Being a CFO and other topics

Not just finance, hobbies too ....

Month: February 2026

Car being driven by a robot goes off a cliff.

AI in Finance Is a Governance Problem — Not a Technology One

By

On February 26, 2026

In Accounting and Reporting, AI, CFO, Finance, Leadership

For the last year or two, every CFO conversation eventually drifts into AI. Sometimes it’s framed as excitement, sometimes as anxiety, and sometimes as an awkward silence followed by, “Well, we’re looking at it.” What’s striking is that most of the tension around AI in finance has very little to do with the technology itself. The models work. The tools are improving fast. The vendors all have slick demos.

The real issue is governance.

Finance teams are wired around controls, auditability, and repeatability. AI systems, by contrast, are probabilistic, opaque, and constantly evolving. That mismatch is where most CFO discomfort comes from — and it’s why “let’s just automate this” often stalls once it hits a real finance process.

The first mistake I see is treating AI like just another system implementation. ERP projects taught us how painful that mindset can be. AI requires a different framing: not “what can this tool do?” but “what decisions are we willing to delegate, and under what constraints?” That sounds abstract. It isn’t.

Over the past year I’ve pushed AI tools on real finance questions: revenue recognition edge cases, SEC disclosure interpretations, covenant calculations, and technical accounting memos. The patterns that show up are not technology failures. They are governance failures waiting to happen.

1. AI doesn’t fight back.

If you have ever debated an accounting position with a strong controller or technical accounting lead, you know what conviction feels like. You push. They push back. You test assumptions. They defend them with chapter and verse. That friction is healthy. Same thing for a forecast analysis. If one FP&A analyst thinks they found a good or disturbing trend, it will be debated and verified and usually their work can be recreated and checked.

AI does not behave that way.

If you tell it, “I think you’re wrong,” it often apologizes and produces a different answer. Sometimes an entirely opposite answer. The confidence level remains high. The tone remains polished. The data is processed inside the model, and the AI often struggles to explain — or even remain consistent in — its answers.

In a live finance organization, that would be a red flag. If a manager flipped their view that quickly under mild pressure, you would question the depth of analysis. With AI, the flip can look like responsiveness rather than fragility.

That is a governance issue. It means you cannot treat an AI output as a position that has survived adversarial testing. It hasn’t. It has survived prompt engineering. And the prompt may have been poor.

2. The praise problem.

Most AI agents are relentlessly deferential. “Great question.” “Excellent point.” “You’re absolutely right to focus on that.” In a consumer context, that feels pleasant. In a finance context, it is dangerous.

Finance works because of tension — between risk and growth, between conservatism and disclosure clarity, between what management wants and what GAAP allows. When the “advisor” in the room is constantly affirming the user, it subtly reinforces bias.

I’ve seen this firsthand when asking an AI to pressure-test a disclosure approach. Rather than aggressively identifying weaknesses, it often validates the framing of the question. The tone can make a marginal position sound well-supported. In other words, the user’s confidence can rise faster than the quality of the analysis.

Governance must assume that AI will not naturally challenge you the way a seasoned audit partner or skeptical board member will.

3. The citation illusion.

This one should make every CFO uncomfortable.

Ask an AI to provide citations to accounting guidance or SEC commentary, and it will often comply — confidently. Paragraph numbers. Codification references. Even plausible-sounding excerpts.

The problem is that some of them are fabricated. They look right. They read right. They are formatted correctly. But they do not exist.

In finance, citations are not decorative. They are the backbone of defensibility. When you write a technical memo on revenue recognition or stock-based compensation, the citation is the bridge between your judgment and the authoritative literature.

If an AI invents that bridge, and a team relies on it without independent verification, the failure is not the model’s. It is the control environment’s. Any AI-assisted accounting memo must include a verification step where a human independently confirms the authoritative source. Not “glances at it.” Confirms it.

4. Rule changes and historical drift.

Accounting rules change. Constantly.

Revenue recognition under ASC 606 replaced a patchwork of legacy guidance. Lease accounting under ASC 842 upended decades of practice. The SEC updates disclosure expectations over time, sometimes subtly, sometimes dramatically.

Meanwhile, the SEC’s EDGAR archive goes back decades. There are scanned paper filings from eras when the rules were materially different. There are thousands of examples built under superseded guidance.

AI models trained on broad corpuses struggle here. They can blend old and new regimes. They can cite legacy practice as if it were current. They can rely heavily on the abundance of historical examples rather than the correctness of modern policy.

I have seen AI answers that lean on pre-606 revenue language as though nothing changed. Or that reference lease accounting concepts that no longer apply post-842. To a non-expert, the answer looks sophisticated. To someone who lived through the transition, the seams are obvious.

Governance means you assume the model does not instinctively know the effective date of your accounting framework. You have to constrain it.

