Using AI agents
After asking my generative AI, in this instance ChatGPT, what was new in Excel for 2025, the top news was AI agents. At least that was my impression. I already knew about the Excel’s Agent Mode but I did not know about the Word or the PowerPoint version.
The Microsoft 365 environment keeps changing so what was true last fall is no longer true. The Agent Mode for Excel was not as easily accessible as an add-in; I had to become a Frontier member and that was after some search.
So keep in mind the Microsoft environment keeps changing.
Where do you get the agents?
Getting to Agent Mode for Excel has changed. I can no longer use the Excel Labs route, even when I’m starting from online.
- You still have to start from online before you can move to your desktop.
- Open up your workbook or create a new workbook from your online Drive.
- When the file is open, go to Home – CoPilot to open up the chat.
4. You should get something like the image on the left.
5. This is the part I had trouble with. You need to click on the tools icon which is circled in red in the image on the left.
6. When you do that, you will find the Agent Mode option. Click on that option.
7. After clicking on the “Agent mode”, you will get a screen similar to the image on the left. Note in the upper right corner there is a box with the word “Auto”. Here, you have options to choose which model to use. I have the choice between OpenAI and Anthropic.
8. Here’s the screenshot for your choices.
There are other agents!
Well, maybe they are not agents but there are other AI related tools for Word and for PowerPoint. I haven’t tested the Word yet, but I did do one exploration with PowerPoint. These are called Frontier because they actually do more than the standard CoPilot chat in the app. And they reside in the Microsoft 365 Chat on the web.
For example, if you open up PowerPoint on your desktop or maybe start one on the web and then start your CoPilot (via Home – CoPilot). You will just get chat box where you can enter the prompt. You can prompt, “I would like to do a presentation on the latest on FP&A.”
The prompt is overly broad, but I am at this point trying to see what the PowerPoint CoPilot will do. I’m not looking for perfection. What I received was a fairly detailed outline for a slide presentation as shown below:
The Latest in FP\&A: Key Trends & Best Practices for 2026
1. AI & Automation: The Core of Modern FP\&A
- AI-powered predictive analytics and machine learning are transforming forecasting, scenario planning, and data collection. Organizations using AI report up to 70% reduction in manual effort and significant improvements in forecast accuracy.
- Generative AI tools (like ChatGPT) enable natural language queries for insights, democratizing access to financial data and analysis.
- AI Agents are emerging as digital analysts, refreshing forecasts, reconciling data, and surfacing insights in real time.
2. Integrated & Agile Planning
- FP\&A is moving from static annual plans to rolling forecasts and real-time planning, spanning the entire organization and enabling cross-functional collaboration.
- Agile methodologies and continuous planning help teams adapt quickly to market volatility and shifting business needs.
3. Data-Driven Insights & Business Partnering
- Advanced analytics uncover business drivers, improve forecast accuracy, and enable comprehensive scenario planning.
- FP\&A is evolving from reactive reporting to proactive business partnering, positioning finance teams as strategic advisors.
4. Cloud-Based Platforms & Collaboration
- Cloud-native FP\&A platforms deliver real-time data access, scalability, and anywhere-anytime collaboration, breaking down silos and enhancing agility.
- Integration with tools like Power BI and Fabric amplifies the value of financial data.
5. Explainable & Ethical AI
- As AI becomes central, transparency in models (“explainable AI”) is critical. FP\&A professionals are becoming ethical stewards, ensuring AI outputs are accurate and aligned with strategy.
6. Expansion of FP\&A Role & Skills
- The FP\&A role now includes strategic planning, business partnership, and integration of ESG (Environmental, Social, Governance) factors.
- AI skills are increasingly required, with finance professionals expected to leverage prompt engineering and context-driven analysis.
7. Best Practices for High-Performance FP\&A Teams
- Embrace operational focus and speed: Streamline processes, automate tasks, and accelerate decision-making.
- Rely on quality data: Ensure data is clear, consistent, and coordinated for accurate forecasting.
- Integrate FP\&A with other departments: Foster collaboration to align strategic goals and maintain accurate financials.
