How to Turn an AP AI Pilot Into a Program: The AI Use Case to Program 90 Day Plan

TLDR: An AI use case to program 90 day plan is the sequence a transformation lead follows when one agent in accounts payable or procurement is live, the CEO wants more, and there is no dedicated team. The plan has three 30 day blocks: lock the evidence from the first agent, build the intake and prioritization machinery, then launch the second and third use cases under the same measurement standard. Most companies skip the middle block. That is why the second use case takes longer than the first.

Best For: Heads of Transformation, Heads of AI, and COOs or finance leaders carrying the AI mandate at enterprises of 1,000 to 15,000 employees in manufacturing, distribution, healthcare services, hospitality, or insurance, who have one AI agent working in finance and no repeatable way to produce the next one.

Updated: October 2026

An AI use case to program 90 day plan is a sequenced operating plan that converts a single working AI agent into a repeatable pipeline of use cases with one owner, one intake path, one measurement standard, and a monthly review. The plan is not a strategy document and not a platform build. It exists because the first use case almost always succeeds on the strength of one motivated person and a cooperative function, and the second one has neither. The 90 days are spent building the things the first win borrowed and never paid back: a baseline, an intake template, a scoring rule, and a named owner for every live agent. For an enterprise in a traditional industry with one AP or procurement agent in production, this is the difference between "we did a pilot" and "we run a program."

Why Does the AI Use Case to Program 90 Day Plan Start With a Finance Pilot?

An AI use case to program 90 day plan usually starts in accounts payable or procurement because that is where the first live agent sits at most enterprises in traditional industries. Finance intake work is high volume, rules based, and already measured, so the first agent produces a number the CFO believes. That number is the only asset the program has on day one, and the plan is built to spend it carefully.

The pattern shows up in the data. Gartner's November 2025 finance survey found 59% of finance functions using AI, with accounts payable automation the second most common use case at 37% of adopters, behind only knowledge management. The same survey found 91% of finance organizations reporting low to moderate impact from their first deployments. That is where most transformation leads reading this actually are.

The situation this plan is written for

The reader of this plan has one agent live, often invoice matching, PO exception handling, or supplier onboarding checks. The CEO saw the number and asked for a program. Functional leaders in operations, HR, and customer service have gone quiet. Quiet is what resistance looks like at this level. Nobody has a team. The transformation lead is doing this alongside a day job, which IBM's 2025 CEO study makes concrete: only 25% of AI initiatives delivered their expected return, and only 16% scaled across the enterprise, mostly because the organization around the initiative never changed.

What the first win hides

The first win hides a missing operating model. The agent worked because the AP manager wanted it to, because the data was already structured, and because the transformation lead personally handled every exception for the first six weeks. None of that transfers. The 4 gaps that stall enterprises after one AI use case are owner, intake, prioritization, and functional buy-in; this plan is the 90 day answer to all four.

What Are the Three Blocks of the AI Use Case to Program 90 Day Plan?

The AI use case to program 90 day plan has three 30 day blocks: Lock, Build, Launch. Days 1 to 30 lock the evidence from the first agent into a baseline and a delta the CFO will sign. Days 31 to 60 build the intake template, the scoring rule, and the monthly review. Days 61 to 90 launch use cases two and three through that machinery, so the program is proven on something other than the pilot.

One sequencing rule governs the whole plan: no new agent enters build until the one before it has a signed baseline and a named business owner. That rule is the mechanism. Everything else in the plan exists to make the rule survivable.

Days 1 to 30: Lock the evidence

The first block does one thing. It turns the pilot's informal result into a defensible number. Most first agents were measured by whoever built them, with a before figure reconstructed from memory. That number will not survive a board question. PwC's 2026 Global CEO Survey found 56% of CEOs have seen no significant financial benefit from AI to date and only 12% report both cost and revenue gains; the difference is rarely the technology and usually the measurement.

The work in this block is to baseline team time before deploying AI on the next two candidate workflows, and to retroactively rebuild the baseline on the live one using system timestamps instead of surveys: invoice received to invoice approved, PO exception opened to PO exception closed, touches per transaction. Record the delta as hours returned per week and exceptions per hundred transactions, with the AP manager's name on the sign off. At the end of day 30, the program owns one number that a CFO will repeat.

Days 31 to 60: Build the machinery

The second block builds three artifacts and nothing else. An intake template that any functional leader can fill in fifteen minutes. A scoring rule that ranks requests on operating data, not on a workshop vote. A monthly review with a fixed agenda. McKinsey's 2025 State of AI found that high performers were far more likely to have redesigned workflows and assigned senior ownership before scaling; the machinery in this block is the small-company version of the same thing.

This is also the block where the plan borrows the 4-part lean AI operating model and PMO setup: one owner, a use case pipeline, a monthly review, and a measurement standard. The transformation lead does not need a team to run it. They need a calendar invite, a spreadsheet, and the authority to say no.

