What Should a Marketing Engineering Agency Build in the First 90 Days?

In the first 90 days, a marketing engineering agency should build five things, in this order: a measured baseline for the one business number you hired them to move, a written "brain" every agent reads before it writes a word, one live system running in your own accounts on your own data by about week six, the guardrails that keep a human in front of anything a buyer sees, and a handoff kit you could run without them. By day 90 you should own all of it, see the first system's output every week, and know whether the number is moving or exactly why it isn't.

What you shouldn't expect in 90 days is a full crew of agents, a finished attribution model, or a guaranteed revenue figure with no baseline under it. A firm that promises those is selling the pitch deck. This guide walks the 90 days week by week, gives you the checkpoint to hold at day 14, 30, 45, 60 and 90, shows how much of your team's time it takes, and ends with the questions that expose a weak plan before you sign.

Who this is for

You run or lead marketing at a B2B company somewhere between $5 million and $75 million in revenue. You've hired a marketing engineering agency, or you're about to, and the proposal in front of you has a slide called "Our 90-Day Plan" that could describe any client they've ever had. You want to know what a good first quarter actually produces, so you can hold the firm to it, or spot the gap before the contract is signed.

If you're still deciding whether you need an agency at all, start with what a marketing engineering agency actually does. If you're comparing firms, the scorecard and the RFP template come first. This article is the one you keep open once the work starts.

The short answer

Here's the whole quarter on one page. The weeks flex by a week or two depending on how clean your data is and how fast your team approves things. The order shouldn't flex at all.

Two systems live by day 90 is a strong quarter. One is acceptable if it's running well and the brain behind it is solid. Zero live systems and a strategy deck is a quarter you paid for twice: once now, and once when the next firm starts over.

Why the order matters more than the list

Most 90-day plans list the right ingredients and cook them in the wrong order. They launch agents in week two, because agents are what you bought, and write the strategy in week eight, after the agents have already spent a month producing copy nobody can defend. Michael Watkins made the case in "The First 90 Days" that a new leader earns the right to make big changes by securing early wins on a foundation people trust. The same logic holds for a new system. The early win only counts if what's under it is true.

Marketing engineering has a specific version of this problem. An agent on thin context produces fluent, generic output at volume. We call that the Context Vacuum: the empty space where your story should be, which the model fills with the average of everything it has read about companies like yours. Greg Isenberg, who has done more than anyone to popularize the marketing engineer role, gives the same advice first: start with the brain. Ten agents reading a tagline will bury you in plausible slop by week four.

B2B makes the order even less forgiving. A consumer brand can launch thirty ads a day and let the numbers find the message. If you sell to 400 hospital systems or 200 regional banks, you'll never get the volume to learn that way, and every buyer you burn with the wrong message may not come back. The first 90 days in B2B have to start from the right story, which is why the brain sits in weeks three and four and the first live system waits until week five.

Weeks 1-2: access, discovery and the baseline

The first two weeks are unglamorous and they decide the quarter. The agency gets access to the systems it will read from and write to, interviews the people who know the buyer, and measures the one number before anything changes it.

What gets done

• Access with an owner. Admin or API access to your CRM, call recorder, website, analytics and any AI tools you already pay for, granted to named people under your accounts. Every credential lands in a shared inventory on day one, not in someone's browser.
• The one number, measured. The business metric from your contract (qualified pipeline per quarter, win rate on deals over $50K, inbound reply time, or how often AI engines name you on your buyers' top questions) measured with a method you both sign. If it's AI visibility, the question set is agreed in writing and run on at least three engines.
• Discovery that goes past the marketing team. Founder sessions, five or six customer interviews, a listen to a dozen recorded sales calls, and a read of the last twenty lost deals. The best language for your buyer's problem usually lives in your recorded calls, and rarely in your marketing.
• A systems map. What talks to what, where data goes stale, and which integrations already fail quietly.

What you should have on day 14

• A one-page baseline: the number, today's value, how it was measured, and the date it will be measured again.
• A list of the systems the agency can access, who granted it, and how it gets revoked.
• A short list of candidate first systems, ranked, with the reason for the ranking.
• A list of what they need from you in weeks three and four, in hours, by person.

The day-14 red flag is a missing baseline. A firm that can't, or won't, write down where you're starting is a firm that will describe success in whatever terms are flattering in month four. The red flags guide treats "no measurement" as the first flag of the build stage for exactly this reason.