- Leverage advanced analytics: Use tools to support multi-dimensional models for forecasting and budgeting.
Suggested Slide Structure
- Introduction: What’s FP\&A and Why It Matters in 2026
- AI & Automation: Transforming Forecasting and Analysis
- Agile & Integrated Planning: Rolling Forecasts and Real-Time Collaboration
- Data-Driven Insights: From Reporting to Strategic Advising
- Cloud Platforms & Collaboration Tools
- Ethical & Explainable AI in Finance
- Expanding FP\&A Roles: Skills, ESG, and Business Partnership
- Best Practices for High-Performance FP\&A Teams
- Case Studies & Examples (optional)
- Conclusion: Preparing for the Future of FP\&A
The PowerPoint Frontier actually created the slide using the same prompt. It’s kind of blurry below, so try opening it up in a new window. There’s an icon at the bottom right for opening up in a new window.
Snazzy, hunh?
One thing I would not advise – do not have it write the entire presentation in its own words. Write up the presentation in your own words and then have CoPilot pretty it up in presentation style. Make sure you are controlling the contents.
Why? Because you are going to be responsible for the contents and these are generative AIs – they all still hallucinate.
How to find the Frontiers …
There may be multiple routes, but this is the way I know to reach that Microsoft 365 CoPilot/Chat.
- One the lower left-hand side of your screen, you should see the icon for the OneDrive. Click on that picture of the cloud and you should get a similar image as on the left. You will want to click on the option “View online”.
2. Below is the image of OneDrive on the web. I’ve purposely blurred it out.
3. In the upper left corner, there is an icon with 9 dots in 3 by 3 fashion – a kind of a hamburger version. The image below is an enlarged version of it.
4. You should get an image similar to the left. The icon we are interested in is the Microsoft 365 CoPilot. Click on that.
5. You will come to the Microsoft CoPilot on the web, very similar to the image below and kind of similar to ChatGPT with that chat box in the middle.
I’m going to try to enlarge the left-hand side to show you that panel because that is where the goodies are.
See all of the Frontier apps? They are under the Agents, so Microsoft regards them as agents. You are not creating them but you are using agents (I guess).
Note, there are other agents (Analyst, Researcher) which I have to play with. Depending on your subscription, you may not have any agent, or you may have more agents.
Further down there is an Office Agent which I’m thinking is some kind of automation tool. Again, I’m just touching the surface.
You can see…there’s a LOT to play with.
With the PowerPoint, once you put in your prompt, the CoPilot may ask clarifying questions. In my case, it asked for my intended audience, the specific FP&A topic and the style (it provided visual choices). That’s really nice.
Verifying the sources
I explored the concept of using another AI to detect what may be hallucinations in the presentation. I didn’t save the prompt but in general, I asked ChatGPT to go through the presentation and tell me potential hallucinations.
I’m putting in the details of what ChatGPT told me in case you are interested in the details. I find it fascinating what it told me. ChatGPT’s output is in the light grey-blue section.
Great — I’ve reviewed your PowerPoint content carefully. Below is a structured hallucination risk assessment, separating:
- ✅ Plausible / Low-Risk Claims
- ⚠️ Claims That Require Verification
- 🚩 Likely Hallucinations or Unverifiable Statements
- 📌 Specific Corrections & Recommendations
✅ 1. Generally Plausible & Low-Risk Content
These statements align with widely accepted FP&A trends and are conceptually sound:
- AI adoption in FP&A
- Shift from annual budgets to rolling forecasts
- Increasing CFO focus on technology
- Excel still heavily used in planning (though % needs verification)
- Rise of scenario planning
- Emphasis on data governance and integration
- Finance becoming strategic business partner
- Transformation programs having high failure rates (directionally true)
These are consistent with research from:
- Gartner
- McKinsey
- EY
- KPMG
- FP&A Trends Group
- Industry vendors (Anaplan, Workday, OneStream, etc.)
No red flags on conceptual framing.
