens afterwards. If everything changed at once, the honest reading of month three is that the new work cannot be separated from the seasonality, the pricing change and the competitor who went quiet.
The ownership test
One question cuts through most of the rest, and it is worth asking out loud in the first week rather than the last: if we parted company on day ninety-one, what would we lose?
- Ad accounts, under your billing, with your organisation as owner and the agency holding user access.
- Analytics and tag management inside your property, not theirs.
- The CRM, the automation platform and the list, with the data exportable by you.
- Creative source files, not only the exported assets.
- A written record of what was built, why, and what it replaced.
If the answer is that you would lose labour and momentum, that is a healthy engagement. If the answer is that you would lose the systems themselves, you did not buy a growth programme, you rented one, and every renewal conversation from here will be held with a quiet hand resting on the plug.
What should not happen
- A revenue promise on a ninety-day horizon in a category with a ninety-day sales cycle. The arithmetic does not permit it, so it is either ignorance or theatre, and neither is what you are paying for.
- Weekly reporting that grows in length while saying less. Volume is a common substitute for a result.
- A rebrand nobody asked for. It remains the most reliable way to spend a quarter looking busy.
- Platform-reported return presented as the headline outcome with no statement of what it excludes or double counts.
- Learning phase, algorithm change and seasonality offered as explanations more than once each. All three are real, and all three are also the most popular places to put a bad month.
Your half of it
Most failed first quarters we see were not execution failures. They were access failures, and the client owned them.
- Data access in week one, not week six. Every week spent waiting on CRM credentials is a week of the diagnostic being guessed at instead of calculated.
- One decision-maker with authority to approve a change of direction. Engagements with four stakeholders and no decider spend the quarter manufacturing consensus.
- Sales call recordings, or at minimum the objections your closers actually hear. Nothing improves conversion faster, and nothing is harder to get hold of.
- An honest number for margin. An agency optimising toward a margin figure you softened will optimise toward the wrong thing, confidently, for a year.
The day ninety scorecard
Five questions. Answer them from the documents rather than from the relationship.
- 01Did the diagnostic tell us something we did not know, including something we would rather not have heard?
- 02Is there a one-page thesis, written before the work started, that we would still defend today?
- 03Is what was built live, documented, and sitting inside accounts we own?
- 04Do we have a baseline recorded before anything changed, and a stated expectation of what should be true in six months?
- 05If we stopped tomorrow, would we keep the systems?
Five yes answers and the engagement has started properly, whatever the revenue line happens to be doing in week twelve. Three or fewer and you are ninety days into paying for a relationship rather than a result. The useful thing to do with that information is say so now, while there are still nine months left to fix it, rather than at the annual review.