Are your systems built for the way you run the business, or are you missing competitive insights hidden in your data? Winning management tools provide you with a measurable competitive edge.

The full process: the 75% you should buy and the 25% you should build

For most of my career the last 25% of a finance system, the matching, chasing, reformatting and re-keying that a standard ERP never quite covers, was done by hand, by capable people with better things to do, or by expensive custom code. AI tools now write that last layer for a fraction of the cost. I can direct the build. The order still matters: think, then buy, then build.

Step one: map the process you actually have

Before any demo, run the five steps on paper. List every report, approval and hand-off in the monthly cycle, name an owner for each, and delete what no one defends. You will likely discover that a quarter of what you were about to automate should simply stop. This typically takes two to three weeks and an abundance of sticky notes on a whiteboard, and it is normally the diagnosis I do at the start of every engagement unless there are other higher-priority needs pressing your organization.

Step two: fit it into an off-the-shelf ERP

A well-chosen mid-market ERP covers about 75% of what a company needs, in my experience, on a core that someone else supports, patches and secures. Choose on fit and total cost of ownership, not on the demo. Questions I ask every vendor:

• Can the general ledger, payables, billing and inventory (if any) live in one system with one chart of accounts?
• How is AI used to shorten and eliminate clerical busyness, and what third-party software tools with AI adaptation make your ERP even stronger for our case use?
• What does the standard close look like out of the box, and what did your last three customers of my size customize?
• What is the all-in cost over five years, including the implementer, the integrations and the people who will maintain it?
• Who is using this system nearby? We would like to talk to and visit multiple mature users. We are in no mood to be a guinea pig.
• How do we get our data out if we leave?

At my last engagement the answer was consolidating several stand-alone accounting sources into one system and integrating third-party disbursement cycle tools, so the department plan needed at least one fewer hire. The consolidation itself produced clean data; the clean data is what made everything after it possible.

Step three: build the rest with AI

The remaining 25% is specific to your business: matching vendor invoices to purchase orders and receipts, chasing approvals, preparing sales and property tax filings, assembling the board packet from the ledger, reviewing leases and contracts for capital and operating spending. Every business is unique and offers many additional use cases. These used to take a skilled person's hours or a consultant's invoice. Today my AI agents, built with tools such as Claude and supervised by your accounting manager, do the routine work, and the CFO reviews the output.

Three rules I hold to:

1. A human reviews every cycle. The accounting manager reviews AI-prepared disbursements and receipts; the CFO reviews journal entries, accruals, monthly financial and KPI artifacts, and the board package. The agent drafts; a person signs.
2. Build on clean data only. An agent on top of a messy ledger produces confident nonsense. Step two comes first.
3. Measure it like labor costs and purchased capital. Every agent gets a cost, an expected saving in hours or dollars, and a review date. If it does not pay, it goes.

What it is worth

Four payoffs, in the order they usually show up: cash sooner (shorter close, billing and collection cycles), lower operating cost (clerical work shrinks, managing capacity grows), lower capital cost (a standard core instead of a custom build), and decisions sooner (reporting reaches the people who can act while it still matters). At my last engagement, a sub-$20 million revenue business, I worked with the team to improve annual cash flow by more than $3 million, starting with daily manager visibility of spending against budget.

Why someone has to own it

Systems do not pay back on their own. The research, mostly from larger companies, keeps pointing to the same cause: someone senior has to be accountable for the result.

• McKinsey (1,491 organizations, 2024): CEO oversight of AI governance was the element most correlated with self-reported bottom-line impact from gen AI. Only 28% of respondents said their CEO holds that role (McKinsey, The state of AI).
• McKinsey and MIT (about 100 companies, 2025): for 44% of operations leaders, digital projects were sponsored by the CEO or board, more than double the rate of the bottom half (McKinsey, Bold accelerators).
• Gartner (782 IT infrastructure and operations leaders, April 2026): 77% of those with a successful AI use case credited integrating AI into existing workflows and securing full executive support. Gartner also says CEOs and CFOs need a more active role in setting funding criteria for major AI investments (Gartner, April 2026).
• KPMG Global AI Pulse (Q2 2026, 2,145 senior leaders in 20 countries, companies above $50 million in revenue): where AI decisions have clearly defined accountability, strategy confidence is 60% against 22%, value realization 57% against 21%, and established ROI 14% against 4% (KPMG, June 2026).

In my experience the same pattern shows up in smaller companies, where the owner is the CEO and the question is who runs it every day. That is why I work as a CFO and chief AI officer (CAIO) in one chair: one accountable person for the numbers and the technology.