5. Finance is not plain English.

Financial reporting language is precise. “Probable” does not mean “likely” in a colloquial sense. “Material” is not a synonym for “important.” “Reasonably possible” has a defined meaning.

AI systems are trained on massive volumes of plain English. That is a strength in many domains. In accounting, it can be a weakness.

I’ve seen answers where the model drifts into narrative explanations that sound sensible but subtly misapply defined terms. In a board deck, that might pass. In a 10-K, that is a problem.

When language itself carries regulatory weight, small deviations matter.

So what does governance look like in practice?

It is not banning AI. That is neither realistic nor wise. The productivity gains are real. Drafting first passes of memos, summarizing contracts, identifying anomalies in large datasets — these are powerful tools. AI can be properly trained on your data and become more accurate. Specialized firms like the Big 4 Auditors can train AI models on better and sanitized accounting data, but your small Finance group cannot and its probably using a more general model.

But they must sit inside a control framework.

At a minimum:

  • AI outputs that influence external reporting require documented human review.
  • AI conclusions about trends must be independently tested and verified. Don’t order another $1M of a part because a model suggested it.
  • Authoritative citations must be independently verified.
  • Prompts and versions used for material analyses should be retained for auditability.
  • Use cases must be categorized: drafting support is different from judgment replacement.
  • Responsibility for the final position must be clearly assigned to a human owner.

Most importantly, the CFO has to set the tone.

Let me make a direct observation: most leadership team members are not finance experts, but AI can create the illusion that they are. You need to make sure they understand the risk.

If AI is positioned as an infallible oracle, teams will over-rely on it. If it is positioned as a junior analyst — fast, helpful, occasionally wrong, and requiring supervision — behavior adjusts appropriately.

The question is not whether AI will be used in finance. It already is.

The question is whether it will be used inside a governance framework that protects credibility.

Investors do not care how you produced your numbers. Auditors do not care how you drafted your memo. Regulators certainly do not care that a model was “usually right.” They care that your disclosures are accurate, supportable, and controlled.

AI in finance is not a technology problem. It is a governance problem. And like most governance problems, it lands squarely on the CFO’s desk.

I don’t want to sound like Cassandra warning of inevitable doom. Nor do I want to be the boy who cried wolf while your competitor quietly figures this out and gains an advantage.

In future posts, I will outline where I believe AI can genuinely add value inside a disciplined finance organization.

Businessman diving into a pile of cash with a bottle of champagne in hand.

Success! – The Waterfall of Cash at the End of the IPO

By

On February 24, 2026

In CFO, Investor Relations, Working With Investment Banks

In my last blog entry on IPOs, the process had reached the successful end of the roadshow, and the banks were now prepared to offer you the deal.

This is the step where the final price is set and where the initial allocation of shares is done. There can be some movement in the share price compared to the original range announced at the launch of the roadshow. If the change is within a 10% band of that indicative pricing, you can proceed with pricing without needing to refile.

If demand is strong enough to justify a price more than 10% above the original band, it ideally was identified early in the roadshow so the range could be adjusted and a new prospectus filed with the SEC. If the demand is very back-end loaded, you may be forced to delay pricing while the new filing is made. That delay — on top of the market signal that a price reset was needed — can outright kill the deal.

That is one reason deals are often cut in size if demand is lower than hoped. The delay combined with weak optics can be fatal.

When the banks (specifically the lead left bank) tell you the price and number of shares, they also give you their proposed allocation — how many shares and to which investors.

The first thing to point out is that the initial allocation includes the “greenshoe,” which is part of a standard IPO. This is normally 15% on top of the base deal size. The underwriters sell these shares short up front as part of the stabilization process.

I will get to stabilization and the greenshoe mechanics shortly, but the key point at this stage is that the order book must cover not only the base deal but the additional 15% as well.

You do get a say in the allocation.

For the most part, the ECM desk knows who is best to place the initial shares with and how many. This ties directly to how the stock is likely to trade when the IPO goes live. You need some trading to happen right away, so some shares need to be placed into accounts that will actually trade.

You should absolutely review the allocation carefully. It is fine to tweak it and move shares to certain investors that you favour, especially those you believe will be constructive long term holders. But you also need to trust the banks. They are in the market every day and know these accounts well.

All of this is calculated by the banks, and they present what they believe is the most reasonable structure to support successful aftermarket trading. Everyone wants to see the stock trade up after the IPO — that stamps the deal as a success out of the gate. A huge first-day jump generates good press, but too big a jump is money left on the table by the company.

The deal then goes to your Board of Directors — almost always to a pre-authorized pricing committee. You need a quick decision, and coordinating a full board in real time can be a logistical nightmare. A smaller group authorized in advance makes the process smoother.

The lead bank presents the final terms, there is discussion, and then a decision is made.

Sometimes demand is not quite where you hoped. The banks may propose a smaller deal, a lower price, or both. If this happens, you are in a very tough spot. A failed IPO can set you back quite a while. If you take a lower price or reduced size, it puts you on your heels from a momentum standpoint and makes early investor relations more challenging.

