Why this session exists
A pile of good use-cases is not a strategy. What turns them into one is sequence and story - the order you do them in, and how you explain the whole thing to the people who hold the budget. This finale hands you both. First we look honestly at where finance AI is actually going over the next 24 months, so your plan is ambitious without being fantasy. Then we sequence the portfolio so the foundation comes first - because skipping it is the single most reliable way to make good pilots fail. Finally we build the one board slide that carries the whole plan.
The agentic future, realistically 6 min live
The 2026 curricula have shifted. Leading programs now teach agentic systems - AI that plans and executes multi-step tasks - and multimodal AI that reads charts, documents, and tables together, as first-class finance topics. That is a real signal about direction. But the useful question for a CFO is narrower: over the next 24 months, what will finance agents actually do in your function, and what will they not do? The honest answer keeps your roadmap credible.
LiveWhat finance agents will and will not do in 24 months3 min▶
An "agent" is AI that plans and carries out several steps toward a goal, not just a single reply. In finance, over a two-year horizon, the realistic scope is bounded and supervised - useful, not autonomous.
- Likely, and worth planning for. A governed close assistant that gathers, reconciles, and drafts variance commentary for a human to approve. A forecast assistant that pulls drivers, runs scenarios, and surfaces the ones worth a look. A filings assistant that reads a 10-K and answers grounded questions. All bounded, all with a person signing off.
- Not in 24 months - and be clear about it. An autonomous CFO. AI that closes the books, files with regulators, or moves money without a human in the loop. Trading or credit decisions made and executed by the model alone. These fail on governance, model risk, and accountability - not just capability.
- The dividing line. Agents shorten the path to a decision; the licensed human still makes and owns it. That line is where your a3 guardrails live, and it does not move just because the demo is impressive.
Self-studyMultimodal, in one paragraph2 min read▶
Multimodal AI reads more than text - a chart, a scanned invoice, a table, and a paragraph together, in one pass. For finance that means an assistant can look at a page of a board pack the way a human does, rather than needing everything pre-extracted into clean rows. It is genuinely useful for document-heavy work like filings review and invoice handling. It does not change the rails: the same privacy, sign-off, and audit-trail rules apply, and a confident read of a messy scan can still be wrong. Treat it as a capable reader that needs checking, not an oracle.
Sequencing the portfolio 6 min live
You have a portfolio of business cases from a5. The strategy is the order you run them in. There are three phases, and they are not optional stops you can skip - each one earns the right to the next. The most common way roadmaps fail is jumping straight to the exciting phase and discovering the foundation was never there.
LiveThree phases: foundation, adoption, transformation3 min▶
Read the table top to bottom - it is the order, not a menu. Value grows as you climb, but so does the cost of having skipped a rung.
| Phase | What you do | What it produces | Rough window |
|---|---|---|---|
| Foundation | Set the AI-use policy, define tool tiers, run literacy across the team (a3 + a4) | A team that can use AI safely, and rails everyone trusts | Now - 6 months |
| Adoption | Put copilots on high-value, low-risk work; run the pilots from your a5 cases with baselines and decision gates | Measured wins, real adoption, a credible track record | 6 - 12 months |
| Transformation | Move proven copilots into governed agentic workflows on bounded tasks, human still signing off | Compounding capability on the work that matters most | 12 - 24 months |
LiveWhy the order is the strategy2 min▶
The phases are not just chronological - each produces the thing the next one needs.
- Foundation buys trust. Adoption only happens on rails people believe in and skills they actually have. Without it, the best tool sits unused.
- Adoption buys credibility. A few measured wins - a close cut by two days, hours redeployed to analysis - are what earn you the mandate and budget to attempt agentic work. You cannot skip to transformation on a promise.
- Transformation buys leverage. Only once a workflow is proven, adopted, and governed is it safe to let an agent run more of it. Automating a broken or unmeasured process just breaks it faster.
