The First Hundred Days of a Measurement Migration

I have spent seven issues making an argument. This one is a manual.

The question I get asked most often, by some distance, is not why should we do this. Most of the CMOs and CFOs I speak with resolved that a while ago. The question is what actually happens if we start. What lands on whose desk in week two. What goes wrong. What I tell my board when the first honest numbers arrive and they are worse than the numbers they replace.

So this issue is the operational version. It is drawn from watching a reasonable number of these migrations, including several that stalled, and it is deliberately specific about the parts that hurt.

One framing note before the timeline. A hundred days is enough to complete the first full loop connected data, a real experiment, a calibrated model, and one budget decision made on the result. It is not enough to finish the migration. Anyone who tells you the whole thing takes a quarter is selling something. What a hundred days buys you is the first proof that the loop closes, which is the thing that makes the next nine months politically survivable.

Three jobs in the first month, and they run in parallel rather than in sequence.

Connecting the data is the easy part, and it is worth saying so. Ad platforms, commerce, CRM, and the offline sources connect in days rather than weeks in most companies. Your marketing operations lead owns this. It is genuinely the least interesting part of the migration and it is where most vendor demos spend most of their time, which tells you something about vendor demos.

What takes longer than people expect is not connection but definition. Two conversion events with the same name and different logic. A revenue figure that includes returns in one system and excludes them in another. Time zones. Currency. Which of the four things called "signup" is the one that matters. Every migration I have watched has lost a week here, and the teams that budgeted for it were noticeably calmer than the ones that did not.

Documenting the baseline is the job people skip and later wish they had not. Before anything changes, write down what your current system says. Reported return by channel, the budget those numbers justify, and the specific decisions made on them in the last two quarters. Freeze it. Date it.

You are going to need this document at day ninety, because the entire value of the migration is expressed as a difference between what you used to believe and what turns out to be true. Teams that never wrote down what they used to believe find themselves unable to demonstrate what changed, which makes the whole exercise feel like an expensive lateral move.

Choosing the first experiment is the most consequential decision of the hundred days. Your CMO owns this one and should not delegate it.

The instinct is to test the biggest, most contested channel, because that is where the money is and where the argument is loudest. This is almost always wrong, and I will come back to why in the section on mistakes. What you want first is a channel where you can get a clean read — sufficient spend to move a measurable outcome, enough geographic spread to construct a credible comparison, and low enough political temperature that the result will be accepted rather than litigated.

John Kotter's Leading Change makes this point better than I can in a marketing-specific frame. His argument that transformations fail when leaders neglect to generate visible short-term wins is, in my experience, the single most predictive variable in whether a measurement migration survives its first year. Your first experiment is that short-term win. Choose it like one.

The geographic experiment runs. Depending on design and category, you are looking at several weeks in market before there is anything to read, and your first calibrated attribution view lands in roughly the same window — a week or two after connection, once conversions have been reconciled.

The marketing mix model takes longer. Four to six weeks is the honest range for a first properly specified model in most businesses, and anyone promising a trustworthy causal read in days is describing a dashboard rather than a model.

This is the period when the migration stops being a project and starts being a feeling.

Somewhere around week five, someone on the team sees the first honest read on a channel they have personally defended for two years, and it is meaningfully lower than the number in the old dashboard. In my experience this is not received as information. It is received as an accusation.

I want to be direct about this because I think it is the most underestimated risk in the entire process. The technical work is not the hard part. The hard part is that a measurement migration is an exercise in telling a group of competent professionals that some of what they believed was wrong. Nobody involved did anything dishonest. The instrument was bad. But the person who has been optimising that channel for two years does not experience the distinction as clearly as the person reading the report.

The CMOs who handle this well do one specific thing, and they do it before the result arrives rather than after. They tell the team, in week one, that the first results will make some current decisions look wrong, that this is the expected outcome rather than a failure, and that nobody is going to be judged on decisions made with the old instrument. Said in advance, it is leadership. Said after the result lands, it is damage control, and everyone can tell the difference.

This is the part I get asked about most, so I want to give it its own section.