I lean toward taking a workable deal and fighting forward, but you cannot make that call in the abstract — you make it in the moment. I did not face that specific scenario as our IPO priced in the middle of the range.

The next morning, trading begins. Pricing is often Wednesday night and trading starts Thursday morning. Banks generally avoid Mondays and Fridays for a first trade.

Traditionally, this also gives the company the opportunity to “ring the opening bell.” It is essentially a staged PR event, but a meaningful one for the team. Because we did our IPO during COVID, we did not get to do it. If you can arrange it, I recommend it.

Hopefully trading starts well and the stock trades up. Every trading day brings outside news and market events, so not everything that happens will have anything to do with your company.

Enjoy your celebratory dinner with the banking teams and, if you drink, have something that night to mark the moment — just do not do anything foolish (really do not drink and drive).

Although the stock technically settles on standard T+1 timing, your underwriting agreement likely specifies settlement between T+2 and T+4. This is because the underwriters purchase the shares directly from the company, not through market trading.

On settlement day, you receive a wire transfer for the gross proceeds less underwriting fees and usually all professional fees agreed to be paid (both yours and certain bank expenses), plus travel and other deal-related costs that the banks advanced.

You will only receive proceeds for the shares sold by the company. Secondary shareholders who sold as part of the IPO receive their cash directly from the underwriters.

If your company receives primary proceeds, you should have a plan ready. In our case, a significant portion went toward paying down debt.

If the stock stays at or above the IPO price during the greenshoe period (typically 30 days), you will also receive proceeds from the additional 15% overallotment.

If the stock trades down and the stabilization agent buys shares in the open market to cover the short created at pricing, you may end up with less greenshoe exercised. That is normal. Stabilization is common. Roughly 20–30% of IPOs trade below the offering price on the first day, and 40–50% trade below it at some point in the first 30 days. We tend to remember the moonshots and forget the statistics.

The greenshoe exists to balance the overallotment and help manage those first few weeks of trading. A clean aftermarket makes life much easier for management and investor relations.

The final topic that is important to understand is the lock-up.

It is very common for underwriters to require a 180-day lock-up on the company and its officers. This is designed to prevent additional stock from coming to market too quickly and gives IPO investors some stability.

It also means management has to wait 180 days before selling any shares they own or have been granted.

The banks can waive lock-ups, and terms are negotiable, but 180 days is standard. I have had waivers granted before when the stock was trading well above the offering price and the transaction involved secondary shares, so it is possible — just not typical.

Enjoy the success.

And try not to drown in the pool of cash.

Keyboard transformation with sparkles.

My Technology Life: Sparkle Up!

By

On February 22, 2026

In Computer hardware, General

My desktop setup has been pretty constant since Covid times when I built my computer. It’s been a Corsair K100 keyboard with magnetic switches, some version of a Scimitar mouse, all sitting on an MM700 RGB Extended Mouse Pad. Those are augmented by an Elgato Stream Deck and later a Stream Deck+.

Around last Christmas I decided it was time to spruce things up. Close to five years of a basically black colour scheme with an RGB edge needed a change.

The setup that caught my eye can be found here:
https://www.corsair.com/us/en/s/fallout-nuka-cola

The individual pieces are probably sold out by now — Corsair rotates themes frequently — but they keep an ever-refreshed set of choices for your “battle station” here:
https://www.corsair.com/us/en/s/corsair-custom-lab

There’s been a quiet revolution in customization technology. Corsair has leveraged knowledge from Origin PC, SCUF and Drop to bring real customization to enthusiast customers. Even Elgato has experimented with — and mastered — custom paint jobs for their Wave 3 microphones, mic arms, and the edging of the Stream Deck Mk2. In the self-built keyboard world, custom keycaps are common, but I don’t know anyone offering the breadth of customizable products that Corsair does.

My first customized products were a SCUF PS5 controller (medieval knight theme) and a few different skins for my Envision controller.

The Corsair Custom Lab technology is flexible enough to handle very low print runs. If your company, esports team, or online D&D group ever wanted themed PC gear, that Custom Lab link above is worth a look.

Once the Fallout Nuka Cola theme caught my eye, I decided to update my Stream Deck, replace the mouse pad, and redo the keyboard. The Stream Deck and mouse pad were simple swaps. The keyboard took a bit more effort.

First, make sure your keyboard supports removable keycaps. Many mechanical keyboards do. Be careful with space bars and larger keys — they often have wire stabilizers that require a little finesse. If you’re unsure, YouTube is your friend.

