A ninety day growth roadmap is a dated sequence of fixes ordered by what depends on what. Month one rebuilds measurement so results can be judged. Month two ships the single largest structural fix.

Month three reallocates budget and proves the fix held. Each item carries one owner and one date.

Picture the handover meeting. Fourteen findings, each one priced and placed in value order.

Then somebody asks what happens Monday, and the room goes quiet. The document that just answered what matters most has nothing to say about what goes first.

I have written about what the diagnosis buys you and how the teardown is run. The audit month is not one of these ninety days. The clock starts the morning the findings land.

A ranked list stops one step short

Ranking answers one question at a time. Each finding gets weighed on its own, value against effort.

Sequencing asks about the whole set at once. The top item on almost every list I have delivered cannot start in week one, and the reason sits further down the same page.

A ranked list tells you what matters most. A roadmap tells you what can actually go first, and those are rarely the same item.

The usual shape: finding one says paid is buying the wrong customer, and the fix is to point campaigns at a later event. Two weeks of work at most.

It is also unstartable: that event does not exist in any form a platform can receive. The most valuable item on the page is blocked by the ninth.

Two kinds of dependency, and only one is obvious

The first is a build edge. Item B cannot be constructed until item A exists. Teams sequence around that without being told, because the block announces itself the moment somebody starts.

The second is a proof edge. Item C can be built this afternoon by anyone with the access.

Ship it before item A lands and you will never know whether it worked. The instrument that would have told you was broken while the result accrued.

Two kinds of dependency Every finding carries both. Only one of them stops you from starting. BUILD EDGE PROOF EDGE Server-side purchase event must exist first Bid toward purchase value Nothing can be built until the event exists. The block is obvious to everyone in the room. Billing reconciliation must be true first Rebuilt pricing page Nothing stops the page shipping this week. The work happens. The learning does not.
Teams sequence around the left panel because the block is visible. The right panel is where quarters get lost, because the work looks free to start.

So the sequencing pass asks three things of every finding. What has to exist before this can be built. What has to be true before its result can be read.

Then the one that moves the calendar: how many items are queued behind this one. An item with three waiting on it earns an early slot whatever its value. That is the unblocker premium.

Why does measurement have to go first?

Every remaining fix produces a number as its only output, and that number decides whether you keep it, extend it, or reverse it. If the instrument disagrees with billing, each output is a guess with a decimal point attached.

Measurement goes first every time. Every win after it stays unprovable until it lands, however small its own number looks.

Going first has a specific meaning. Four or five events at most, each with a definition finance and marketing have both signed, firing server-side where an ad blocker cannot reach them.

Reconciliation runs weekly until the gap sits inside a tolerance the founder agreed to before seeing the first number. A tolerance negotiated afterward is one you keep renegotiating.

Then freeze a baseline: one dated page holding every number as it stood the day before anything changed, read-only for the quarter.

Without it, day 90 becomes an argument about where you started, and whoever remembers the most flattering January wins it.

What lands in each of the three months

The lanes below are the common arrangement. Their widths change with the business; their order rarely does.

Ninety days, drawn as a graph Solid arrow: cannot be built yet. Dashed arrow: cannot be read yet. DAYS 1 TO 30 DAYS 31 TO 60 DAYS 61 TO 90 Cut the obvious waste days 1 to 10, subtractions only Measurement rebuild Weekly review rhythm One structural fix READ WINDOW Channel reallocation READ must exist first result unreadable until this lands read window: the span where a result becomes legible The last two weeks carry no new work. They exist to hold the read.
Five lanes, two edges. Reallocation is buildable in week one and unreadable until the rebuild lands, which is the only reason it sits in month three.

Month one: build the instrument, cut the obvious waste

Two things run in parallel: the measurement build, and the stop list. The stop list needs no instrument to justify itself, so it outranks even measurement.

Every audit turns up spend that is wrong under any reading of the data: a campaign chasing an event that stopped tracking with revenue a year ago, a tool renewing at full price with three logins. Month one's only spend changes are subtractions.

Month two: ship the one structural fix

Structural means a change to how the business converts or monetizes rather than how it advertises: the funnel step where most people quit, the plan mix that caps what a customer can be worth, or the rule that decides who gets in.

Each lifts every channel at once, because every channel runs through it. Exactly one of them: what runs out first is legibility.

Month two is also when the operating rhythm starts, a weekly review with kill-and-scale thresholds set before the first numbers arrive.

Month three: reallocate, then check the fix held

Two things become possible now. The first is an honest channel read, so budget moves on evidence, usually off a channel protected by a flattering platform number.

