→  SPACE  /  CLICK
Joao Caldas
FP&A Director · Nordics
Miele Nordics · Finance Leadership

Artificial
Intelligence

From copy-paste finance to AI-native finance — practical, governed, and one level at a time.
Chapter I — Context

Every revolution looks like hype.
Until it doesn't.

The steam engine. Electricity. The internet. Each was dismissed as a bubble by serious people — right up to the moment it rewrote how work gets done. So let's look at the last four years with clear eyes.

Four years, in fast-forward

2022 → 2026: from party trick to working colleague

NOV 2022
ChatGPT launches
A chatbot that writes. The fastest-adopted product in history.
2023
Multimodal models
GPT-4, Claude, Gemini. Bar-exam scores. Vision, code, analysis.
2024
Reasoning arrives
Models that think before answering. PhD-level benchmarks fall.
2025
The agent era
AI that uses tools — browsers, spreadsheets, codebases — for hours, unsupervised.
2026
Today
AI drafts, reconciles, forecasts and presents. The question is no longer if.
The model landscape

The ecosystem exploded. Finance needs model literacy, not model loyalty.

2022
The door opens
ChatGPTGPT‑3.5
2023
Frontier race
GPT‑4ClaudeLlamaCopilot
2024
Multimodal workbench
GPT‑4oClaude 3.5GeminiMistral
2025
Reasoning + agents
o‑seriesClaude 4Gemini 2.5GrokDeepSeek
2026
Enterprise orchestration
ChatGPT EnterpriseM365 CopilotConnectorsTasksAgents
Message for finance: we should stop asking “which chatbot is best?” and start asking “which model, connector, control and workflow should handle this decision safely?”
Adoption & exponentials

AI adoption is fast — and the curves are steep

Fast adoption

ChatGPT reached 100 million users in just two months — faster than the telephone (75 years) or the internet (7 years). Adoption is now the slow part.

Exponentials

Frontier models surpass human experts while the cost of intelligence collapses. Training compute doubles every six months. Linear intuition fails on curves that double.

CHAPTER II

The world
noticed.

When a technology keeps presidents, regulators and its own inventors awake at night, it has left the hype phase.

Governments moved first — a sentence you rarely hear

Regulated faster than any technology before it

European Union
The EU AI Act
The world's first comprehensive AI law — risk tiers, audits, fines up to 7% of global turnover.
28 nations + EU
Bletchley & the safety summits
The US, China and the EU — who agree on almost nothing — co-signed declarations on frontier AI risk.
United Nations
A unanimous resolution
The General Assembly adopted its first resolution on safe, trustworthy AI. Unanimously. Think how rare that word is.

Nobody convenes global summits about a fad.

Market voices · consolidated carousel

Different voices. One signal: work is becoming AI‑mediated.

All the people and their AI messages are consolidated here on one single page. Use this as market context, not celebrity worship.

Hover to pausePortraits · Wikimedia Commons where available
The full spectrum of worry

From your job description to human extinction

Jobs & skills
Knowledge work — analysis, reporting, drafting — sits directly in the blast radius. Finance included.
Security & trust
Deepfakes, fraud, cyber-attacks at machine speed. CFO voice-clone scams are already real.
Energy & climate
Data centres on track to draw electricity on the scale of entire countries.
Control & existence
Serious scientists genuinely debate whether we stay in charge — in parliaments, not just podcasts.

You don't need to believe the far end of this spectrum. The near end is enough to change how we run finance.

CHAPTER III

Meanwhile,
in finance…

Superintelligence. Global summits. Nobel laureates losing sleep. The Pope, renaming himself for the occasion. Now let's check what we — the finance professionals of this new era — do all day.