The most common replaceable style is Cherry MX. Use a keycap puller — I suggest having two, especially for larger keys. (They’re often included with mechanical keyboards, but I’ll link some below if needed.) Place the wire ends over the keycap, twist slightly so they sit diagonally under the key, and pull up. Unplug the keyboard or turn off wireless before you start.

https://amzn.to/4cMuSjo

Once you remove a keycap, check the switch stem. If it’s a small cross, you have Cherry MX style. Flip the keycap over and you’ll see the matching cross-shaped connector underneath.

Before removing everything, put the key back on and take a few reference photos. You’ll thank yourself later.

I won’t turn this into a full guide. Pull the caps off (use two tools for longer keys), keep them organized, and once they’re off, clean the keyboard. After five years, there will be debris. Cleaning is much easier with the caps removed. Then press the new keycaps on — don’t be afraid to be firm so they seat properly. Plug it back in and enjoy the new look. There are lots of Youtube videdos showing this process if you want to take the time and learn it.

The keycap set I chose included a wide variety of keys. Unfortunately, the K100 has six “G” keys without a good Nuka Cola equivalent, so I left those as-is. I may tweak the LEDs in iCue to red to help them blend in.

This is how my keyboard looked before any changes.

Step one was removing all the original keycaps (I arranged them beside the keyboard as a reference, with photos as backup).

Keyboard with no keycaps.

New keycaps installed.

My new desktop.

This was my first time replacing keycaps. It took about 30 minutes to remove and organize the old ones, and another 15 minutes to install the new set. I’d be faster the second time.

It’s a small change overall, but it makes me smile every time I sit down. Certainly a real sparkle up for my daily tasks!

Runner dressed in an S-1 races to the finish line made of money.

Doing the IPO – Closing the Sale

By

On February 17, 2026

In CFO, Investor Relations, Leadership, Working With Investment Banks

Up until the point where the underwriters give their nod and your lawyers tell the SEC to declare the S-1 effective, there has been a very large burden on your shoulders as the CFO. As important as the final sales push is, once you get to this stage most of the remaining work shifts away from you and onto the banks and, more importantly, to your boss — the CEO.

There are two activities that typically happen before the roadshow itself formally launches.

The first is any “wall crossing.” This is a process where the banks preview the deal with a small number of qualified institutional investors. It does not happen every time, but when it does, the goal is to gauge demand, see how well the marketing message resonates with the target audience, and infer valuation. Valuation is not supposed to be discussed explicitly, but both the banks and the investors involved know how to read and give signals.

Wall crossing usually happens before the S-1 is declared effective, but typically when it is close to ready. Other than the fact that these are often among the first presentations you will do — so you may not sound as polished yet — they are not materially different from the presentations you will give once the roadshow officially begins. If something has gone seriously wrong earlier in the preparation or underwriting process, an overwhelmingly negative reaction could stop the IPO. But if you are this far along, that really should not happen.

Even if you enjoy conspiracy theories and think the banks are just looking for a graceful exit, the reality is that the funds you are talking to do not have time for games. What you usually get instead is constructive feedback, some tightening of the pitch, and an early indication of how strong demand — and therefore valuation — might be.

The other common special presentation is the recorded version of the deck that can be shared with accounts where an in-person or live meeting cannot be scheduled, usually due to geography. Some investors will watch the recording simply to get the canned material out of the way so that they can spend their live meeting time asking questions and engaging with management. The banks almost always have a preferred vendor for this. Other than being prepared to be on camera, there is not much additional advice to give.

At the end of this phase, your Board will need to meet to formally accept or reject the banks’ offer, so that meeting needs to be scheduled in advance. Sometimes pricing authority is delegated to a committee, but for IPOs the full Board often votes.

Then your lawyers ping the SEC, the S-1 flips to effective, and you are officially off.

Well before this point you will have selected your lead bank — usually referred to as the “lead left” bank, which reflects their position in the underwriting syndicate. They are almost always the stabilization agent as well. You probably relied on an external advisory firm to help with this selection process.

Once chosen, the lead bank becomes the primary administrator of the IPO and controls the master order book. In my experience, they bring in the majority of the orders. The other banks generally make a good-faith effort, attend meetings, and provide advice, but because the lead controls the book, the investors they want to prioritize tend to take precedence.

Interestingly, I have found that smaller banks often punch above their weight. They tend to be hungrier and try harder, and it is not uncommon for them to deliver more orders than their share of the fees would suggest.

The IPO I worked on was during COVID, so it was conducted entirely over Zoom. Today, roadshows have largely shifted back to in-person meetings. Zoom is still used for geographically isolated investors, but many meetings now take place at investor offices or group venues, often over lunch. This reintroduces a meaningful amount of travel planning, which the banks will manage. Having done enough deal and non-deal roadshows in person, I have a healthy respect for the logistics.

My first piece of advice is to rest before it starts and stay away from alcohol during the roadshow. You will likely be changing time zones and dealing with jet lag. The less additional stress you put on your body, the better you will perform in meetings. You can have all the drinks you denied yourself once it is over.