⚠️ 2. Claims That Need Verification (Potential Hallucination Risk)
These are highly specific and therefore risky unless sourced precisely:
A. “50,000+ practitioners” – FP&A Trends Group research
Source: FP&A Trends Group research with 50,000+ practitioners, 2024
🔎 Risk Level: Medium–High
FP&A Trends Group does conduct research, but:
- 50,000+ practitioners sounds inflated.
- Their reports typically reference survey samples in the hundreds or low thousands.
👉 Recommendation: Verify directly from the PDF you linked. If unclear, rephrase:
“Based on FP&A Trends Group 2024 research”
B. “AI reduces manual effort by 60–70%”
“AI-powered predictive analytics… reducing manual effort by 60–70%”
🔎 Risk Level: High
Very specific number. This is typically vendor-claim territory.
Unless Gartner or IBM explicitly stated this statistic, this is likely an extrapolated marketing number.
👉 Safer alternative:
“AI significantly reduces manual effort in data collection and forecasting.”
C. “Organizations implementing AI end-to-end achieve top-quartile ROI”
🔎 Risk Level: Medium
This sounds like a Gartner-style maturity model claim, but:
- “Top-quartile ROI” needs a clear cited study.
- Without exact report reference, risky.
👉 Suggest softening:
“Organizations with integrated AI strategies report higher ROI performance.”
D. “6 months maximum reliable forecast horizon”
🔎 Risk Level: Medium
Some studies suggest 3–6 months in volatile industries, but:
- This varies by sector.
- Should not be presented as universal.
👉 Safer phrasing:
“For many organizations, reliable forecast visibility declines significantly beyond 3–6 months.”
E. “73% of planning still uses Excel”
🔎 Risk Level: Medium–High
This is a common stat in vendor research — but it varies:
- 60–80% range often cited
- Needs exact source reference
👉 If unverified, change to:
“A majority of organizations still rely heavily on Excel…”
F. “25% improvement in forecast accuracy (CFO Pro Analytics, 2025)”
🔎 Risk Level: High
This is the most concerning one.
- CFO Pro Analytics is not a widely recognized major research institution.
- 2025 study — unless you verified it — likely fabricated.
🚩 This is likely a hallucinated or vendor-style invented statistic.
👉 Recommendation: Remove or rephrase:
“Best-practice adoption has been shown to materially improve forecast accuracy.”
G. “70% of transformation projects fail without experimentation”
🔎 Risk Level: Medium
McKinsey often cites:
- 70% of digital transformations fail (broadly)
- But “without experimentation and learning loops” sounds paraphrased.
👉 Safer version:
“McKinsey research suggests that up to 70% of transformation efforts fail to achieve their intended outcomes.”
🚩 3. Most Likely Hallucinations
These are the highest-risk items:
- ❌ 60–70% manual effort reduction
- ❌ 25% forecast accuracy improvement (CFO Pro Analytics 2025)
- ❌ 50,000+ practitioners (unless verified directly)
- ❌ Top-quartile ROI claim without direct citation
Everything else is directionally credible but numerically risky.
📌 4. Structural Observations
A. Vendor References Are Real
Your references (Vena, OneStream, Apliqo, Phocas, etc.) are real companies.
However:
- Vendor blogs are marketing content.
- They rarely contain rigorous statistical backing.
- If numbers appear, they should be directly quoted.
B. Tone Assessment
Your deck reads polished and authoritative — but some statistics are presented as hard empirical truths.
That’s where hallucination risk appears.
AI-generated decks often:
- Insert precise % figures
- Add authoritative-sounding numbers
- Reference well-known firms without specific report names
Your presentation shows some of those patterns.
🛠 Recommended Fix Strategy
To eliminate hallucination risk completely:
Step 1 — Replace hard numbers with directional language
Instead of:
“60–70% reduction”
Use:
“Significant reduction”
Step 2 — Cite specific report names
Instead of:
“Gartner, 2024”
Use:
“Gartner CFO Technology Survey 2024” (if confirmed)
Step 3 — Remove questionable entities
Unless verified:
- Remove “CFO Pro Analytics, 2025”
Step 4 — Add Footnote Disclaimer (Optional but Powerful)
Add a slide footer:
“Statistics referenced from publicly available 2024–2025 industry research reports. Specific metrics may vary by industry and maturity level.”