At some point in the first quarter your board or your CFO will see a number that is worse than the number they saw last quarter. Return on your largest channel will be lower. The confidence interval around it will be wider than the false precision it replaced. Someone will ask, reasonably, whether the new system is working.

Here is the sentence I would build the entire conversation around.

The numbers did not get worse. The reporting got honest.

Then show three things, in this order. The old number. The new number. The specific experiment that produced the difference. Not the methodology — the experiment. Boards do not want a lecture on causal inference; they want to know that a real thing was done in the real world and this is what it showed.

The strategic move, though, is not the day-thirty conversation. It is what you said at day zero.

If you told your board in advance that reported performance would decline as measurement improved, and explained why, then day thirty is a confirmation of your judgment. If you did not, day thirty is a surprise that arrives sounding exactly like an excuse — and it does not matter that it is true, because the shape of the conversation is already set.

There is a genuine psychological asymmetry underneath this that is worth naming. Kahneman and Tversky's work on loss aversion is the usual reference — losses are felt roughly twice as intensely as equivalent gains. A board that watches a reported return fall from three-point-eight to two-point-one experiences a loss, even though nothing about the business changed and the second number is the one that was always true. Rory Sutherland's version of the point, which I find more useful in practice, is that people judge a change by the transition rather than the destination. The destination here is excellent. The transition looks like a decline. Manage the transition explicitly or it will manage you.

One more thing about this conversation. Bring your CFO into it before you need them, not when you need them. A CFO who has been part of the migration since week one arrives at the day-thirty board meeting as a co-author defending a shared decision. A CFO who first hears about it when the numbers drop arrives as a second interrogator. Same person. Entirely different meeting.

By roughly day sixty you have an experimental result, a calibrated attribution view, and a marketing mix model either delivered or close.

What happens next is the part that makes the migration real, and it is worth being precise about the mechanics because they are commonly described wrongly.

The experimental result does not get applied as a multiplier on top of the model afterwards. It informs the model before the model runs — it enters as prior knowledge about how a channel behaves, so that the model is built already knowing what the experiment demonstrated. The distinction sounds academic and is not. A number adjusted after the fact is a fudge factor and everyone in the room can smell it. A model built on experimental evidence is a different object, and it survives interrogation by a sceptical CFO in a way the first thing does not.

It is also worth saying what this is not. The methods are not three independent opinions being averaged until they agree. They do different jobs — the model handles strategic allocation across the whole mix, the experiment supplies ground truth on specific questions, the calibrated attribution view gives granularity for weekly decisions. Expecting them to produce identical numbers is a misunderstanding of what each one is for. Rajeev has written about this distinction in The Incrementalist with considerably more rigour than I can bring to it, and it is the piece I would put in front of a technical stakeholder who wants to interrogate the approach.

Then you make one decision. One reallocation, executed, with a stated expectation of what it should produce and a date on which you will check.

Not five decisions. One.

The single reallocation is the artefact that ends the first hundred days, and its purpose is not primarily financial. Its purpose is to demonstrate that the loop closes — that evidence produced a decision, the decision was executed, and the outcome can be checked against what was predicted. Once a company has done that once, it will do it a hundred times. Until it has done it once, every conversation about measurement remains theoretical, and theoretical initiatives are the ones that get cut when the quarter gets difficult.

Four things, reliably.

Data definitions, as described above. Budget a week. You will use it.

The agency relationship. This is the one nobody plans for. A meaningful share of agency compensation in this industry is still tied to spend under management. A measurement system whose honest answer is frequently spend less here is, structurally, a threat to your agency's revenue. I am not suggesting agencies argue in bad faith — the good ones do not. But you should expect methodological objections to arrive with unusual energy, and you should have decided in advance whether your agency is a participant in the migration or a party to be managed through it. The best outcome I have seen is bringing the agency in during week one and being explicit that the compensation model will be revisited rather than allowing it to become the unspoken subtext of every methodology conversation.

Reporting continuity. For a period you have two number systems and one board deck. This is genuinely awkward and there is no elegant solution. What works is a sunset date, set at day zero, after which the old numbers do not appear in decision-making forums. What does not work is running both indefinitely and letting each meeting decide which to believe.