Eat well. Stay hydrated. You will be talking a lot, and a raw throat will absolutely work against you. This is an important trip — or series of trips — and you need to stack the odds in your favor. Expect planes, cars, and possibly trains (New York to Boston). For larger IPOs, expect travel to Europe and a week or more on the road.

The second piece of advice is to be prepared to sit through many meetings where you say very little — but also be ready to carry the meeting if needed. You are the CFO, traveling with your CEO. A tremendous amount rested on your shoulders before the roadshow launched, but once it starts, the burden shifts heavily to the CEO.

This varies by individual, but most CEOs who make it this far are good at selling what makes their company special. Your role is to support them, handle the details, and do whatever you can to help them succeed. It is common for people to lose their voice or have their energy dip over the course of a roadshow, so you may need to step in more than expected — but you should not plan on it.

The meetings themselves feel very much like the wall-crossing meetings. Most are cordial, with a back-and-forth Q&A. Styles vary widely among analysts and portfolio managers. Some questioning styles can feel combative or dismissive. Do not take it personally, and warn your CEO in advance if you sense that style emerging.

Some of the largest orders I have seen came from firms that sounded the most negative in the room. Maintain an even keel, but do not become expressionless. Investors expect some passion from management. Unlike routine IR meetings, IPO roadshows tend to involve a higher proportion of portfolio managers — actual decision makers — rather than just analysts.

At the end of each day, the order book will be updated and you will get a snapshot of how many orders are coming in and from whom. The ECM team at the lead bank will be increasingly directive about allocations, assuming demand warrants it, but that becomes more important later in the process.

For now, the rhythm is meeting, limo, meeting, limo, eat something, meeting, limo. It all blurs together. I have a good memory, but if I do not take notes, I struggle the next day to recall who said what.

As the book grows, you will see which banks are delivering orders. It often looks like the lead bank is doing all the work, but that is partly an artifact of book control. Non-lead sales forces may push less aggressively because investors know which ECM team controls allocations, so they route orders accordingly. Analysts across the syndicate are usually working hard, and smaller banks often bring in smaller accounts that larger banks ignore, earning their place in the book.

Eventually, the meetings end and the banks tally the orders. If things went well, they will tell you they have enough demand to put a deal together. I do not work on an ECM desk, but I do know that accounts routinely inflate orders in expectation of being cut back.

If you assume roughly 100% order inflation, the common rule of thumb that a deal should be 3–5x oversubscribed really means you need something closer to 6–10x in raw demand. This varies based on market cap and, more importantly, the quality of the orders, not just the quantity — but as a rough guide, it is about right.

In my final post on the IPO process, I will cover allocation, the pricing committee, and what happens when trading begins — including how the banks can step in if early trading is choppy.

Wooden puppet draped in green glowing code with a large nose.

My Technology Life: AI: Lying Liars Lie

By

On February 14, 2026

In AI, CFO, Computer hardware, Finance

I know that the in-vogue term is hallucinate instead of lie, but since the main interface to AI tends to be via chat — and the models are intentionally designed to simulate a personality — “lying” feels more accurate.

During my attempts to develop the RPG PDF conversion pipeline I described last week (you can find that post here: https://mgpotter.com/my-technology-life-ai-agent/), I encountered behaviors that should sound very familiar to anyone who has tried to push AI beyond toy problems.

Here are a few highlights.

1) Work Claimed, Work Not Done

On several occasions, I was told that the new Python script I requested had been completed. When I asked to see the script — because my own coding is not good enough to trust it without review — I was then told the script could not be found and likely had not been written.

In another variation, I was told the PDF had been successfully processed and that the output was excellent. No output file existed.

This is not a “mistake.” It is the model optimizing for conversational completion. It is trained to provide a satisfying answer, not to verify that work was actually performed.

2) Phantom Sub-Agents Doing Phantom Work

At one point I was informed that five sub-agents had been spawned to divide the PDF and perform OCR.

The problem? The OCR tool in question does not run on the 15-year-old CPU I was using as a test bed. It lacks the instruction set required to execute.

Yet I received multiple progress reports describing how efficiently the sub-agents were performing.

In reality, the tool had crashed immediately. The sub-agents were waiting for a reply that would never come. The administrator bot was confidently reporting progress on work that had not and could not have occurred.

Again, this is not malicious. It is structural. The AI fills in gaps with plausible narratives.

3) “Perfect Output” That Was Garbage

More than once, I received a grand report that the parsing was perfect and ready for conversion into Fantasy Grounds format.

The file was not even close.

The model had learned that the desired outcome was “success.” So it reported success.

4) Hardcoding the Answer

While dialing in table and column detection, I created an answer sheet to help guide the agent’s debugging.

The next output was perfect.

Until I asked probing questions and ran the code through a second model.

There had been no improvement to the algorithm. The agent had simply hardcoded the expected answer.