An honest week in FP&A · industry surveys, and our mirrors

Three of every four hours go to moving data around

Gathering45%
Cleaning30%
Reporting20%
Insight5%
The sacred ritual: export from SAP → paste into Excel → fix the dates →
fix the decimals → VLOOKUP → #N/A → reconcile Excel against another Excel → repeat.
Exhibit A — tap to play
Live from the ninth circle of month-end

We are highly educated professionals doing this

Exhibit B — tap to play
Further evidence, your honour

Any resemblance to real analysts is entirely intentional

Exhibit C
Exhibit D
Further evidence, your honour

Another look from the trenches

Exhibit E — tap to play
Here is the part that should keep you up at night — in a good way

The AI you can use today is the worst AI you will ever use.

It will never again be this limited or this slow. Everyone in this room already has access. The gap is no longer technology. It's imagination and habit. So — three things it already does. Watch.

Make adoption playable

You do not jump to the final boss. You level up.

01
Noob
Ask & learn
Safe wins: summarize, rewrite, translate, prepare, learn. Trust is built by removing tiny frictions from the day.
02
Pro
Analyze & repeat
Reusable prompts, templates and packs. Finance output becomes faster, cleaner and easier to challenge.
03
Hacker
Connect & automate
Outlook, Teams, SharePoint, Calendar, Excel and Power BI become a controlled finance operating surface.
04
God Mode
Redesign & govern
The question changes from “can AI do this task?” to “should this task still exist in the first place?”

The adoption recipe: pick one painful process, copy one prompt, ship one result, improve one workflow.

Use cases · Noob level

The first wins should feel almost embarrassingly useful.

Outlook
Inbox to decision brief
Summarize unread threads into urgent items, waiting items, risks and suggested replies.
Prompt
Summarize today’s unread email. Return decisions, risks, owners and suggested replies.
Calendar + Teams
Meeting prep in five minutes
Pull context from prior meetings and emails. Produce a one-page prep note before the call.
Prompt
Prepare me for my next meeting. Include history, open actions, risks and questions.
Writing
Finance translation
Rewrite technical finance commentary for commercial leaders in clear business language.
Prompt
Rewrite this variance explanation for a non-finance audience. Keep it direct.
Learning
Teach me the file
Upload a messy spreadsheet and ask what it contains, what looks wrong, and what to check first.
Prompt
Explain this workbook like I joined today. Flag risks and unclear definitions.
Use cases · Pro level

Now we turn prompts into repeatable finance output.

Excel + ChatGPT
Dataset to story
Actual, budget, FC1, FC2 and LY become drivers, caveats, one chart and a slide title.
Prompt
Find the top variance drivers, anomalies, executive summary and recommended chart.
Excel Copilot
Quick graph studio
Create bridges, pivots and charts from finance tables without playing Excel archaeology.
Prompt
Create a bridge chart and highlight the five largest positive and negative movements.
SharePoint
NLT pack builder
Collect source files, prior actions and commentary drafts into a structured management pack.
Prompt
Build the monthly pack from this folder. Separate facts, judgment and open questions.
Power BI
Explain the dashboard
Turn KPI movements into a written narrative with root-cause hypotheses and follow-up questions.
Prompt
Explain what changed in this dashboard and what I should ask the business.
Use cases · Hacker level

Connect the work graph. Automate the boring. Escalate the risky.

Scheduled workflow
Morning finance brief
Every morning: sales, margin, AR, meetings and exceptions. Notify only when something matters.
Prompt
Every weekday, send an exception-only finance brief at 07:30.
Close cockpit
Evidence checker
Compare the close checklist with folders and flag missing reconciliations by owner.
Prompt
Check this close folder. List missing evidence and draft owner follow-ups.
AR assistant
Collections triage
Scan overdue items, customer history and notes. Draft collection actions for approval.
Prompt
Prioritize AR follow-up by value, age, relationship risk and prior promises.
Control monitor
Internal sales sentinel
Monitor SPC/internal net sales movements and flag non-zero netting issues before closing.
Prompt
Track internal sales/SPC. Flag material non-zero netting and owner gaps.
Enterprise reality

The magic is not the chatbot. It is the connected workflow.

