Phil Simon's Blog

Equal-Opportunity Hallucinator

Large language models can't distinguish fact from fiction. That's hardly news. Mildly intelligent college students know as much. They understand the risks of using AI chatbots willy-nilly to crank out term papers. Yet, the fact that ChatGPT and its ilk routinely hallucinate somehow evaded senior leadership at a top-tier consulting firm.

In today's post, I'll explain why. But first, here's a quick primer on how the consulting industry works.

Consulting 101: Profit Pressures and Turf Wars

Boutique consulting shops with a few dozen employees pick a lane and stick to it; internationally recognized advisory ones with thousands of workers do not. Rather, they offer different arrays of services and solutions because their clients are so different. Boeing requires different consulting services than Chase and Banner Health do. One size does not fit all.

Take a job at Accenture, BCG, KPMG, McKinsey, PwC, and other big-ticket professional-service firms. You'll quickly discover that they tend to organize themselves into segments—or verticals, to use their parlance. Common examples include healthcare, retail, financial services, manufacturing, higher education, and the public sector.

These verticals operate as quasi-autonomous businesses within their massive parent companies. By themselves, they would be enormous enterprises. Exhibit A: Deloitte's public-sector vertical generated more than $12 billion last year—about one-sixth of the company's total revenues.

Impressive, but that amount is never enough. Pressure to increase each vertical's bottom line is incessant, especially at publicly traded firms. Can someone say stock options?

Regardless of whether John Q. Public can buy its stock, firm mandates to each vertical are clear:

A Case Study in No-Code Failure

👤I've changed the names in this story.

YYZ is a successful mid-sized VC firm. A few years back, it conducted an extensive search for a new managing partner. Among the candidates who interviewed for the position, Adam clearly stood out from his peers. Yeah, he knew the industry, but no one else asked about YYZ's existing systems and tech. In short, they were a mess. Disparate Excel spreadsheets and a torrent of emails ruled the day.

Before accepting the position, Adam stressed the need to make significant tech changes after joining. First up: the firm needed to turn chaos into order—and that required adopting new tech. The popular magnet app Notion was squarely in his crosshairs.

Photo by Isaac Smith / UnsplashBackground

In his previous position, Adam had used the popular no-code tool. He was an unapologetic fanboy. Notion allowed everyone at his previous employer to easily retrieve key information, including:

The status of each deal.The ROI of each investment.A summary of weekly, monthly, quarterly, and annual performance. Total funds invested by industry. Which employees were closing which deals—and which were the most successful.

People spent minimal time reconciling errors. Accurate reporting was a breeze. Things just worked. Why shouldn't YYZ function the same way?

A New Tool Begets the Same Problems

Adam spent the first few months creating a bespoke Notion system. Not long after launching it, though, things at YYZ began rapidly devolving. Information in Notion sometimes conflicted with the company's bank statements, accounting system, internal emails, and signed, legally binding documents. Late-night data investigations were becoming commonplace.

In one particularly galling example, YYZ nearly remitted the same six-figure Series-A payment to a portfolio company twice. Six weeks later, a promising biotech startup opted to go with another VC firm because it was more organized.

Beyond these alarming one-off incidents, other system-related problems plagued the company. Although it used a top-tier tool, answers to simple business questions proved elusive. Here are a few:

Where do we stand with each deal?How much have we invested in each company and industry by stage? What is the status of each term sheet we've sent out? Who was the last person to talk to the potential portfolio company about a potential investment?

Adam wasn't the only one peeved. The other YYZ partners were none too pleased at its current state of affairs.

Of course, it didn't need to be this way. Let's dive into what happened.

Root Cause Analysis

Notion was supposed to function as a single source of truth at YYZ. Instead, it had morphed into a morass of incomplete, outdated, missing, or duplicate information. Reports from it had become mere guides—starting points to find the truth. Some were downright meaningless. GIGO at its finest.

You may infer that YYZ's plight stemmed solely from its tech. Rest assured, it did not. As is usually the case, humans are to blame.

Duplicate Records and Conflicting Information

A galling lack of internal controls allowed anyone at YYZ to enter just about anything into Notion. Let's start with duplicate entries. Many employees who couldn't locate an existing record simply created a new one.

Duplicate Portfolio Company in YYZ Notion Database: | Click on the image to enlarge it.

You needn't be an expert in data modeling or master data management to appreciate the problems that this practice causes.

And then there were the issues related to an individual record's existing property—Notionspeak for a database field or column, if you like. Some employees created new, overlapping, and unnecessary properties that told different stories. Take deal status, for instance. NotionAI could help identify YYZ's discrepancies among three conflicting values, but it could not resolve them.

In a few cases, people took a DIY approach and created their own, separate Notion databases to track their deals. One rogue worker—a promising new hire—ignored Notion entirely and opted to use a Google Sheet.