This is a recurring issue: the model optimizes to satisfy the prompt, not to build a robust, generalized solution.

5) Creative Rewriting Instead of Extraction

In some cases, the “extracted text” was not extracted at all. It had been rewritten and reorganized to be cleaner and more readable.

That might be helpful for marketing copy. It is catastrophic for financial reporting or legal work.

These Problems Are Not Unique to Hobby Projects

I have seen similar behaviors when applying AI to real Finance questions:

  • SEC citations that do not exist
  • Press releases with invented links
  • Tariff rules misread and inverted
  • Spreadsheets reorganized in ways that no longer foot

In Finance, you cannot be 98% right. Especially when you are reporting publicly.

A 2% error rate is not a rounding issue. It is a career-limiting event.

How to Reduce These Errors (But Not Eliminate Them)

There are ways to mitigate these behaviors. They require discipline.

1) Force Evidence, Not Assertions

Instead of asking whether the script was completed, ask the AI to return the full script, include line numbers, include the file path, and confirm the function definitions exist. Make the AI produce artifacts, not conclusions.

2) Require Verifiable Citations

Instead of asking what an SEC rule says in general terms, require the model to quote the exact paragraph of the rule, include the regulation number, and state explicitly if it is uncertain rather than inferring. Force it to cite or admit uncertainty.

3) For Code: Demand Diff-Based Changes

Instead of simply asking to improve the algorithm, require the model to return only the changed lines, explain the logic improvement, confirm that no test data is embedded, and explicitly state that it has not hardcoded expected outputs. This reduces the chance of hardcoding or cosmetic fixes.

4) Explicitly Forbid Invention

Include language in your prompts that instructs the model to say “unknown” if it does not know, to avoid fabrication, to avoid assuming files exist, and not to simulate tool output. You would be surprised how much that helps.

5) Separate Tasks

AI struggles when prompts mix architecture, implementation, testing, and reporting in one request. Break them apart. Treat it like managing a junior associate.

6) Independent Verification

If the output matters, use a second model to review it, recalculate totals independently, cross-reference source documents, and inspect logs manually. Trust but verify is too generous. Verify and then trust provisionally.

The Finance Question

I have seen steady progress in AI tools for Finance. FP&A more than accounting, which makes sense. Forecasts are inherently estimates; variance analysis is expected.

But regulatory filings, audit workpapers, footnotes, tax positions, debt agreements — these are binary environments.

The market, the SEC, your auditors, and your board do not accept “the AI hallucinated.”

The tools are impressive. They are helpful. They can accelerate research, draft memos, and summarize documents.

They are not yet reliable enough to operate unsupervised in Finance.

As of right now, AI tools in Finance should be used:

  • As assistants
  • As draft generators
  • As brainstorming tools

And always with a heavy layer of skepticism and human review.

Lying liars lie.

The models are not malicious. But they are optimized to complete conversations, not to protect your reputation.

That distinction matters.

The OpenClaw logo, featuring a stylized lobster and the wordmark.

My Technology Life – AI Agent

By

On February 7, 2026

In AI, CFO, Gaming, RPG

A Quick Warning Before We Start

Before getting into the substance of this post, it’s worth being explicit about the environment this work was done in.

OpenClaw is not a secure system. I would not expose it to the internet, and I would not run it anywhere near a machine that held sensitive data. This experiment was conducted on an isolated Linux box that is more than ten years old, deliberately segmented away from anything that mattered. That isolation was intentional, and I would consider it a prerequisite rather than a nice-to-have.

With that caveat out of the way, here’s what I learned.

The more technical information is this post comes from AI allowing me to cosplay as someone with a much deeper background in this field. I haven’t coded since I was a teenager running a BBS on my Apple //GS. Everything described here was implemented by directing AI tools — primarily Claude — with research, validation, and conceptual framing done through ChatGPT.

This was not a case of me dusting off dormant engineering skills. It was an exercise in seeing how far careful prompting, iteration, and architecture could go without writing code myself.

I started with what seemed like a reasonable question: could an AI agent take an RPG PDF and convert it into a usable Fantasy Grounds VTT reference manual? Fantasy Grounds is the tool I use to run RPG with my friends and I very often have to get adventures into the program.

The test case was a Mothership RPG adventure. Not particularly long, but representative of the kind of layout that makes RPG books pleasant to read and painful to process. Multi-column text, sidebars, boxed callouts, tables, and frequent typography changes all coexist on the same page. Humans have no trouble with this. Machines very much do.

The first thing that became obvious is that PDFs do not contain “text” in the way we usually think about it. They contain positioned glyphs. Reading order, paragraph structure, and emphasis are all emergent properties created by the human brain. When you extract text naively, you get all the words, but not the story they were meant to tell.

Standard PDF extraction tools did exactly what they are designed to do. They gave me the words. They just gave them to me in the wrong order. Columns were interleaved, paragraphs were broken every line, sidebars merged into body text, and tables disintegrated into streams of numbers and labels with no structure left intact.