Days 61 to 90: Launch two and three

The third block launches use cases two and three. Two, not one. A program that has produced two agents is still a pair of pilots. Three is the smallest number at which the intake and scoring rule have been tested on a request the transformation lead did not personally originate. The candidates usually come from the same finance and procurement neighborhood: three way match exceptions, supplier master data cleanup, contract renewal flagging, expense audit sampling. Gartner's June 2025 forecast that over 40% of agentic AI projects will be canceled by end of 2027 names the cause as unclear business value and inadequate controls. A second use case launched without a baseline looks like that a year later.

The Intake Template: What Every Request Must Contain

An AI program intake template is a one page form that a functional leader fills before any use case is scored, and it is the most important artifact in the 90 day plan. The template forces the requester to name the workflow, the system of record, the current volume, the current cycle time, the exception rate, and the person who will own the agent after go live. A request that cannot answer those six fields is not ready to score.

The template is deliberately boring. Deloitte's 2025 State of Generative AI in the Enterprise found that two thirds of organizations expected 30% or fewer of their experiments to scale within six months; a form that filters out requests with no volume data and no owner is how a program gets its scale rate above that line.

How the scoring rule uses the template

The scoring rule ranks every intake on four dimensions: impact, ease, technical feasibility, and buy in. Each is scored from operating data in the template, not from a vote. Impact is monthly volume multiplied by cycle time. Ease is inverse of exception rate. Feasibility is whether write back to the system of record is needed. Buy in is whether the post go live owner is named and has signed the form. The 5-dimension method for prioritizing AI use cases with data covers the scoring in more depth; the 90 day plan only needs the four fields above to rank three candidates.

What the monthly review decides

The monthly review decides three things and then ends: which live agents are above or below their baseline delta, which intake requests enter build next, and which agents are retired or paused. The review takes an hour. The transformation lead runs it, the CFO or COO attends, and the functional owners present their own agents. That last detail matters more than any slide. When the AP manager presents the AP agent, the program has a second owner.

Pilot Expansion vs. Program Build: Which One Are You Doing?

Pilot expansion and program build look identical in the first month and diverge sharply by the third. Pilot expansion adds a second agent using the same person, the same informal measurement, and the same reliance on one cooperative function. Program build adds the second agent through an intake, a baseline, and a review that would work if the transformation lead left tomorrow. The 90 day plan is the second thing.

The trade off is speed in month one. Program build is slower for the first 30 days because the baseline and intake work produces no visible agent. BCG's September 2026 Applied AI Index found that only 7.5% of companies have reached the stage where AI value compounds, with 2.8 times the EBITDA growth of laggards, while 41% are scaling and the remainder are still in pilot mode; the slow first month is what separates the 41% from the rest.

Where a generic plan is not enough

For an enterprise where the next use case touches the general ledger, requires write back to a multi entity ERP, or sits under SOX controls, a generic 90 day plan is not enough. The answer there is a dedicated program build that does three things: designs the human in the loop control pattern before the agent is built, defines the approval thresholds and audit trail with internal audit in the room, and sequences the rollout one entity at a time. The properties that matter are an owner inside the function, a control design signed by audit, and a baseline per entity. Anything that skips those is a pilot wearing a program's clothes.

What Does the AI Use Case to Program 90 Day Plan Measure?

The AI use case to program 90 day plan measures four things per agent and one thing per program: hours returned per week, exceptions per hundred transactions, cycle time delta, and human in the loop load per agent; and pipeline pass rate for the program as a whole. Pipeline pass rate is the share of intake requests that reach production within two review cycles. Below one in three, the intake is too loose or the scoring rule is broken.

Why hours returned beats headcount

Hours returned per week is the right first metric because it is measurable from timestamps, it does not require a headcount decision, and it is what functional leaders will defend. EY's December 2025 research found that most US companies are reinvesting AI productivity gains instead of cutting roles, so the capacity effect is the number the board will eventually ask about. Hours returned also sidesteps the surveillance worry: the baseline is built from system events, not from watching people.

The 90 day benchmark

By day 90 a program built on this plan should show one agent with a signed baseline and delta, two more in production with baselines measured before build, three intake requests in the queue from leaders the transformation lead did not recruit, and one monthly review completed with the CFO present. Capgemini's 2025 agentic AI research found that only 2% of organizations had deployed agents at scale and 14% had begun partial deployment; three measured agents in 90 days puts an enterprise ahead of most of its peers without any platform spend.

Common Objections From Operations Leaders (And What to Say)

Operations leaders raise the same three objections to a 90 day program plan: there is no team, the second use case will be as slow as the first, and functional leaders will not fill in forms. All three deserve a direct answer rather than a slide, because each one is partly right and the plan is built around the part that is.

"We don't have a team for this"

You do not need one for the first 90 days. The plan requires one owner with the authority to say no, a functional owner per agent, and an hour a month from the CFO. MIT's 2025 State of AI in Business report found that 95% of enterprise pilots produced no measurable profit and loss impact, and that the successful minority typically had a single accountable business owner rather than a large central team. The team comes later, if the pipeline pass rate justifies it.