So why was data quality at YYZ an utter mess? We need to keep asking why.

New Notion Template: Venture Capital & Investment TrackerA simple way to track investments, funding rounds, valuations, and much more.Phil SimonPhil SimonHuman Factors, Not Tech Ones

In a way, Notion was destined to fail at YYZ—as was any new tool. The company failed to properly train its employees on its new system and related processes. What's more, Adam and his team failed to lock down Notion databases to prevent the very issues they'd soon encounter. To some extent, everyone was an admin.1

Many people blew off training sessions because they were too busy or had noodled with Notion on their own time.

GIGO at its finest

Adding salt to the wound, communication at YYZ was generally deficient. The company would make small but critical changes to its Notion system but fail to tell its workforce about them.

On Amoeba and NotionLow-code and no-code tools don’t fall neatly into a single category.Phil SimonPhil SimonSimon Says: New Tools Require Planning

The moral of this little yarn: Notion, Airtable, Coda, Loop, and other apps have enabled a new, fundamentally different type of tool builder. (It's a key point in my book Low-Code/No-Code). Citizen developers lack formal coding knowledge, but they have nonetheless created innovative solutions that have solved real problems. In many cases, their IT departments have thanked them for taking months of work off their plate.

In a way, no-code/low-code tools are just like their antecedents: Without sufficient database design, product support, change management, accountability, and training, they are just as likely to cause problems as they are to solve them.

Remember that truism the next time a software vendor's wares look too good to be true. The same holds true if you think you can build the plane as you're flying it.

Footnotes To be fair, Notion's lack of record-level permissions at the time exacerbated the problem.↩ need help building and deploying A NEW NOTION SYSTEM at your organization?

Let me help your organization avoid the mistakes that YYZ and countless others make as they adopt Notion.

HIRE ME

On Employee Expenses, Lazy Managers, and AI Agents

Starting in 2000, I worked at Lawson Software as an application consultant. The job typically involved descending on different organizations and setting up their new HR and payroll systems. As you'd expect, we consultants would need to book our own flights, hotels, rental cars, and the like. Even when flying coach, our weekly expenses could easily reach $2,500 or more.

Unfortunately, Lawson didn't provide us with company credit cards.1 As a result, my colleagues and I had to pay for our expenses out of our own pockets. We would then submit them to our managers for approval. Upon successful submission, Lawson's internal system sent daily short emails reminidng them to either:

Approve their pending expenses; orReject them with an explanation.

In theory, consultant reimbursements would arrive in their bank accounts soon after that. In other words, the float would be ephemeral.

Photo by NASA Hubble Space Telescope / UnsplashTheory and Practice Collide

Sadly, some employee loans weren't so fleeting. The cause for the delays wasn't technical. As is often the case, it was human.

My first manager (call her Denise here) routinely ignored these emails. A few weeks after I had submitted about $5,000 in expenses, I followed up with an email and a voicemail only to hear crickets.

The AI Slop Paradox

Odds are that you've heard the term AI slop in the last few months. Wikipedia defines it as follows:

a term for low-quality media made with generative artificial intelligence.

If you've been living in a bubble, here are some recent stories on the topic:

Spotify's new content policies won't stop its proliferation.AI worksplop will destroy productivity.It may portend the death of the creator economy.OpenAI's Sora 2 app just dropped. It lets anyone create godawful videos, and it's the #1 iOS app as I write these words.

It takes all of about four minutes to find AI-generated drivel in the form of art, articles, and songs. Even before Sora 2 arrived, people used AI to create insipid videos.

Concerns over AI slop may ultimately turn out to be legitimate, alarmist, or something else. Put me in the first camp. Like Rust Cohle, consider me a pessimist.

AI Slop Catches Up to Its Human Equivalent

Let's not forget, though, that old-fashioned humans continue to generate a never-ending supply of slop. Don't believe me? Allow me to present Exhibit A:

AI Book Settlement Tracker

My latest Notion template just dropped. It lets authors track claim status and settlement payments for all AI lawsuits.

Here's the preview video:

The cost is $9.94. Here's a little more about it:

As with any Notion template, you can customize it to meet your needs—and I'm not just talking about your specific titles and new lawsuits. Add new charts, automations, templates, properties, settlement override amounts, and database views to your heart's content. Invite co-authors or agents to collaborate on it if you like.

Read the whole post on the Racket Publishing site for more information on it.

If you subscribe to this site and you'd like to buy it, then head over to the Benefits page to view your coupon code.

BUY THE TEMPLATE

Maslow Revisited

At some point, any college student studying business comes across Maslow's Hierarchy of Needs. TL;DR: Not all worker desires are created equal. Employees will never become self-actualized unless management pays them on time. You know, that sort of thing.