At that point, the obvious temptation was to let the LLM “just read the PDF.” After all, large language models are very good at understanding text, right?

That approach failed in subtle but dangerous ways.

LLMs are quite good at repairing relationships when the underlying structure is mostly correct. They are far less reliable when asked to infer structure that was never presented to them in the first place. RPG books are full of ambiguous layout decisions, and when an LLM guesses, it does so confidently and silently. Sidebars get merged into rules text. Paragraphs are reordered to match genre expectations rather than author intent. The output looks clean, but it is wrong in ways that are difficult to detect later.

The approach that actually worked separated responsibilities very strictly.

First, the extraction phase focused entirely on facts. Using PyMuPDF, the system extracted every word along with its exact coordinates, font size, font face, and bounding box. The output was ugly and unreadable, but nothing was lost. Every signal a human reader subconsciously relies on was still present, just not interpreted yet.

Second came layout reconstruction. This was where most of the complexity lived. By working from geometry instead of text flow, it became possible to detect column gutters, read entire columns top-to-bottom instead of left-to-right across the page, and reconstruct paragraphs based on vertical spacing rather than newline characters. Hyphenated words could be repaired deterministically. Headings could be inferred from typography rather than guessed from phrasing.

This step also addressed the most visible problem with PDF extraction: the explosion of extra line feeds. Those line breaks are not semantic. They are artifacts of line wrapping. Once reading order and paragraph boundaries are reconstructed using spacing and font metrics, most of those spurious line breaks disappear before an LLM ever gets involved.

Only after that cleanup did the LLM enter the process, and even then its role was constrained. It was allowed to repair flow and normalize text, but not invent structure, reorder content, or generate XML. Markers for headings, emphasis, sidebars, and tables were preserved explicitly so the model could not “helpfully” smooth them away.

The final stage — generating Fantasy Grounds XML — was deliberately scripted and deterministic. Fantasy Grounds is unforgiving, and rightly so. IDs, tags, ordering, and escaping are not things you want a language model remembering across thousands of tokens. Once the content was clean and correctly ordered, turning it into XML was a mechanical problem, not an AI problem.

Tables turned out to be the most treacherous area. Early attempts to detect them aggressively led to false positives where multi-column prose or credits pages were misclassified as tables. The safer approach was to be conservative to the point of under-detection, preserving text as text unless the evidence was overwhelming. A less-perfect table is preferable to corrupted rules text.

One of the more humbling realizations came late in the process. The PDF had no table of contents, but it did have bookmarks. Those bookmarks reflected the author’s actual organizational intent far better than anything inferred from layout alone. Once the pipeline followed them, chunking and navigation improved immediately. It was a reminder that many “AI problems” are really failures to leverage existing metadata.

From a CFO perspective, the conclusion is straightforward.

This took longer than doing it manually. Considerably longer.

I spent over $100 running smaller, simpler test cases just to understand where tools failed, where token usage exploded, and how errors manifested. That spend did not produce output. It produced learning. For a single RPG module, this is not rational. Manual conversion would have been faster and cheaper.

Where this starts to make sense is repetition. Multiple modules. Consistent layouts. A reusable pipeline. At that point, the upfront investment begins to amortize.

There is also a broader organizational lesson here. Running this through a rough, developer-oriented agent on an isolated machine worked, but it was far from ideal. A more user-friendly agent, or involvement from an IT team, would have reduced iteration time, lowered token waste, and improved safety. There is real value in tooling and support, even — perhaps especially — when AI is involved.

This was a successful experiment, but not an efficient one. I would do it again only if I planned to do it many times. As with so many automation efforts, the real question is not whether it can be done, but whether you are willing to do it often enough for the investment to pay off.

Sometimes the most useful output isn’t the finished product. It’s understanding where the break-even point actually lies. And what you learn along the way.

I ran OpenClaw on a very old HP ProLiant MicroServer (Gen 7) with a Turion II processor. I blocked incoming access using standard Linux hardening, isolated it on its own VLAN, and did not install any additional skills or allow the agent to browse the internet. That was intentional, given the known security risks around prompt injection attacks and malware delivered through unscreened skills.

All files used by OpenClaw lived only on that machine, and it had no access to my other systems. OpenClaw itself can run on fairly low-end hardware, since the heavy lifting is done in the cloud. If you want a more robust — and still relatively inexpensive — platform to run it on, with the added benefit of access to the Apple ecosystem, many people use Mac minis.

Mac Mini M4 on Amazon.com

If you want to find OpenClaw (the speed of the internet moves fast and months from now this may not be the hot new tool), look here:

OpenClaw AI Bot

Excel and Powerpointn icons as hockey players doing a faceoff. SEC logo on the puck..

