Aryng

"Garbage in, garbage out has been true since the first spreadsheet macro. AI doesn’t change that principle. It just gives the garbage a far more persuasive voice and a much bigger platform to be wrong on...", writes ⭐️ Piyanka Jain ⭐️ in the latest edition of her newsletter. Link in comments.

"Are we using AI tools?" is the wrong question. The right one is "Are we ready to use AI tools?", and most finance teams aren't, not because they lack ambition, but because their data isn't in shape yet. Every CFO already runs the most underrated operation in the company: reconciling numbers that live in Stripe, QuickBooks, Gusto, Salesforce, and half a dozen other systems that were never built to agree with each other. That reconciliation effort eats close to half of FP&A's time before a single AI tool ever enters the picture. Point AI at that mess and it won't clean it up. It'll produce a confident, fluent, wrong answer, one that sounds exactly as certain as a correct one would. Here's the part that gets missed: you don't need every field in your data warehouse clean before AI becomes useful. You need what we call the CFO-100, the roughly hundred numbers that actually show up in your board deck, your investor update, your QBR. Reconcile that vital core, assign clear ownership, and you have a real foundation. Then, and only then, does the type of AI you choose start to matter. After all, finance doesn't need a model that sounds confident. It needs one that's deterministic: traceable, auditable, repeatable, the same input producing the same output every time, because that's the bar an audit committee actually holds you to. Data readiness isn't a prerequisite to check off quickly. It's the whole game. #CFO #FPandA #AIreadiness #FinanceLeadership Pedro Knotschke | ⭐️ Piyanka Jain ⭐️

CFOs are being asked "are you using AI?" as if that's a question with a clean answer. It isn't, because AI isn't one thing. A probabilistic language model generating your board narrative and a deterministic system calculating MRR from fixed rules are both "AI," in the same way a scalpel and a bandage are both medical tools. What actually matters is the task, the destination of the output, and whether an error is recoverable before it does damage. Those three things determine more about AI adoption in finance than any vendor demo will. We built a complete decision framework around this: three zones for classifying any finance task, five questions to answer before AI touches it, and a vendor evaluation protocol designed for the demo room, including the one question that almost never gets asked and that predicts post-implementation failure better than anything else on the list. ⭐️ Piyanka Jain ⭐️ | Pedro Knotschke Link in comments.

Anthropic states it plainly in their own documentation: even with temperature set to zero, their model will not be deterministic. Identical inputs may produce different outputs across API calls. Read that again if you work in finance. Most AI tools being sold into finance right now are probabilistic at their core. They generate answers by predicting which word is statistically most likely to come next. Sometimes they get the same answer twice. Sometimes they don't. The difference between those two outcomes is invisible in the interface. This is the architectural problem I kept running into before we built AskEnola. Finance has a different contract with numbers than almost any other function: same inputs, same rules, same output, every time. That's what makes a number auditable, defensible, and signable. The way we solved it: the model never touches the math. SQL runs against your actual data, business logic executes as a rule, and the LLM steps in only to write the sentence around a number that was already calculated. The number is never the model's to decide. We are writing a bunch of stuff for CFOs who are evaluating AI tools and aren't sure what questions to ask. Start with this one: does your system calculate the number, or generate it? Link in comments.

Is your favorite AI incentivized to cost you more? With Shanti Greene and Jake Mendlovic Link in comment

“Why so serious?” 🤡 Because when AI starts sounding confident and wrong… things get interesting. Episode 3 of Not Another AI Podcast explores the AI trust problem, from hallucinations and bias to transparency and accountability. Can we really trust the technology we use every day? 🎧 Watch here: https://lnkd.in/gt-TqpND ⭐️ Piyanka Jain ⭐️ Shanti Greene Ant Newman #AI #ArtificialIntelligence #Podcast #TechPodcast #GenerativeAI #TrustInAI

The conventional wisdom is that CFOs are slow to adopt AI because they're risk-averse. That's not quite right. 96% of CFOs say AI integration is a priority, even though only 59% are using it in their finance function. The gap isn't reluctance. It's a very specific architectural problem that most vendors aren't talking about honestly. Here's the core of it: accounting is deterministic. Same inputs, same rules, same output, every time. However, the AI solutions being sold to finance teams right now are largely probabilistic. They generate plausible answers. For most use cases, plausible is fine, but for a CFO whose signature ends up on the number, plausible is the failure mode. We wrote about what would actually close this trust gap: what auditability really means, the three levels most tools claim vs. what they deliver, and the one architecture that holds up in an audit room. Link in comments. ⭐️ Piyanka Jain ⭐️ Pedro Knotschke Aditya Gautam #CFO #AIinFinance #FinanceLeadership #FinancialReporting