Weeks 3-4: the brain

The brain is the written source of truth every agent reads before it produces anything. It's a small set of plain documents, kept in your accounts, that answers three questions an AI model can't answer for you: who buys, how you sound, and what your company believes that a competitor wouldn't say.

What the brain contains

One rule keeps the brain healthy for years instead of weeks: agents can suggest additions, and only a named human turns a suggestion into company policy. Without that rule the brain drifts toward whatever the agents found easiest to say.

At PitchKitchen we document the brain as the Magnetic Messaging Framework, thirty-one sections covering who the company is, who it's for, what it stands for and the story it tells. You don't need our format. You need some format that's written, approved and readable by a machine. Our marketing engineering page shows the three core files and how they map to the framework, and how to make brand guidelines machine-readable covers the mechanics.

What you should have on day 30

• The brain files, approved in writing by the person who owns the company's story (usually the CEO).
• A short test: the agency runs five real tasks (a cold email, a page section, a call summary) with and without the brain, and shows you both. If you can't tell the difference, the brain isn't done.
• The first system chosen, with its inputs, outputs and the person who approves its output.

If your positioning is genuinely unsettled, weeks three and four will expose it. That's useful, and it's also the most common reason a 90-day plan slips. A firm that can do the strategy work will say so and adjust the timeline. A firm that can only build will quietly encode your current confusion into every agent. How to tell whether an agency can actually build covers the opposite failure: firms that can talk strategy and can't ship.

Weeks 5-6: the first live system

The first system should be the one with the highest value from data you already have and the smallest blast radius if it's wrong. In B2B that almost always means an agent that reads before it writes to anyone outside the building.

Three good first systems for a B2B company

Notice what isn't on the list: outbound at volume. Automated outbound is the most common thing agencies want to build first, because it's visible and it produces activity numbers fast. It's also the system most likely to burn buyers you can't replace if the brain is still thin. Build it after the language system has told you what your buyers actually say.

What you should have on day 45

• One system running in your accounts, on real data, on a schedule, without the agency pressing a button.
• Its first weekly output in front of the person who owns it.
• Every buyer-facing output landing as a draft, with the approval step built into the tool itself. "Always ask me first" written in a prompt is a polite suggestion, so the approval has to be a step the agent physically can't skip.
• A log showing what the system did, what it read, and where it failed.

Weeks 7-8: the second system, and hardening the first

By week seven the first system will have broken at least once. An API limit, a renamed CRM field, a call recorder that changed its export format. That's normal. What matters is whether anyone found out. The agency's job in weeks seven and eight is to make every failure loud: an alert to a named person, an automatic retry where it's safe, and a documented manual fallback where it isn't.

We learned this the expensive way on our own system. PitchKitchen runs on nine agents in Claude Code, all reading our own framework, and some of them break. One of them, Bob, exists only to watch the others and flag anything that stops. A crew without that kind of watcher looks healthy right up until you discover it's been silent for three weeks.

The second system goes live in the same window, usually the next one down the ranked list from day 14. It should reuse the brain and at least one integration the first system already proved, which is why it takes half the time.

What you should have on day 60

• Two systems running, each with a named owner on your side.
• An alert path for each integration, tested by actually breaking it once.
• A short change log: what was adjusted in the brain or the prompts since day 30, and why.

Weeks 9-10: the measurement loop

Measurement in a marketing engineering engagement has one job: tell you what to do next week. A dashboard with forty charts doesn't do that. A short weekly memo does, if every line ends in one of four verbs: keep, kill, change or test.

• Keep: the customer language system surfaced three phrases that now open every discovery call.
• Kill: the lost-deal notes for deals under $10K, because reps ignore them.
• Change: the AI answer tracker's question set, because two of the ten questions turned out to be ones buyers never ask.
• Test: a homepage headline built from the most frequent buyer phrase, measured against the baseline.

By day 75 you should see the first trend line against the day-14 baseline. It may be flat. Ninety days is short for any number that depends on buyers, and AI engines in particular update on their own schedule. A flat line with a clear explanation and a test underway is a healthy result. A flat line explained by "these things take time" with nothing being tested is not.

For AI search specifically, the guide to marketing engineering agencies for AI search and GEO lists the six artifacts a legitimate AI visibility engagement should hand you by day 90. Use it alongside this plan if AI visibility is your one number.

Weeks 11-12: training and handoff

The last two weeks turn a system the agency runs into a system your company owns. Even if you plan to keep the agency running it, you should be able to fire them on day 91 without losing anything.

