Build an AI-Native Business in 90 Days, Without a Budget

When I wrote about the difference between using AI and being AI-native, the most frequent response from executives was a polite objection: the distinction is real, but redesigning how work happens is a luxury reserved for companies with transformation budgets.

I spend my working life inside those budgets, and most of what they buy is the expensive undoing of habits that formed when the company was small. A founder has nothing to undo yet, which is why the quarter ahead is worth more than the three-year program a large enterprise will fund.

Here is the plan I would run if I were building a company today with no budget beyond my own calendar. In 90 days, with no new software and no outside help, you can do the three things that separate an AI-native business from one that merely uses AI: Write down how the work actually happens, take ownership of your data, and redesign who decides and what gets measured.

Why the big companies are paying for your quarter

In June, Genpact and HFS Research published a survey of more than 2,000 senior executives, and 85% of them said that accumulated enterprise debt — meaning undocumented processes, unusable data, aging systems and unprepared people — is limiting the value they get from AI. The researchers put the trapped value across the Global 2000 at roughly $18 trillion, found that only 33% of enterprise data is AI-ready and estimated that inefficient or manual processes consume about 40% of a typical employee’s week. More than half of those enterprises have no funded initiative to fix any of it.

Every item on that list is a habit before it is a legacy. Nobody decides to have undocumented processes; they go unwritten while the company grows, and the McKinsey Global Survey published this year found that nearly three-quarters of AI high performers have fundamentally redesigned their workflows, against roughly one-quarter of the rest, while the share of smaller organizations scaling AI agents sat flat at 22%. The gap between the two groups is not spend, and it is not model access. It is redesign, and redesign is cheapest before the habits set.

The first 30 days: Write down how the work actually happens

Choose one process that touches revenue directly, such as quote to cash, customer onboarding or returns, and resist the urge to choose three at once. Sit with the people who run it and have them narrate the real path, including the spreadsheet nobody admits to, the approval stuck in one inbox and the exception handled from memory, because the documented process and the one that actually runs are rarely the same document.

Then mark every step as either production or judgment. Production is the drafting, gathering, comparing and summarizing that a model can increasingly perform; judgment is the deciding, prioritizing and taking of responsibility that a human must own. By day 30, you should hold a one-page map a new hire could follow on their first morning, and if a new hire can follow it, so can a model. A process nobody can describe is a process no model can run and no auditor can defend.

Days 31 to 60: Take ownership of your data before it owns you

With the map in hand, trace where the information for each step lives, and expect to find it scattered across three tools, several inboxes and one personal drive. For every field that matters — meaning customer, product, price, status and date — name a single owner and agree on a single definition, because enterprises I work with across the Americas, Europe, Australasia and Japan are spending fortunes to rediscover what counts as an active customer, a question you can settle in a single afternoon.

Then check the exits. Open every tool you pay for and confirm that your data can leave it, in full, in a usable format, on your schedule — and if it cannot, that is the first renewal to renegotiate. Data you cannot export is data you are renting, and a company that rents its own context is not AI-native, however many models it subscribes to.

Days 61 to 90: Redesign who decides and what gets measured

Return to the judgment steps on your map and give each one a name, a definition of what a good outcome looks like and a clear rule for when a human reviews before anything reaches a customer. Then look hard at the approvals that remain, because most approval chains are information poverty in disguise, and once last month’s data is visible to everyone, many of them survive only out of habit.

Finally, retire the metrics that measure the tool and install the ones that measure the business. Baseline the cycle time, cost to serve and customer turnaround of your chosen process on day 61, measure again on day 90, and let the difference, rather than the number of licenses, tell you whether anything changed. In my experience, the first fully redesigned process returns more than the first 10 pilots combined, because a pilot proves that a model works, and a redesign proves that the company does.

What to do on day 91

Three moves follow. First, run the same quarter on the next process, because the method compounds and the second map takes half the time. Second, only now buy tools, since you can finally specify what a tool must do and measure what it did. Third, make the habit a condition of growth, so every new hire inherits a map, and every new vendor passes the export test before a contract is signed.

In August, I offered a test: If every model disappeared on Monday morning, would your business notice? Here is the companion test once you have run this quarter: If you hired 10 people on Monday morning, would they know by Tuesday how the work happens? The companies spending billions are buying back the quarter you still have, and the only transformation budget it requires is the discipline to spend it.

When I wrote about the difference between using AI and being AI-native, the most frequent response from executives was a polite objection: the distinction is real, but redesigning how work happens is a luxury reserved for companies with transformation budgets.

I spend my working life inside those budgets, and most of what they buy is the expensive undoing of habits that formed when the company was small. A founder has nothing to undo yet, which is why the quarter ahead is worth more than the three-year program a large enterprise will fund.

Here is the plan I would run if I were building a company today with no budget beyond my own calendar. In 90 days, with no new software and no outside help, you can do the three things that separate an AI-native business from one that merely uses AI: Write down how the work actually happens, take ownership of your data, and redesign who decides and what gets measured.