Writing
Writing on enterprise planning, the office of the CFO, and what AI changes in finance. Each piece links to the original on LinkedIn.
AI can already connect to Oracle EPM and automate much of an administrator's work. The remaining gap is translating a logical financial design into a physical application, which means teaching AI how an experienced architect decides what to build.
Read on LinkedInThe second video demo. Three AI models set their own assumptions for a fictional food manufacturer facing a 2027 downturn, and every scenario runs through one Excel model. The combined forecast cuts 2027 EBITDA from $4.8 million to $3.2 million, with the models spread between $2.9 million and $3.4 million.
Read on LinkedInAnnual reviews reconstruct a year from memory, and calibration happens out of view. A six-step quarterly process where AI maintains the evidence, both sides are evaluated against the same agreed framework, and the disagreements that remain are visible to everyone.
Read on LinkedInAs planning software works through an identity crisis, the definition of what to calculate is becoming the real asset. A declaration layer keeps that logic outside any one platform, so Oracle, Anaplan, OneStream, Python, or Excel can be built from it and held to the same verification checks.
Read on LinkedInA $10 billion infrastructure finance role whose responsibilities never mention AI, what AI-enabled FP&A work looks like in practice, and why hiring for it may move faster than waiting on an enterprise AI committee.
Read on LinkedInSeven steps from a governed account catalog to a monthly P&L, balance sheet, and indirect cash flow, with AI building the historical mapping and reconciling actuals in minutes rather than days.
Read on LinkedInPart four of the critical thinking series. A 2013 meeting about Hyperion server blades, and why curiosity keeps us in the first diamond long enough to understand the problem.
Read on LinkedInThe first video demo: a functioning model and orchestration solution producing an on-the-fly scenario in under 4 minutes, and what comes next on the path to autonomous forecasting.
Read on LinkedInFour takeaways on Phase 0 in a post-AI world, including why pre-implementation now has two objectives and why the data will never be ready.
Read on LinkedInA utility client never automated its actuals load because every objection was legitimate. Years later the copy and paste continues.
Read on LinkedInPart three of the critical thinking series. Why 'I was wrong' is the most avoided sentence in corporate life, and what saying it does for trust.
Read on LinkedInThree independent LLMs run the same scenario against a copy of the model, and the spread between them becomes the confidence score. Agreement is not accuracy.
Read on LinkedInAI writes the RFP, AI writes the responses, AI scores them, and a human decides over lunch. Why selection should be rebuilt around prototypes.
Read on LinkedInDozens of implementations, and never once have consultants and client agreed on the planning grain. A 2x2 diagnostic for finding the disagreement early.
Read on LinkedInAn account catalog, a design file, and a folder structure are enough for Claude Code to build the model, run auditable scenarios, and write its own manual. Where a gap in experience still shows.
Read on LinkedInThat's the mistake, and almost everyone makes it. Part two of the critical thinking series, on seeing the room through the audience's eyes.
Read on LinkedInLeading AI models given the same model to build. All three balanced the balance sheet, then disagreed on what matters. A passing check is not a correct model.
Read on LinkedInPart one of the critical thinking series, with a hand-drawn double diamond and a steering committee that spent a whole meeting making the wrong thing faster.
Read on LinkedInMost finance teams aren't deploying it, and many aren't especially interested. Five reasons why, and a question about which one applies to yours.
Read on LinkedInClose and compare, split the variance, learn from what's left, run it forward, repeat. Part three of the machine learning series.
Read on LinkedInWhen actuals arrive, the miss splits into exceptional events, human decisions, and true model error. Only the last one retrains the model. Part two.
Read on LinkedInML predicts the drivers. The model calculates everything that follows. A model should never guess a number a formula can compute exactly. Part one.
Read on LinkedInA client and a consultant describe the same program and both are telling the truth. On the gap between delivered scope and delivered value.
Read on LinkedInA recruiter asked whether I'd rather hire finance people who know AI or AI people who know finance. Neither. The gap is problem-solving.
Read on LinkedInAn empty office, a laptop, every LLM available, and a public company. Build a driver-based forecast the CFO can run scenarios on, live. You have 60 minutes. How would you do?
Read on LinkedInI said Claude shouldn't be used to balance a balance sheet. Then I built a three-statement forecast in four hours and ran the integrity check. Every statement tied.
Read on LinkedInTake your time. The list is short, and faster AI delivery makes it shorter unless partners are chosen on their ability to execute.
Read on LinkedInA ranch and a steakhouse on summer vacation, both run by families thinking decades out. What long-lived private businesses know about planning.
Read on LinkedInA prospect asked a question I couldn't answer, so I said so. On the most honest thing a consultant can say in a first meeting.
Read on LinkedInFind the last missed milestone and listen to how people reacted. Nothing sounds toxic, and that's the point.
Read on LinkedInAn old friend challenged my decision to run the firm with no budget, ever. How to give investors a projection, and how to hold people accountable to something other than an outdated target.
Read on LinkedInQ1 closes. In under two minutes the actuals are validated, the forecast and 10-year plan are refreshed, and a 1,000-path Monte Carlo hands the CFO an EPS distribution. Nobody opens a spreadsheet.
Read on LinkedInFor as long as I can remember, planning transformations have started with the same question: which platform? Why that question no longer comes first.
Read on LinkedInAfter the kickoff, the sponsors leave and the team designs a future it was never incentivized to reach. Why 'nobody ever got fired for buying IBM' is an admission, not a compliment.
Read on LinkedIn