AI That Actually Ships: Howard Blazzard at the CFO Roundtable

At the invitation-only K-State/CoBank CFO Roundtable on May 12, 2026, Prosody’s Howard Blazzard took the room through a subject every finance leader is wrestling with — what artificial intelligence actually changes about the work. His session, Artificial Intelligence for Agricultural Co-op CFOs, wasn’t a hype reel. It was a practical field guide in four parts: where AI is genuinely being used in cooperative finance today, the data and governance foundation it depends on, the risks a CFO has to manage, and how to start. It was grounded in interviews with three agricultural cooperatives and in Prosody’s own hands-on results — including the talk itself, which Howard built with AI assistance and then reviewed and approved, practicing exactly what he preached.

What AI is actually being used for in co-op finance today

Rather than survey the technology, Howard mapped what co-ops are really doing with it, on a spectrum from proven to piloting to emerging:

• Robotic process automation and AP automation. The proven, lower-risk starting point: rules-based automation for invoice processing, GL reconciliations, payroll uploads, and member-statement generation, plus AI-powered OCR that learns invoice fields without manual template setup and can auto-code invoices from historical GL patterns. His key distinction: RPA follows rules; AI learns patterns. Start with RPA — it frees real staff time and builds the board confidence you’ll need later.
• Document AI. The example that lands immediately with CFOs: condensing CoBank credit agreements, audit reports, board-prep packages, vendor contracts, and regulatory guidance from hundreds of pages to their key themes in minutes. One co-op’s team ran a 600-page climate-disclosure peer analysis this way. The tools are available today — ChatGPT, Microsoft Copilot, Claude — with no custom build required.
• Financial and variance analysis. FP&A platforms are moving from static PDFs to conversational analytics — asking your data questions in plain language — and cash-flow forecasting that detects patterns in grain movement before borrowing spikes. The prerequisite he flagged: three-plus years of clean data. Audit your data depth before you invest.
• Logistics and member-facing AI. Energy-logistics routing that integrates weather data, remote tank monitors, and fleet scheduling to cut miles driven and prevent members from running dry — fewer miles, zero outages, lower inventory — plus emerging member-facing uses like precision-ag recommendations and credit scoring.

A recurring caution ran through this section: as vendors bundle AI assistants into everything, watch for “AI bloat,” and be deliberate about which built-in features you actually turn on.

Prosody’s own results

Howard grounded the “does this really work” question in Prosody’s own practice, where the firm uses Claude as a junior research assistant and copy editor — AI drafts first, humans lead every refinement:

• Opportunity Zone 2.0 research. Census-tract analysis, regulatory cross-referencing, and first-draft compliance memos cut research time roughly 70% — from about 40 hours per engagement to 12 — while letting the team analyze three times as many census tracts.
• PIDP grant writing. Parsing NOFOs and prior award patterns, drafting benefit-cost analyses and merit-criteria narratives, synthesizing partner MOUs and letters of support, and cross-checking against NOFO and Buy America requirements turned around first drafts about 60% faster and saved 50-plus hours per application cycle — with more consistent narratives. (These are efficiency gains in Prosody’s grant writing process, not a claim about awards won.)
• Project management. A project-specific master tracker goes live and triaged on Day 1 of an engagement, with version history and provenance notes that keep filings defensible — cutting end-to-end delivery time by about a third and producing a full field-level audit trail.

The through-line: this is what lets a small firm punch above its weight — and it only works with a human in the loop at every step.

Building the foundation: data quality and governance

Howard’s sharpest line was the one every CFO should tape to the wall: AI does not fix bad data — it amplifies it. The path runs from scattered data across multiple ERPs and silos, to “golden records” (a single, clean, validated source of truth), to AI outputs you can actually trust.

Governance, he argued, is a three-legged stool: IT owns the infrastructure and pipes the data flows through (many co-ops are still modernizing legacy ERPs); Accounting owns validation and audit — the function that catches hallucinations before they reach a report; and FP&A translates AI output into CFO decisions. Board oversight sits on top, setting AI policy, risk tolerance, and investment approval. The CFO’s job is to champion data governance as an AI prerequisite — not to hand it off as a separate IT project.

Managing the risks

Risk management is the CFO’s core competency, and Howard framed AI risk in the same discipline, across four buckets:

• Data risk. Grain positions, member financials, and credit data exposed to cloud tools that may train on your inputs. Review your data-classification policy before deploying anything.
• Operational risk. Confident-sounding but wrong outputs, and finance decisions made on hallucinated numbers. Build human verification into every workflow — never let AI output go straight into a report.
• Compliance risk. Vendor contracts that may conflict with bank reporting, USDA requirements, or state privacy law. Get legal and audit review of AI agreements before deployment, not after.
• Vendor risk. A fast-consolidating market and the danger of depending on a vendor that fails or pivots away. Evaluate stability, data portability, and exit provisions before signing — and favor a small number of strategic partners over sprawl.

He gave particular weight to shadow AI — like shadow IT, but with fewer guardrails: staff pasting patron financials, hedging strategies, and member lists into free consumer tools with no data-retention controls and no audit trail, creating invisible exposure and potential fiduciary liability for officers and directors. The answer isn’t to ban experimentation but to govern it: a clear strategy captures the efficiency gains while protecting members, the board, and proprietary data.

Getting started

Howard closed with a phased, year-long adoption path built to survive real organizations:

• Pilot & Learn (months 1–3): a cross-functional champion team from finance, operations, grain, and IT runs a 90-day pilot on two or three high-volume, low-risk tasks, with baseline training in prompt fundamentals and the co-op’s own data-handling rules.
• Measure & Expand (months 4–6): track KPIs a board understands — hours saved per task, error rates before and after, cost per deliverable — and turn early wins into organizational buy-in.
• Scale & Embed (months 7–12): formalize approved tools, data boundaries, and review workflows into a board-adopted policy, and move AI from “project” to “process” with continuous ROI reporting.

His framing: AI fluency isn’t a technology project — it’s a culture shift. Top-down mandates tend to fail through indifference; blended teams working measurable use cases with real leadership support are what succeed.

The takeaways

Howard left the room with six: AI is a force multiplier when it’s implemented properly; unlocking it requires a genuine shift in how teams approach problems; a human stays in the loop, always; choosing the right tool for the right task is critical; governance comes before scale; and early movers compound their advantage. Or, as he put it: the question isn’t whether AI will change how cooperatives operate — it’s whether you’ll lead that change or follow it.

Related reading

• Two Conferences Worth Your Time
• Prosody Labs
• Prosody Labs OZ 2.0 Policy Monitor

Want to see what purpose-built, well-governed AI could do for your finance team’s back office or grant work? Howard and the Prosody Labs team are always happy to talk shop. Get in touch →