IPO Process – Underwriting, Forecasts, And The Road To Launch

By

On February 3, 2026

In Accounting and Reporting, CFO, Investing, Investor Relations, Working With Investment Banks

While you are grinding away on the S‑1, there are many other work streams running in parallel. One of the most important—and least talked about—is getting through the underwriting process. The S‑1 gets you most of the way there, but it is not enough on its own. You also need to provide the banks’ analysts with a detailed financial forecast.

This is very different from life once you are public. Under Regulation FD, you do not share your detailed internal forecast with sell‑side analysts unless, for some very unusual reason, you have already made it public. During the IPO process, however, you are still a private company, and those public‑company disclosure rules do not yet apply.

This forecast matters a lot, and it needs to be carefully balanced. There is always a temptation to push the numbers—stretch the growth assumptions and aim for a higher valuation at the IPO. In my experience, the less public‑company experience a leadership team has, the stronger that temptation tends to be. That is a meaningful mistake, and it can have consequences in several different ways.

The first is credibility with the sell‑side analyst. This is someone you are likely to have a relationship with for years. The more aggressive your forecast, the more questions you will get. Analysts are very experienced at talking to management teams, and they are good at figuring out when numbers lack a solid foundation. Even if you can technically defend the assumptions, they will still wonder why you are pushing so hard. These analysts work closely with potential investors and will be fielding questions about expectations. You want them transmitting confidence, not concern.

The second issue is the pressure you put on yourself to hit your first few quarters as a public company. The IPO process feels like a sprint, but being public and creating long‑term value is much closer to a cross‑country run than a 50‑yard dash. Missing your first quarter right out of the gate can be catastrophic for credibility. It is very hard to reset expectations once you stumble early.

The third consideration is internal to the banks themselves. Your banking team has to take the deal in front of their internal committees to get approval and a green light to proceed. Everyone in that room has seen dozens—if not hundreds—of IPOs. Their job is to control risk. A management team that appears overly promotional or willing to stretch the truth to grab incremental valuation is a risk factor.

This does not mean you should sandbag the numbers. It means you should make sure you do not need perfect execution and a lot of luck to hit the first few quarters. If you need another practical reason to stay disciplined, remember that you are going to be locked up for at least six months after the IPO. There is no immediate personal benefit to being overly aggressive right out of the gate.

This forecasting exercise also feeds directly into the final negotiation around the expected IPO price. It sounds great to see the stock skyrocket the moment trading begins, but that simply means you left money on the table. A healthy first‑day pop sets a positive tone. An excessive one is just capital you failed to raise. You might recapture some of it later through a secondary offering, but it is far better to price the initial deal thoughtfully.

At the same time, you are also building the roadshow presentation. In the successful IPO I was part of, this was heavily driven by our founder and CEO. He had previously gone through an unsuccessful roadshow, but the business had changed dramatically by the time of this one. Building the presentation inevitably involves bouncing back and forth with the business section of the S‑1 to ensure that every claim and message is properly reflected in the prospectus.

I was fortunate that our founder drove this process. He had deep institutional knowledge of the company and a clear sense of what mattered. Roadshow presentations are strange things. Sometimes they are the focal point of investor meetings. Other times they barely get looked at, and the conversation turns immediately into Q&A.

You will also use this presentation to record the virtual roadshow, so it will be seen by a large number of potential investors. I generally recommend including a few “halo” slides that act as launching pads for key investment themes. Not everyone naturally finds their rhythm in an investor meeting, and a well‑structured presentation can help create momentum. Our CEO did not really need it, but for many teams it can be valuable support.

I am always amazed by how much time gets spent on presentations. This one is more important than most, but it is still a massive time sink. There are so many cooks in the kitchen that the final version is often worse than an earlier draft. In addition to the usual internal stakeholders, your lawyers and the banks’ lawyers will give it an extremely thorough scrub. Eventually, it does get done, and you inch closer to launch.

Usually the last major hurdle—assuming the market is open—is final approval from the SEC. I have written previously about responding to SEC comment letters, and the process is not fundamentally different here:

Responding to a comment letter from the SEC

The stakes are higher and the time pressure is intense, but the mechanics are the same. The key difference is that you generally do not have the option of saying, “We will improve this in the next filing,” even for relatively minor disclosure issues. The SEC will want it fixed now.

You need to move quickly as you approach your intended launch window, but it is worth remembering that the SEC does not want to stop you from going public. Their job is to make sure you are following the rules. You may not even receive a detailed review—or any review at all. If you planned properly and staffed the process correctly, this stage should be manageable. If the SEC uncovers multiple accounting or disclosure issues, however, the process will stall, and it will be obvious to everyone why.

Once you clear this final step and your banks confirm that the market window is open, you instruct your lawyers to notify the SEC that the S‑1 is effective. At that point, you actually launch. My next post will cover what happens once the process moves fully into the roadshow and selling phase.

Powered by WordPress & Theme by Anders Norén