From "AI could help everywhere" to "do this first"
In a1 you mapped where AI is real in a finance function and named your three biggest time-sinks. Tonight you turn that raw list into a ranked shortlist. Most AI programs do not fail because the technology was weak - they fail because the team started with a hard, high-risk, high-visibility use-case, got burned, and lost the room. The fix is boring and powerful: score every candidate the same way, then sequence deliberately.
Value vs risk: the scoring grid 5 min live
Two questions decide the order of every AI bet in finance. How much value does it create, and how much risk does it carry if it goes wrong? Score both on a simple low-medium-high scale, plot the candidate on a 2x2, and the sequence almost writes itself. High value plus low risk is where you start. High value plus high risk is where you go later, with governance in place. Everything else waits or is dropped.
LiveHow to score value - and be honest about it2 min▶
Value is not "how impressive the demo was". It is the concrete good the use-case creates for your function, measured four ways. Score each candidate low, medium, or high on the whole picture - and be honest, because you will have to defend the number in a5.
- Time saved. Hours per month this returns to your team. A weekly report that takes six hours and drops to one is real, recurring value.
- Error and rework reduction. Fewer restated numbers, fewer late-night corrections, fewer manual re-keys. Quality counts even when the clock does not move much.
- Capacity freed. What your best people do instead - analysis and judgement rather than copy-paste. This is the value CFOs care most about but measure least.
- Dollars. Where it is honestly attributable - faster collections, caught duplicate payments, avoided penalties. Do not inflate it, but do not ignore it.
LiveHow to score risk - the axis that keeps you off the front page3 min▶
Risk is why two use-cases with identical value belong in different quarters. Drafting a variance note and deciding someone's credit limit are worlds apart, even though both "use AI on financial data". Score risk on three dimensions and take the highest of the three.
| Risk dimension | Low | High |
|---|---|---|
| Data sensitivity | Already-public or synthetic data | MNPI, client PII, unpublished financials |
| Decision impact | A human reads a draft and decides | The output decides on money or a person |
| Regulatory exposure | Internal note, no filing | Feeds a filing, a regulator, or a customer |
Notice the pattern: risk climbs the moment a use-case touches sensitive data, makes a decision instead of a draft, or reaches outside the building. Anything scoring high on any one dimension is not banned - it is a "govern first" bet for a3, not a first pilot.
The use-case inventory 5 min live
Here is a starter map across the whole finance function, already scored the way Part 1 describes. Do not treat the verdicts as gospel - your data sensitivity and your controls will shift a row up or down. Treat it as a worked example of the scoring, and as a checklist so you do not miss a candidate hiding in treasury or tax.
| Use-case (function) | Value | Risk | Verdict |
|---|---|---|---|
| Variance commentary (FP&A) | High | Low | Start here - quick win |
| Forecasting assist (FP&A) | High | Medium | Strong - human owns the number |
| Close reconciliation (accounting) | High | Medium | Strong - verify every match |
| AP invoice extraction (AP/AR) | Med-High | Low | Quick win - easy pilot |
| Cash-position summary (treasury) | Medium | Low | Easy pilot - internal only |
| Credit scoring (risk / credit) | High | High | Big bet - govern hard (a3) |
| Fraud / AML flags (risk) | High | High | Assist only - human decides |
| Earnings summary (investor relations) | Medium | Med-High | Internal draft only - MNPI care |
| Tax research assist (tax) | Medium | Medium | Assist - licensed sign-off |
Read the table top to bottom and the sequencing story is obvious. The rows verdicted "quick win" and "easy pilot" are internal, low-sensitivity, and draft-and-analyse - exactly the safe first bets from a1. The "govern hard" and "human decides" rows carry decision impact and regulatory exposure; they are real opportunities, but they belong after you have earned trust and stood up governance, not on day one.
LiveWhy the same task scores differently in two firms2 min▶
The inventory is a starting point, not a verdict for your firm. The same use-case can be a quick win in one function and a govern-first bet in another, and the difference is almost always on the risk axis.
- Data changes the score. An earnings summary built from an already-published 10-K is low risk. The same summary built from a pre-release draft is MNPI handling - high risk - even though the task looks identical.
- Controls change the score. Fraud flags where a human reviews every alert are an assist. The same model auto-declining transactions is a decision - and a regulated one.
- So score your version. Do not import a verdict from a webinar. Take the candidate, run it through your data, your decision impact, and your regulatory exposure, and place your own dot on the grid.
Two treasury teams, one use-case. Both want an AI cash-position summary. Team A feeds it yesterday's already-reconciled balances for an internal morning note - low risk, ship it. Team B wants it to move funds between accounts automatically - now it is a decision touching money, needs approval gates, and lands in a3's territory. Same words on the slide, different quadrant.
Sequencing: earn trust before you climb 4 min live
A scored shortlist is only half the answer. The other half is order. The teams that win start high-value and low-risk, keep it internal, and build a verification habit before anything reaches a customer or regulator. The teams that stall try to prove the program with the hardest, most visible use-case first - and one public mistake sets them back a year.
LiveThe order that earns trust - and the order that loses it2 min▶
- Stage 1 - internal, over your own data. Drafting and analysis your team reads and uses, on data you supply. Variance commentary, an invoice-extraction pilot, an internal cash note. Low blast radius, fast to value, builds confidence.
- Stage 2 - build the verification habit. Before you climb, make "a human checks each result" the default, not the exception. This habit is the thing that lets you safely take on harder use-cases later - and it is what a3 formalises into governance.
- Stage 3 - customer or regulator facing. Only now do you approach decisions that leave the building. By this point you have wins to point to, a team that trusts the tools, and the controls to govern the blast radius.
This is the same climb as a1's three eras, seen from the sequencing angle: get excellent at assistive and augmented work internally, before you let anything agentic or customer-facing run. a6 turns this into a dated 24-month roadmap.
Score 5 candidates and pick your first bet ◐ 15 min · with your team
No code, just the grid and a conversation. This is the artifact a3 governs and a5 turns into an ROI story - so do it honestly, not optimistically.
List five real candidate use-cases from your function - start with the three time-sinks you brought from a1, then add two more from the inventory in Part 2 that apply to you.
Score each one on value (low / medium / high) using the four value dimensions, and on risk (take the highest of data sensitivity, decision impact, regulatory exposure).
Plot all five on the 2x2 grid. Circle the one sitting furthest into the top-left - high value, low risk, internal. That is your first bet.
Write one sentence for each of the other four: is it "second", "later, after governance", or "not now"? Being able to say why you are not doing something is as important as your first bet.
Official sources covered
This session teaches the use-case identification and scoring approach from public finance-AI curricula. Certificates, graded assessments, and paid tooling stay with the official programs - noted honestly. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · What makes a strong first AI bet for a finance function?
Start top-left on the grid: high value, low risk, internal. It delivers fast, has a low blast radius, and builds the trust you need before climbing to harder bets.
2 · Two use-cases create identical value. Why can the risk axis still put them in different quarters?
Risk is scored on data sensitivity, decision impact, and regulatory exposure. A use-case that decides on a person's money or feeds a regulator must be governed first, whatever its value.
3 · What is the right sequence for rolling out AI across the function?
Earn trust before you climb. Get excellent at internal, low-risk work and make human verification the default, then approach decisions that leave the building - which a3 governs and a6 sequences.