A recent experience at the gym reminded me that, when it comes to customers, today's decision-makers ought to follow Maslow's timeless advice.

Forgetting the Basics

By way of background, my health club last year introduced rudimentary AI functionality into its app. I prompted it and it responded as follows:

The oddly named L•AI•C is far from perfect, but it answers many obvious fitness-related questions in English. In fact, I do enjoy GTX classes.

My high-end offers this shiny new thing. Great, but is it a trailblazer? Hardly, as I've said many times, AI is everywhere today.

Forgetting the Basics

And yet...

Vibe Coding With Opal

A few months ago, Google dropped yet another new AI tool. This one—Opal—lets users “build, edit, and share mini-AI apps using natural language.” You can start from scratch, or you can begin by selecting one of its pre-built apps.

In this post, I describe my results noodling with the beta tool. I'll also explain what this type of experimentation portends for the future of work.

First Experiments

I picked one of the gallery's pre-built apps. Video Marketer promised to create “AI video ads with your product and target audience.”

Sample From Opal Gallery | Source: Google

I punched in some basic information for my book The Nine: The Tectonic Forces Reshaping the Workplace. Opal asked for the product's primary audience. I quickly entered my reader personas. Here's Opal's bizarre eight-second creation:

Video of a Woman Reading a Hallucinated Book Called The Gine | Source: Google Opal

The Year's Laziest Marketing Email

The year 1998 saw the launch of a very different type of search engine. Apart from its funky name, the new website approached the problem from an entirely different perspective. Within months, it blew up and left AskJeeves, AltaVista, AlltheWeb, and their ilk in the dust.

Google grew rapidly and organically from its inception. (The Oxford English Dictionary in 2006 added google as a verb.) In its early days, the company spent exactly zero dollars on marketing. Co-founders Larry Page and Sergey Brin were software engineers who dismissed marketing as a legitimate business function. If you built a kick-ass product, they reasoned, then you didn't need to tell people about it.

Larry Page and Sergey Brin | Source: Wired

John Batelle's 2005 book The Search: How Google and Its Rivals Rewrote the Rules of Business and Transformed Our Culture is well worth reading, even today.

🤔 A Fun Counterfactual
What would have happened if Yahoo! had ponied up a mere $1 million—yes, million —for Google in 1998? It's not like the company lacked the cheddar. Its market cap was roughly $30 billion back then.

In 1999, Larry and Sergey relented and dubiously hired someone to spearhead Google's nascent marketing efforts: Douglas Edwards. Eventually, he wrote a book about his experience. I'm Feeling Lucky: The Confessions of Google Employee Number 59 is fascinating.

Adios, Algorithms. Hello, Email. Remember Me?

Over my career, I have created more websites, reports, dashboards, and databases than I can count. And then there are systems—lots and lots of systems. Some my clients demanded them; others I built strictly for myself. (RacketHub is a labor of love.) During my foray into higher ed, I taught courses in the W. P. Carey Department of Information Systems.

I'm not complaining. I enjoy creating systems because I think in terms of structured data. Why they implode is the subject of my first book. Hell, my previous website was the embarrassingly long philsimonsystems.com until I grabbed its current, snappier domain name. You get the idea.

A few years ago, I built a bespoke one that retrieved content and let me consume it on my terms. It was decent but imperfect. In this post, I'll explain why and how I built the old system, its limitations, and why I ditched it for a decidedly low-tech successor. Feel free to steal borrow it or modify it as needed.

Photo by Markus Spiske / UnsplashBackground

I prefer doing things without distractions, especially reading. Not surprisingly, the powers that be don't make it easy to focus. None of this was news to me, but Adam Alter's exceptional 2017 book on the subject revealed the extent of the problem. Tech companies intentionally design their apps, websites, and devices to distract us from what we're doing. You needn't be a neurologist to know that living in a state of continuous partial attention isn't ideal.

The Flip Side of Forced System Migrations

I spent the majority of the aughts helping organizations move from legacy enterprise systems to more contemporary ones. I even penned a book about the subject.

That experience taught me a great deal, but one truism sits at the top of the list. Individuals, teams, departments, and organizations change systems for two reasons:

Because they want to.Because they have to.

In most cases, group motivations fell into the second bucket. Management would have loved to have stayed with its current systems, but—for whatever reason—stasis wasn't an option.

The usual suspects included:

Vendors sunsetting their existing systems.Cost and security concerns.A new CXO comes down with a severe case of shiny object syndrome.An acquisition or merger.

I didn't keep a formal tally, but I suspect that, three-quarters of the time, employees moving to new systems wanted to exactly replicate their existing business processes to a T—even with new, more powerful tools. If I only had a nickel for every time someone resisted new tech and uttered the sentence, “But that's not how we do it now.”

Change Management