Process design is the first AI investment
For companies without a transformation budget, the practical route to AI capability starts with how work gets done. A quarter spent documenting one revenue-linked process, organizing the data it depends on and clarifying human decisions can create more durable value than a collection of disconnected pilots.
The distinction between using AI and becoming AI-native comes down to whether the business has redesigned its work around the technology. Buying access to models doesn’t resolve unclear handoffs, scattered information or decisions nobody owns. Those conditions limit what AI can do, regardless of the number of tools in use.
That’s a particularly relevant opening for founders and smaller companies. Large organizations often spend heavily to unwind habits and systems that accumulated over years. A younger business has fewer established routines to replace. It can use that flexibility to make processes explicit before informal workarounds become permanent.
The proposed 90-day effort needs no new software or outside support. It centers on three practical tasks: map one process that affects revenue, establish ownership and standards for the data involved, then assign decision responsibility and measure business outcomes.
Why operational gaps matter to AI performance
A June survey by Genpact and HFS Research of more than 2,000 senior executives found that 85% said accumulated enterprise debt was limiting the value they receive from AI. The researchers estimated that this debt—such as undocumented processes, unusable data, aging systems and unprepared employees—was associated with roughly $18 trillion in trapped value across the Global 2000. They also reported that just 33% of enterprise data was AI-ready, while inefficient or manual work consumed about 40% of a typical employee’s week. More than half of surveyed enterprises had no funded effort to address those constraints.
Those figures point to a management problem as much as a technology problem. If staff rely on undocumented judgment or information stored in isolated systems, introducing a model won’t automatically make the work faster or more reliable. The missing foundation is often a shared understanding of the process and the information it uses.
McKinsey’s Global Survey on AI found that nearly three-quarters of high-performing AI organizations had fundamentally redesigned workflows, compared with roughly one-quarter of other organizations. The share of smaller organizations scaling AI agents, meanwhile, remained at 22%. The figures suggest that workflow redesign separates stronger results from experimentation. A smaller company can begin that work without waiting for a major technology program.
Days 1–30: Map one process from start to finish
Select a single workflow tied directly to revenue, such as quote-to-cash, customer onboarding or product returns. Keeping the scope narrow makes the work manageable and gives the team a clear baseline for judging whether changes help.
Ask the people who perform the work to describe what happens in practice. Capture the spreadsheets used along the way, approvals that sit in inboxes, exceptions handled from memory and any other informal steps. A formal procedure may describe the intended path, but the real process often contains workarounds that determine how long a task takes.
For each step, distinguish routine production from human judgment. Routine production includes drafting, collecting information, comparing records and summarizing material—tasks AI may be able to support. Judgment includes prioritizing, making consequential choices and accepting responsibility for an outcome. As work is mapped, keep human judgment visible rather than treating every step as a candidate for automation.
By the end of the first month, the team should have a concise process map that a new employee could follow. That document gives employees a common reference and helps the business identify where a model might fit. If nobody can explain a process clearly, it’s difficult to assign work to a model, assess its output or identify who is accountable when something goes wrong.
Days 31–60: Put the data in order
Use the process map to identify what information each step requires and where that information lives. Customer records, product details, prices, statuses and dates may be spread across business applications, email and personal files. For every important data field, assign one person to maintain it and agree on a consistent definition. A business can spend a great deal of time debating what qualifies as an active customer; that question may be settled quickly once someone owns the definition.
Data ownership also means checking whether information can move between systems. Test whether each paid tool can export the business’s data in full, in a usable format and when the company needs it. If an export is limited or difficult to obtain, that affects how useful the information will be in other workflows. It also gives the company a concrete issue to address at renewal.
This work connects to the broader question of who controls the data and context a business creates. AI depends on access to relevant information, but access alone isn’t enough: teams need to know what a field means, who maintains it and whether they can retrieve it. Without those basics, even a capable model may work from incomplete or inconsistent inputs.
Days 61–90: Assign decisions and track business results
Return to the judgment points identified in the process map. For each one, name the person responsible, define what a satisfactory outcome looks like and set a clear rule for when an employee must review work before it reaches a customer. This gives staff and any AI system a more useful boundary: routine tasks can move ahead, while consequential decisions remain assigned to a person.
Review the process’s approval steps as well. Some approvals protect against real risks; others may persist because information is difficult to see or because the organization has always required them. Once the relevant data is available to the people involved, the team can assess which checks still serve a purpose.
Then measure the workflow itself. Record its cycle time, cost to serve and customer turnaround around day 61, and measure those same indicators again on day 90. The comparison offers a clearer view of operational change than a count of AI subscriptions or pilot launches. A model can perform well in a test while the wider process remains slow; redesign evaluates whether the business has improved how work moves from beginning to end.
These measures won’t explain every cause of a change, but they provide a grounded starting point. If results have improved, the team can examine which process changes contributed. If they haven’t, the map, data standards and decision rules give employees specific places to investigate. That makes the next iteration more informed than simply adding another pilot.
Day 91: Repeat the work with a sharper brief
After one process has been mapped, organized and measured, the company can apply the same method to another workflow. The first effort establishes the approach; subsequent maps may take less time because employees already understand what information and decisions to record.
Only after the process is clear does a software purchase have a defined job to do. The company can set requirements based on actual workflow needs and assess whether a tool changed cycle time, service costs or customer turnaround. That’s a better basis for evaluating a purchase than the size of a feature list or the number of licenses available.
Make the discipline part of how the company grows. Give new employees a process map they can use, assign responsibility for keeping it current and check that new vendors let the business export its information. These habits help prevent today’s improvised work from turning into tomorrow’s operational debt.
The useful test after 90 days is simple: could ten new hires learn how the chosen process works by their second day? If the answer is yes, the business has created a foundation for repeatable work, clear accountability and measured AI use. The next quarter can extend that foundation to another process—and any later spending can follow a more precise operating need.