The handoff kit

• A runbook a new hire could use to operate each system: what it does, when it runs, what to do when it fails.
• An ownership inventory: every workflow, prompt, knowledge file, dashboard and agent configuration, and the account it lives in. All of them should be yours.
• A credential list with who holds each one and how it's revoked.
• Training for the people who approve outputs and the person who maintains the brain, measured by use, not attendance.
• A next-quarter proposal: what to keep, what to add, what to stop.

The handoff test is simple. Ask the person on your team who'll own the systems to walk you through one failure and one brain update, without the agency in the room. If they can, the quarter worked.

The day-90 review

Put a 60-minute review on the calendar on day one, so neither side can let it drift. Bring the day-14 baseline and ask six questions:

1. 1What's the number today, measured the same way as on day 14?
2. 2Which systems are running, and when did each last fail?
3. 3What did the brain change in our marketing that we can point to?
4. 4What did we kill, and what did that free up?
5. 5Could we run everything without you tomorrow? Show us.
6. 6What would you do differently if you started this quarter again?

The last question is the most revealing. A good firm has an answer ready. A weak one will tell you everything went to plan, which is never true in a first quarter.

What a 90-day plan shouldn't include

Overreach is the most common way a first quarter fails. Push back on any plan that puts these inside the first 90 days:

• More than three agents live. Each one needs a brain, an owner, an alert path and a review habit. Five systems at 60 percent are worth less than two at 100 percent.
• A CRM migration. It's a separate project with its own risk. Build on the CRM you have and fix the fields you need.
• A rebrand. If the visual identity needs work, schedule it after the story is written, so the design carries the right message.
• Guaranteed revenue with no baseline. A guarantee is only as good as its baseline, instrument and remedy. Without all three it's marketing.
• "Phase two discovery" at day 60. Discovery belongs in weeks one and two. A second discovery phase usually means the first one didn't happen.

How the plan changes by company size

The order stays the same at every size. What changes is how many people the brain has to align and how much of the site and stack the first quarter touches.

Budget follows the same bands. What a marketing engineering agency costs covers pricing models and ranges, and what a $25K, $50K or $100K engagement should deliver breaks down the scope each budget should buy.

How much of your team's time the 90 days take

Agencies underquote this more than anything else, and it's the most common reason a plan slips. Plan for roughly the following. These are planning ranges, and your agency should give you its own numbers, by role and by week, in the proposal.

If nobody on your side can own the systems after day 90, say so before you sign. The right plan then includes either training someone or a run phase with the agency, priced separately.

How we run the first 90 days at PitchKitchen

We're one of the firms you might be weighing, so here's our own plan, stated plainly, for you to hold against everyone else's. The 90-Day AI Website Rebuild runs in three blocks. Weeks one to three pull the truth: founder sessions, customer interviews, the competitor AI keeps naming, your buyers' top ten questions agreed in writing and the baseline measured. Weeks four to eight write the story as the Magnetic Messaging Framework and rebuild the homepage on it, live inside your current site. Weeks nine to twelve grow the core pages, send the story out across the web and track the AI score weekly against the baseline.

Three agents start with every client who continues into our run phase: Echo (customer language), Phoenix (lost-deal reopener) and Iris (AI answer tracker), the same three first systems recommended above. We'd be a poor fit if you want high-volume outbound plumbing built in Clay. A GTM engineering shop will serve you better there. We're built for B2B companies from $5M to $75M whose real problem is that the story isn't clear yet, and who want the systems to run on it once it is.

Questions to ask about any agency's 90-day plan

• What exactly is live in our accounts on day 45, and who approves its output?
• How will you measure our baseline in the first two weeks, and with what instrument?
• Where does the brain live, who approves changes to it, and what happens to it if we part ways?
• What will break first, and how will we find out?
• How many hours a week do you need from each person on our team, by phase?
• What's in your day-90 handoff kit? Can we see one from a past client, redacted?

For the full interview set, see the ten questions to ask before hiring a marketing engineering agency. If you're writing an RFP, section 10 of the template asks for this plan in a shape you can compare across firms.

The bottom line

A good first 90 days is boring to describe and obvious to inspect. A number written down in week two. A brain your CEO signed in week four. One system on real data by week six, a second by week eight, and a weekly memo that tells you what to do next. Everything in your accounts, and a team that can run it on day 91. If the plan in front of you can't be checked against those milestones, ask for one that can.