Somebody's favourite channel. Usually branded search, retargeting, or an affiliate programme. There is no process fix for this. There is only the leadership move described above, made early.

Month four is when migrations die. The novelty is gone, the first results have landed, the hard reallocation is now due, and the quarter has its own problems. Three patterns account for most of the failures I have watched.

One. Parallel systems with no sunset date. Running old and new alongside each other feels like prudence and functions as a stalemate. Two number systems means every budget meeting begins by relitigating which system to trust, and the old one tends to win — it is familiar, it is more flattering, and it is what the incumbent process was built around. Set the sunset date at day zero, put it in writing, and hold it.

Two. Starting with the hardest channel. Teams choose their largest and most contested channel first because it is where the money is. It is also the highest-noise, most politically loaded, most likely to produce an ambiguous first result. An ambiguous first result in month two is how a migration acquires the reputation of being inconclusive, and that reputation is very difficult to shed. Start where the read will be clean. Earn the right to test the contested channel by having already been right about an easier one.

Three. Treating it as a tooling project rather than a change in who decides what. This is the deepest of the three and the most common. If the migration is owned by analytics, produces better reports, and never touches the process by which budget is actually allocated, then nothing has changed. The company has bought a more accurate description of decisions it is going to make the same way regardless. Measurement migrations that succeed are accompanied by an explicit change in decision rights — what the model is allowed to trigger automatically, what requires human approval, who signs off on a reallocation above a threshold. If nobody's authority changed, the migration did not happen.

One. Whether a hundred-day migration playbook becomes a standard vendor deliverable. Most vendors in this category sell a platform and leave the operating-model change to the customer, which is roughly like selling someone a piano and wishing them luck. The vendors who ship a genuine change-management programme alongside the software will win more of the second year than the first. We are not yet as good at this as we should be.

Two. Whether the day-zero board briefing becomes standard practice. The single highest-leverage intervention in this entire document is a CMO telling the board, before starting, that reported numbers will decline as measurement improves. It costs nothing. It converts the most dangerous moment of the migration into evidence of foresight. I would like to see this become as routine as a pre-mortem.

Three. Whether agency compensation models shift in response. If measurement gets honest and agency compensation stays tied to spend under management, the misalignment becomes structurally untenable rather than merely awkward. I expect the first serious moves toward outcome-linked agency compensation within eighteen months, and I expect them to come from agencies rather than clients — because the agency that proposes it first gets to define the terms.

Two things.

The first is that a hundred days is a clean narrative device and real migrations are messier than any timeline suggests. Companies with heavy offline retail exposure, long purchase cycles, or fragmented regional data take longer, sometimes considerably. A business-to-business company with a nine-month sales cycle cannot close the loop in a hundred days at all, because the outcome being measured has not happened yet. If you are in that category, the sequence still holds but the clock does not, and any vendor — us included — who implies otherwise is compressing your reality to fit their case study.

The second is that I sell the thing I have just described the implementation of, and this issue is therefore also, structurally, a document that makes buying easier. I have tried to make it genuinely useful independent of who you buy from, and I think the day-zero board briefing, the sunset date, and the decision-rights point are all true regardless of vendor. But the reader should apply the usual discount, and I would rather name that plainly than pretend a CEO writing an implementation guide for his own category is a neutral party.

If you have run one of these migrations — successfully or otherwise — I would genuinely like to hear what month four looked like. My sample is large enough to see patterns and not large enough to be confident about them, and the failures are considerably more instructive than the successes.

Thanks for reading the eighth one.

Tobin Co-Founder & CEO, Lifesight August 2026

Coming in Issue #9 (in two weeks) What Causal Measurement Actually Says About Brand Spend

Nearly every marketer I meet assumes that rigorous measurement is an argument for cutting brand investment — that the money will flow to whatever can be proven fastest. In our customer base the finding runs consistently in the other direction, and I want to show why, what the honest limits of the evidence are, and what a CMO should do with a number that says spend more on the thing that is hardest to justify.

Forward this issue to whoever would run the first hundred days at your company.