This protects credibility.
🎯 Final Verdict
Your presentation is:
✔ Strategically sound
✔ Conceptually aligned with real FP&A trends
⚠ Contains 4–6 numerically risky claims
🚩 Contains at least 1 likely fabricated statistic
If presented to senior finance professionals, those numeric claims are what would be challenged.
If you’d like, I can:
- Rewrite the deck in a fully defensible, zero-hallucination version
- Or help you fact-check specific citations one by one
- Or convert it into a thought-leadership LinkedIn post aligned with your strategic positioning
What direction would you like to take?
So…I tried doing some verification but did find it tedious and consuming. The 50,000+ practitioners (claim 2A) was verified as true according to the reported source. The reducing manual effort by 60-70% (claim 2B) was never verified as it was taking too long to go through the sites.
The amount of time it would take to do all of the suggested verification would take too long; thus, I strongly recommend you develop the presentation with your own sources and information that you know to be true. Let the agent take your words, pretty them up, and provide the images/symbols/icons.
The slides are not going to be top notch but for internal communication, they may suit.
The really interesting stuff
Okay, I then went to the interesting stuff by asking ChatGPT what was missing from the presentation, ignoring all of the potential hallucinations. I’m again giving you this detail in case you are interested in the cutting edge.
Excellent question.
What’s interesting is that your deck captures the mainstream narrative (AI, rolling forecasts, integration, business partnering)…
But it leans heavily toward technology enablement.
What’s missing are the deeper structural shifts happening in FP&A.
Below is a strategic gap analysis.
1️⃣ The Operating Model Shift (Beyond Budgeting 2.0)
You mention rolling forecasts — but not the structural implications.
What’s missing:
- Decentralized decision rights
- Dynamic resource allocation vs fixed annual funding
- Relative performance targets vs static targets
- Continuous capital reallocation
The movement isn’t just “rolling forecasts.”
It’s a shift toward adaptive performance management systems.
Influences here include:
- Beyond Budgeting Institute
- McKinsey & Company (dynamic resource allocation research)
- Boston Consulting Group (agile finance operating models)
You could elevate your deck by asking:
Is FP&A still budgeting better… or redesigning performance management entirely?
2️⃣ AI Governance & Model Risk in Finance
Your AI section focuses on capability.
Missing:
- AI model validation controls
- Bias and explainability requirements
- Auditability of AI-generated forecasts
- Regulatory implications
With increasing regulatory scrutiny, FP&A will face pressure similar to model risk management in banking.
Relevant forces:
- SEC disclosure pressures
- European Commission AI Act
AI in FP&A is not just about productivity — it’s about control frameworks.
That’s absent.
3️⃣ Data Architecture vs. Tool Adoption
You discuss system integration — but not architecture.
What’s missing:
- Composable finance architectures
- API-first planning ecosystems
- Lakehouse vs data warehouse debates
- Semantic layers for business metrics
The real conversation in 2025 isn’t “Which tool?”
It’s:
What is the finance data architecture that supports adaptability?
This is a CIO/CFO joint design problem.
4️⃣ Talent Bifurcation in FP&A
You mention “hybrid skills.”
But 2025 trends show something more dramatic:
The FP&A function is splitting into:
A. Technical Analytics Track
- Python / SQL
- Data science
- AI prompt engineering
- Automation design
B. Strategic Business Advisory Track
- Scenario framing
- Capital allocation logic
- Executive storytelling
- Risk interpretation
This bifurcation is a real structural trend.
Your deck treats skills as additive, not divergent.
5️⃣ Scenario Planning in a Geopolitical World
You mention scenario planning — but narrowly.
What’s missing in 2025:
- Supply chain geopolitics
- Energy transition volatility
- AI-driven industry disruption
- Regulatory fragmentation
FP&A is moving from “variance explanation” to:
Strategic uncertainty modeling.