"The second use case will be as slow as the first"

The second use case will be slower in build and faster in production if the intake and baseline work is done first. The first agent spent its time on discovery that nobody wrote down. The second spends its time filling a form that forces the discovery upfront. Stanford's 2025 AI Index documented that the cost of running AI at a given capability level fell roughly 280 fold between late 2022 and late 2024. The slow part of use case two is never the technology.

"Functional leaders will not fill in an intake form"

Some will not, and that is information. A leader who will not supply monthly volume and a named owner has told the program that the use case is not ready. Bain's 2025 Global Private Equity Report found that fewer than 20% of portfolio companies had operationalized AI use cases with concrete results, and the common factor in that minority was a business owner who could state the outcome before the build. The intake form is how the program finds those people.

This analysis was developed using methodologies and operating experience from Assembly.

Frequently Asked Questions

What is an AI use case to program 90 day plan?

An AI use case to program 90 day plan is a sequenced operating plan that turns one live AI agent into a repeatable pipeline. It has three 30 day blocks: lock the evidence from the first agent, build intake and scoring machinery, then launch two more use cases under the same measurement standard, with one owner and a monthly review.

Why does the plan start with an accounts payable or procurement agent?

Finance intake work is where most first agents live because it is high volume, rules based, and already measured. Gartner's 2025 finance survey found accounts payable automation was the second most common finance AI use case at 37% of adopters, which makes it the most common starting point for a program.

What is the difference between pilot expansion and program build?

Pilot expansion adds a second agent using the same person and informal measurement, while program build adds it through an intake, a baseline measured before build, and a named functional owner. The two look identical in month one and diverge by month three, when pilot expansion runs out of the transformation lead's time.

What happens in days 1 to 30 of the plan?

Days 1 to 30 lock the evidence from the first agent into a baseline and delta the CFO will sign. The work is rebuilding the before figure from system timestamps, recording hours returned per week and exceptions per hundred transactions, and getting the functional manager's name on the sign off.

What happens in days 31 to 60?

Days 31 to 60 build three artifacts: a one page intake template any functional leader can complete, a scoring rule that ranks requests on operating data instead of a workshop vote, and a monthly review with a fixed agenda. No agent is built during this block, which is why most companies skip it.

What happens in days 61 to 90?

Days 61 to 90 launch use cases two and three through the intake and scoring rule. Two are launched rather than one because a program with exactly two agents is still a pair of pilots; the third is the first use case tested on a request the transformation lead did not originate.

What should an AI use case intake template contain?

An intake template contains six fields: the workflow trigger, the system of record, monthly volume from the system, current cycle time from timestamps, exception rate, and a named post go live owner inside the function. A request that cannot answer all six is not ready to be scored or built.

How is the scoring rule different from a workshop vote?

The scoring rule ranks use cases on operating data from the intake form: impact as volume multiplied by cycle time, ease as the inverse of exception rate, feasibility as whether write back is required, and buy in as whether a functional owner has signed. A workshop vote measures enthusiasm; the rule measures the workflow.

What is the sequencing rule inside the plan?

No new agent enters build until the previous one has a signed baseline and a named business owner. That single rule is the mechanism behind the plan. It is what prevents the second and third use cases from borrowing the transformation lead's time the way the first one did.

What does the monthly review decide?

The monthly review decides three things: which live agents are above or below their baseline delta, which intake requests enter build next, and which agents are paused or retired. It takes an hour, the CFO or COO attends, and functional owners present their own agents rather than the transformation lead.

Which metrics does the plan track?

The plan tracks hours returned per week, exceptions per hundred transactions, cycle time delta, and human in the loop load per agent, plus pipeline pass rate for the program. Pipeline pass rate is the share of intake requests reaching production within two review cycles; below one in three signals a broken intake.

Why measure hours returned instead of headcount?

Hours returned per week is measurable from system timestamps, needs no headcount decision, and is the number functional leaders will defend. EY's December 2025 research found most companies reinvest AI productivity gains rather than cut roles, so capacity is what boards eventually ask about.

Do you need a dedicated team to run the 90 day plan?

No dedicated team is needed for the first 90 days. The plan requires one owner with authority to say no, a named functional owner per agent, and an hour a month from the CFO. A team is justified later only if the pipeline pass rate shows demand that one owner cannot process.

What does a good day 90 look like?

A good day 90 shows one agent with a signed baseline and delta, two more in production with baselines measured before build, three intake requests queued from leaders the transformation lead did not recruit, and one monthly review completed with the CFO in the room. Anything less is pilot expansion.

When is a generic 90 day plan not enough?

A generic plan is not enough when the next use case touches the general ledger, needs write back to a multi entity ERP, or sits under SOX controls. Those situations need a dedicated program build with a human in the loop control design signed by internal audit and an entity by entity rollout.

Why do most second AI use cases take longer than the first?

Second use cases take longer because the first one borrowed an operating model it never built. IBM's 2025 CEO study found only 16% of AI initiatives scaled across the enterprise, mostly because ownership, intake, and measurement were never formalized after the pilot. The 90 day plan builds those three things before the second agent starts.