The second is the hold check. Four weeks of post-change data is where novelty wears off, and you find out whether you fixed the thing or moved it somewhere nobody was looking. Whatever did not fit the quarter goes onto a carry-forward list with its edges attached.

The ship date is set by how long the result takes to read

Every fix has a lag between shipping and being legible, and that lag varies by more than an order of magnitude. Typical spans at this size:

  • A funnel step reads in one to two weeks, once enough people have walked the changed step.
  • A tracking rebuild reads the day reconciliation matches, though channel comparison wants 30 clean days on top.
  • A pricing or plan mix change needs a billing cycle plus the first renewals, so 60 days or more.
  • A lifecycle program reads in stages, its first flow in about three weeks and its retention effect later.

Work backwards from day 90. Anything carrying a sixty day read has to ship inside the first thirty days, so a pricing item may outrank a funnel item worth more.

The second constraint is separation. Two fixes that move the same number cannot share a read window, because the result arrives as one blended figure.

Ship a checkout change and a price change in one fortnight and the argument about which moved conversion outlasts the quarter. Stagger anything touching one metric by a full read window.

Skip both constraints and day 90 looks the same every time: five workstreams open, each two thirds finished, and the two that shipped landed days apart on the same metric.

The audit was right. The ranking was right. The quarter still bought activity where it needed evidence.

Ninety days is long enough to fix the thing and see whether the fix held. Thirty days only ever proves that something changed.

What do I have to decide myself?

Most of the ninety days runs without the founder. Five decisions do not, and four land in the first fortnight.

  • The steering number, week one. Every event definition follows from it. Deciding in week six means rebuilding in week seven.
  • Whether price and packaging are open, week one. Many of the highest-value findings live there. Say so early if they are closed.
  • The stop list, week two. Ending a retainer, switching off a colleague's campaign, cancelling a championed tool: relationship decisions with money attached.
  • One owner per lane, and what that owner drops, week two. Naming the owner is the easy half. Naming what comes off their plate determines whether anything ships.
  • The go or hold at day 45. The second structural item either has support to start or waits for next quarter. That call on schedule stops one plan becoming five.

What does done look like at day 90?

Done is a state somebody can check, worth writing down at the start so nobody redefines it later.

  • Platform-reported and banked revenue reconcile inside the week-one tolerance, and have held for a month.
  • Finance and marketing compute the steering number the same way, from the same source.
  • The structural fix has been live four weeks or more, with its result on the page rather than in dispute.
  • Every channel has a read against real revenue, so a budget move can be defended with a number.
  • The weekly review has run twice without the person who built it in the room.
  • One page lists what was switched off and what that saved, beside the frozen baseline.

Notice what is missing. No revenue target, no promised acquisition cost. Ninety days is enough to make the machine legible and land one real fix; the compounding arrives afterward.

How does the roadmap change if the audit found something else?

Three verdicts cover most of what comes back, each bending the calendar differently.

Broken measurement. The instrument absorbs the opening month, the structural change waits for the second, and the first believable channel comparison arrives in the third. Say in week one that you are trading visible progress for the ability to judge anything.

Broken funnel, measurement broadly sound. Only the specific steps need instrumenting, so the structural fix starts in week three and two read cycles fit inside the window.

Broken unit economics. The pricing change gets pulled to the front of month one, ahead of work worth more on paper, because a sixty day read cannot begin in month two. Spend freezes while it runs.

The verdict sets the shape Same ninety days. The boundaries move to fit the slowest read on the list. DAY 1 30 60 90 Measurement broken one fix, one read Measurement Structural fix Channel read Funnel broken, data sound two fixes, two reads Instrument First fix Read Second fix Unit economics broken one slow read, spend held Price change One billing cycle, then the first renewals Read no spend reallocation while this read runs Row two is the shape you want. Row three pulls a middling item to the front because its answer takes sixty days to arrive.
Row three is the one founders argue with, because the item moving to the front rarely matches what they would have picked from the ranked list.

Build the graph yourself

None of this needs an outside operator. Put every finding on its own line with what must land ahead of it, what must be believable before its outcome counts, and how long that outcome takes to surface.

Promote whatever has a queue behind it, start the slowest-surfacing items soonest, and keep the closing fortnight clear. One owner and one date per line.

What comes out is a plan where week one is obvious and you can say afterward which change did what. These ninety days are the fix phase inside the wider growth operating system.

For companies past product-market fit, usually $15M in revenue and up, I run the diagnosis at a fixed $12,500, roadmap included.

Where that turns into ongoing operating help, retainers open at $20,000 a month and the arc is laid out here.

If you are holding the findings and have no answer for week one, sequencing them is the job. Let's talk.