The team that jumped to phase three. A finance group, impressed by an agent demo, tried to automate the whole close before anyone had run a copilot or written an AI policy. The agent produced plausible-looking entries no one trusted, there was no baseline to prove value, and the auditors had questions no one could answer. They quietly shelved it and restarted with literacy and one small copilot pilot - the foundation they had skipped. A year of momentum lost to skipping two rungs.
The board narrative 4 min live
A board does not fund a use-case list; it backs a story it can repeat. The whole leader track collapses into five beats a board can follow in a few minutes - where we are, the bet, the guardrails, the milestones, and the ask. This is the a5 business case zoomed out from one use-case to the portfolio, and it belongs on a single slide.
LiveThe five beats of the story3 min▶
Tell it in this order. Each beat answers the question the board asks next.
- Where we are. The honest starting point, with a number. "59% of finance functions use some AI; we run two copilots today and have no shared policy." No hype, no shame - just the baseline.
- The bet. The direction and why now. "Nearly four in five CFOs plan to deploy generative AI within 24 months; we intend to lead our sector's adoption, not react to it." One sentence of ambition, grounded.
- The guardrails. The a3 rails, named up front. "AI assists, the licensed human decides. Privacy, sign-off, and audit trail on every use-case." A board trusts the leader who raises risk before being asked.
- The milestones. The three phases as a timeline with checkpoints. "Foundation by month six, two measured pilots by month twelve, first governed agentic workflow by month twenty-four - each with a decision gate."
- The ask. The specific decision you want today. Budget, a mandate, a named sponsor, the next review date. Make it easy to say yes.
Notice the shape: this is the a5 one-pager - problem, bet, cost/guardrails, value/milestones, ask - lifted from one use-case to the whole portfolio. If you can defend one business case, you can tell this story.
LiveOutline of the board slide2 min▶
One slide, five zones, read top to bottom in three minutes. Keep it to a page on purpose - if the story does not fit, the strategy is not finished.
- Header - the bet. One line: where finance is going and our stance. This is the title, not a footnote.
- Left - where we are. Two or three honest facts about today, with the market context number for scale.
- Center - the roadmap. The now → 6 → 12 → 24 month strip with a theme per stage: literacy and quick wins, copilots, governed agents.
- Under the roadmap - the guardrails. One line, always visible: "AI assists, the human decides - privacy, sign-off, audit on every case."
- Bottom - the ask. The single decision, in bold. Budget, sponsor, next review date.
Sketch your 24-month roadmap ◐ 15 min · on paper
Pull together everything from a1 through a5 into one roadmap sketch and the bullets for one board slide. You are not building anything - you are proving the pieces connect into a story you could defend in the room where the budget lives.
- Draw the strip: now, 6, 12, and 24 months. Give each stage a one-line theme (literacy and quick wins → copilots → governed agents on bounded work).
- Under each stage, name one or two concrete moves - pull the real ones from your a5 business cases and a4 team plan.
- Mark where the foundation ends. Be honest: is anything in your plan trying to skip it?
- Write the five board-slide bullets: where we are (with a number), the bet, the guardrails (your a3 rails, verbatim), the milestones, the ask.
- Draft the ask in one sentence - the exact decision, sponsor, and next review date.
Official sources covered
This finale teaches from a 2026 finance-AI curriculum and published implementation practice. Certificates, graded assessments, and paid tooling stay with the official programs - noted honestly. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · In a 24-month finance-AI roadmap, what must come first?
Foundation first: policy, tool tiers, and literacy. Skipping it is the most reliable way to make later pilots fail - a team with no skills will not adopt the tool, and rails written after an incident are damage control, not governance.
2 · What is a realistic scope for finance agents over the next 24 months?
In 24 months the honest scope is bounded and supervised: agents shorten the path to a decision, but the licensed human still makes and owns it. Autonomous decisions on the close, filings, credit, or trading fail on governance and accountability, not just capability.
3 · What does a board narrative need to do its job?
A board backs a story it can repeat, not a use-case list. The five beats - where we are, the bet, the guardrails named up front, the milestones, and a specific ask - are the a5 business case zoomed out to the portfolio, and they belong on one slide.