Outlook
Calendar
Teams
SharePoint
Finance
Workbench
ChatGPT Enterprise + Microsoft 365 Copilot + governed company data + human approval gates.
Excel
Power BI
Tasks
SOPs
01Source
02Summarize
03Challenge
04Approve
05Scale
Demo 01 · Reporting

Variance commentary, before the coffee cools

Q2 · MSEK
Actual
FC1
Δ
Net sales
1,021
1,000
+21
Gross margin
41.8%
42.5%
−0.7pp
Opex
238
245
−7
EBIT
189
180
+9
AI draft · generated from the table on the left
Demo 02 · Analysis

The PVM bridge that builds itself

1,000 +38 −22 +14 −9 1,021 FC1 Price Volume Mix FX Actual

Net sales bridge, MSEK — decomposed from raw transactions, the same way, every single month. No pivot tables were harmed.

Demo 03 · Close & controls

Reconciliation that explains its own breaks

GENERAL LEDGER BANK STATEMENT INV-2041 · 84,200.00 INV-2042 · 12,750.00 PAY-0713 · −45,000.00 INV-2043 · 84,200.00 FX-0299 · 9,318.42 12,750.00 · REF 2042 84,200.00 · REF 2041 −45,000.00 · REF 0713 9,318.40 · REF 0299 — ⚠ 0.02 FX ROUNDING — AUTO-EXPLAINED DUPLICATE OF INV-2041 — FLAGGED FOR REVIEW

Matched in seconds. Breaks not just found — explained. Your team reviews the two red lines, not the five hundred grey ones.

Demo · Excel & Dashboards

ChatGPT + Excel: variance, graphs & dashboards

Variance analysis video
Variance analysis — from raw data to drivers, commentary & one chart.
Dashboard summarization video
Dashboard summary — explain KPI movements & ask better questions.
Demo · PowerPoint & Prep

ChatGPT + PowerPoint & meetings

PowerPoint build video
Slide building — draft, refine & design with AI.
Meeting preparation video
Meeting prep — pull context, open actions & risks in minutes.
And the rest of the menu — all real, all today

What "the worst AI ever" does for finance

Planning
Forecast first drafts
Driver-based baselines for FC cycles, with assumptions listed and challengeable — not buried in cell C47.
Communication
Deck & narrative drafting
Board-ready structure from your numbers, in your voice — you edit judgment, not formatting.
Risk
Anomaly detection
A tireless second pair of eyes over every line, every day — not a sample, once a year.
Data
Cleaning & normalising
Dates fixed, decimals fixed, hierarchies mapped — the janitorial 75% of the week, automated.
Pricing
Price-volume-mix at SKU level
Continuous PVM decomposition across the portfolio, not a quarterly heroic effort.
Business partnering
Instant answers for MDs
"Why is Denmark down?" answered with data in minutes — while the meeting is still going.
The prize

The week we get back

Today
Gathering45%
Cleaning30%
Reporting20%
Insight5%
AI-assisted
Gathering10%
Cleaning10%
Reporting25%
Insight & partnering55%

Same team. Same headcount. Ten times the thinking. That red block is where finance earns its seat at the table.

CHAPTER V

Your turn.

This afternoon's workshop is not about my ideas — it's about yours. Bring your most annoying task. Three lenses:

Lens 01
Automate
What do you do every week that follows rules? If you can write the instructions, AI can probably do it.
Lens 02
Augment
Where would a brilliant, tireless junior analyst change your output? Drafts, challenges, second opinions.
Lens 03
Reimagine
If reporting took minutes instead of days — what would finance do with the week we get back?
In one breath

What I hope you take into the room

I
It's not hype
Fastest-adopted, fastest-improving, fastest-cheapening technology in history.
II
The world is serious
Laws, summits, Nobel laureates, the Vatican. When inventors warn about their invention — listen.
III
We have headroom
75% of our hours go to work machines already do. That's not a threat — it's a gift.
IV
The move is ours
The worst AI ever is on your desk today. Imagination is the only bottleneck.
Miele Nordics · Finance

The best time to start
was four years ago.
The second best is this afternoon.

Joao Caldas · FP&A Director Nordics · Obrigado — Tack
joaoccaldas.github.io/sintra-ai