*** Use this bundle page to register for two or more in the Corporate Finance Series and receive 15% off. ***
Most people in this room will spend far more time reviewing and approving AI-assisted work than building it themselves — though plenty build workflows of their own too, for their own review and analysis work, or to hand off to a team. Either way, the moments this session is built for are the same: approving a workflow someone else built, getting pulled in when something that used to run fine suddenly isn't, or fielding a request to skip a safety step because someone's in a hurry — your own request to yourself included. This session gives you the judgment for exactly those moments — supporting, debugging, and guiding a workflow, whether it's a team member's or your own, without becoming the bottleneck and without waving through something you don't understand.
The session opens by running a complete accounting workflow (ONBOARD) start to finish, live, and naming every element as it appears on screen — this is the workspace, this is the file that governs the assistant, this is the evidence trail. It briefly locates that framework against COSO's internal control model before moving on, so the session opens with a recognized standard underneath it rather than an invented one. The next three hours take those elements apart one at a time: what data may safely reach a tool, what an executive is authorizing when they approve an agent file, why a number can disagree between two systems, how to tell whether the script that ran is the script that was approved, and how to classify a claim as fully supported, partially supported, or unsupported.
Most blocks keep their hands-on element as a decision rather than a keystroke — sort and defend real tasks, classify data fields, diagnose a broken join, read a script against a checklist — with one deliberate exception early on where participants actually type. Since this is a fully online session, exercises run as individual work followed by a poll or open group discussion, rather than small-group table work.
Discussion Leader: Svetlana Toohey
Lana Toohey, CPA, MAcc, is Controller at SDII Global LLC, a multi-entity forensic engineering and consulting firm serving clients across the United States and Canada. She leads financial reporting, consolidations, foreign currency translation, lease accounting, and operational finance initiatives for a complex and growing organization.
Lana began her career as an auditor at Deloitte, building a strong foundation in discipline, structure, and technical precision. She later spent nine years with Mettler Toledo, supporting capital investments, factory transformations, restructurings, business-unit consolidations, and a major SAP migration.
These experiences shaped her leadership approach: strong fundamentals, collaboration, and respect for compliance in an increasingly complex business environment.
In recent years, Lana has focused on a critical question: How can finance operate smarter? That focus led her into automation, AI copilots, and coding. She now applies AI-enabled development and productivity tools to design better processes, troubleshoot reporting, automate data tasks, and accelerate analysis.
This journey led to PythonMuse (www.pythonmuse.com), a learning platform helping accounting and finance professionals use Python, AI, and automation in a responsible, audit-ready way.
Lana’s mission is to help shape the future of accounting by preparing finance professionals to work effectively alongside AI and an increasingly agentic workforce. Her message is simple: AI will not replace you—but professionals who know how to use it will redefine the role."
AI foundations — capability tour (24 min)
Opens with what's in it for this audience: most will review and approve AI-assisted work far more often than build it, though plenty build too — either way, the session's job is the judgment to support, debug, and guide a workflow, theirs or a team member's. Then names the audience's range directly — some attendees already use AI daily, and that's fine, the next few minutes aren't the point. What an LLM does and does not reliably do in accounting work; a live demonstration of a confident, fluent, wrong answer, before anyone touches a keyboard. Engagement: posture self-assessment poll.
Elements of a workflow — live demonstration (20 min)
A complete accounting workflow run start to finish, narrating each element as it appears on screen, briefly located against COSO's control model along the way. This becomes the map the rest of the session points back to.
Markdown files (12 min)
The plain-text format everything else is written in, and why. The session's one hands-on-keyboard moment: a short, low-stakes exercise authoring a single file.
The work harness (14 min)
When a simple AI chat is enough, and when a structured harness is required. Individual exercise: sort your own tasks (brought from your close) into chat vs. harness; then a poll and group discussion.
Masking internal data (20 min)
Masking as a control decision, not a formatting one — what to require before any of your organization's own data reaches a tool. Individual exercise: classify data fields, then decide whether a masked file may leave the organization; poll on the outcome.
The agent file (22 min)
What an executive authorizes when they approve one — role, rules, data boundaries, and the difference between written guidance and enforced behavior. Individual exercise: draft one, then compare a couple of answers as a group.
Skill definitions (18 min)
What makes a procedure repeatable, and what breaks when a required element is missing. Individual exercise: diagnose a broken example.
Data shape — PDF, CSV, Excel, database (26 min)
Why some data sources are trustworthy on arrival and others need to be earned. Individual exercise: diagnose a broken join caused by data-shape mismatch.
Version control, in one question (18 min)
Not a tool tutorial — can you prove the script that ran is the script that was approved? Individual exercise: interpret a diff.
Reading a script (20 min)
A six-point checklist, applied to a real analysis script — explicitly not a request to write or edit code. Individual exercise: judge whether a script did what was asked.
Validating output (20 min)
Claim-by-claim classification of AI-drafted commentary: fully supported, partially supported, unsupported. Individual exercise: classify a full set of claims against source data.
None
Participants are asked, before the session, to bring one real workflow from their own month-end close — the tasks, the data, the tool(s) currently used — that they'd want to see brought under this kind of control. Several exercises work directly with what participants bring, so the group discussions compare real tasks rather than invented ones.
Finance executives at every level who approve, oversee, or build AI-assisted accounting work — CFOs and finance directors alongside controllers and finance managers. Most of this audience spends more time reviewing than building, though many build workflows themselves too. The session assumes the professional skill this audience already has (judging work) and builds the specific vocabulary and control questions needed to apply that skill to AI-assisted output — whether the workflow in question is a team member's or your own. No coding background is assumed or required at any point; one short segment does invite participants to type, but nothing later in the session depends on it.