Firms drawing from frameworks used by:
- Royal Dutch Shell (pioneer in scenario planning)
- World Economic Forum global risk outlook
Your deck frames volatility operationally — not structurally.
6️⃣ ESG Integration into FP&A Models
Noticeably absent.
In 2025, leading organizations integrate:
- Carbon pricing assumptions
- Sustainability capex forecasting
- ESG-linked financing metrics
- Scope 1/2/3 modeling
Even if ESG hype fluctuates, regulatory reporting and investor expectations remain embedded.
Ignoring ESG makes the deck slightly dated.
7️⃣ Performance Measurement in AI-Driven Businesses
A deeper issue:
Traditional KPIs break down in AI-enabled business models.
Questions emerging:
- How do you forecast when marginal cost approaches zero?
- How do you value data assets?
- How do you measure productivity when automation replaces headcount?
This is frontier-level FP&A thinking.
8️⃣ Finance as Orchestrator of Enterprise Intelligence
Your deck says “Strategic Business Partner.”
But the cutting-edge trend is:
Finance becoming the enterprise decision intelligence hub.
Meaning:
- Integrating financial + operational + behavioral data
- Creating predictive strategy dashboards
- Driving capital reallocation continuously
This is more than partnering — it’s system orchestration.
9️⃣ Capital Allocation Discipline in a High-Rate World
You didn’t address macro context.
Post-zero-rate era implications:
- Higher hurdle rates
- Stricter investment screening
- Shorter payback expectations
- Portfolio rationalization
FP&A is regaining importance in capital discipline.
This is a material 2025 theme.
🔟 Measurement of AI ROI Itself
Ironically, the biggest 2025 trend:
FP&A must measure the ROI of AI investments.
Most organizations struggle to quantify:
- Productivity gains
- Revenue uplift
- Risk reduction
- Model maintenance costs
This meta-layer is missing from your deck.
🎯 Strategic Summary
Your deck covers:
- Technology adoption
- Process modernization
- Collaboration
What it misses:
- Structural operating model redesign
- Governance implications
- Macro-economic context
- Talent bifurcation
- ESG integration
- Capital allocation rigor
- Enterprise decision intelligence
- AI ROI measurement
In short:
Your presentation reflects FP&A 2.0 (modernized)
The emerging conversation is FP&A 3.0 (adaptive, AI-governed, capital-intelligent).
So, ask what’s missing, what’s unusual, what’s cutting edge.
Sidebar
A little bit of an unrelated news and not AI related, despite the image. It’s incorrect.
The Epstein files are still going on! It is not dying.
I can’t tell who is really driving the sustained interest to keep the Epstein file front of mind. Some podcast bro hos (Joe Rogan, Andrew ???, Shawn Ryan) appear to be turning on him for not releasing the files as promised and people really want to know which elite have been involved in this sordid network.
But the files are now having an effect worldwide. Andrew Mountbatten Windsor (formerly Prince Andrew) was arrested on Thursday 2/19 on suspicion of committing crimes while trade envoy. Specifically, charges of allegedly transmitting confidential government information to Epstein.
There are emails that showed the transmission. Not a good look.
The last time a senior royal member was arrested was in 1647! Roughly 379 years.
There was another English politician who had to resign. I think some politician or royalty in Sweden and Norway got hit with the backlash.
Epstein knew a lot of elites around the world. Were they all doing sordid stuff? Probably not – except maybe the former prince (not a good look).
The English police went through his multiple residences, probably pulling out his computers and phones to dig through digital details.
Think about that – the English having access to all of his conversations with Epstein. And potentially other so-called members of the network. So, the details we don’t get out of the US files, we might get out of the English files. It is going to depend on the kind of conversations Epstein and Andrew had.
I imagine elites who knew both Epstein and Andrew are worried. So far, it’s a lot of wealthy businessmen, politicians and academicians.
My friend told me last month, right after the Venezuela invasion, that 2026 was going to be a year of crises but I was thinking US crises, according to some astrologists. I was thinking US crises. Curiouser and curiouser.
I don’t strictly believe in astrology but this is curious.
You must be logged in to post a comment.