Automating Spreadsheets is Madness

The big consulting firms tell you to install ServiceNow and Salesforce and Workday, but not how to automate your business

Introduction

Usually in this space we talk about healthcare and health insurance, and today is a little different, but we will apply our conclusions to healthcare in the final analysis below.  Today we are going to talk about business processes, or any processes, how they get built, codified and maintained, who does that work, and how AI is going to fit, or not, into all that.    

The Situation

The C-suite is excited. They see ChatGPT, they see "no-code" tools, and they think they've solved the software problem. "We don't need developers, we can just use AI."

Here's what actually happens: they prompt an LLM to generate some code, paste it into a spreadsheet, email it around, and call it "automation."

This is not automation. This is emailing spreadsheets with extra steps.

The average executive has never written a line of code. They don't understand what software actually does — they think it's just "putting instructions in a computer." So when someone tells them AI can write code, they believe it. They're not stupid; they're just outside their expertise. Like asking a cardiologist to evaluate a bridge design.

The problem is they make decisions about software based on what vendors tell them, what consultants recommend, and what "seems logical" to someone who's never had to maintain a production system at 2 AM.

Logical Conclusion

If this continues:

• You'll have 47 "AI-powered" spreadsheets that nobody understands
• Critical business logic lives in someone's Outlook inbox
• When something breaks (and it will), nobody can fix it because nobody wrote it
• You'll pay consultants to "clean up" the mess
• Your competitors who actually know how to build software will eat your lunch
• The business will become completely dependent on whoever remembers how the "automation" works

This isn't a future prediction. This is every company that's ever tried to "do AI" without hiring people who actually understand software.

The Short Answer

You cannot automate what you cannot define, and you cannot define what you do not understand. If your C-suite cannot write a specification, they cannot evaluate a solution, AI-generated or otherwise.

You need developers. Not because AI is bad, but because you need someone who knows what questions to ask, what can go wrong, and how to fix it when it does.

The Slightly Longer Answer

Let's break this down.

What is Automation

True automation is code that runs without human intervention in a production environment. It has tests. It has error handling. It has logs. It can be rolled back if something goes wrong. It can be audited. It can be security-scanned. It has a deployment pipeline.

Emailing a spreadsheet is not automation. It's a game of telephone played with Excel files.

What does AI actually Do

AI generates text that looks like code. It isn’t code.

• It doesn't know your business logic — it guesses
• It doesn't know your data model — it assumes
• It doesn't know your security requirements — it doesn't care
• It doesn't know your compliance obligations — it wasn't trained on your policies

An AI can write a function that calculates a deductible. It cannot tell you whether that calculation complies with your state's insurance regulations, whether it handles the edge cases your actuaries know about, or whether it matches what's in your policy documents.

What Happens when it Breaks

When your AI-generated "automation" fails at 11 PM on a Friday:

• Who debugs it?
• Who finds the root cause?
• Who fixes it?
• Who tests the fix?
• Who deploys the fix?
• Who verifies the fix didn't break something else?

If the answer is "we'll figure it out" — congratulations, you've just discovered why software is hard.

The Consultant Trap

Here's what actually happens in most companies:

1. C-suite sees AI hype (or any hype)
2. They hire a consultant to "implement AI"( or the new hotness)
3. Consultant generates some prompts, builds a prototype in a sandbox
4. Prototype works in demo
5. Production deployment fails because nobody anticipated the real-world complexity
6. Consultant bills for "Phase 2: stabilization"
7. Phase 2 "works" but requires manual intervention
8. Company hires full-time staff to maintain the manual intervention
9. Five years later, you have a $2M annual budget for something that was supposed to "cost nothing"

This is not a failure of AI. This is a failure of understanding what software actually is.

I’m convinced that the big consulting firms have started to copy and paste your name and C-Suite into that same document that they give everyone.  “Install ServiceNow.  Install Salesforce.  Install Workday.”  That isn't helping your business in the slightest and all these programs need their own developers to configure and keep running and infrastructure to manage and run on.

That means you have all the headaches of custom software, with none of the benefits.  You have also paid for out-of-the-box solutions that don’t fit your needs at all.

If they can’t really help you run your business, why in the world would you hire them to “implement AI?”

The Real Cost

The cheap solution is always the expensive solution when you factor in maintenance. AI-generated code that "works" but isn't maintainable costs more to keep running than hiring a developer who writes code that lasts.

The average "no-code" solution requires more code to fix than if you'd just written code in the first place.

The "AI assistant" that helps you write code doesn't know your codebase. It doesn't know your conventions. It doesn't know your team's patterns. It generates something that looks right and breaks in production.

The Skill Gap

The reason your C-suite thinks "anyone can do this" is that they've never done it themselves. They see software as "instructions" — like a recipe. They don't understand:

• State management
• Concurrency
• Race conditions
• Memory leaks
• Security vulnerabilities
• Data consistency
• API design
• Technical debt
• Architectural decisions that can't be unmade

When you ask an AI to "automate your claims processing," it doesn't know that claims have regulatory requirements, that state variations exist, that your actuarial team has specific business rules, or that your legal team has approved certain language.

It just generates text.

The Solution

You need developers. Not because AI is useless, but because AI is a tool that needs someone who understands:

• What the system should do
• What could go wrong
• How to verify correctness
• How to maintain it over time
• How to scale it
• How to secure it

That person is not an AI.  It's a developer.

The executive who thinks "AI replaces developers" is like the executive who thought "email replaces postal mail" or “we’ll make a zillion with a dotcom” or “We have to have a BlockChain stat” or "we will get big insights and control the world with Big Data.”  They are confusing hype with productivity.  Without actually being able to produce their company’s product or service, they can’t possibly steer the company in the correct direction.

AI is a productivity tool for developers. It is not a replacement for developers.

What to do Instead

If you actually want to automate your business:

1. Hire people who can write software
2. Have them evaluate your actual requirements
3. Build real systems with proper architecture
4. Maintain those systems over time
5. Measure results, not "AI usage"

Continue emailing spreadsheets or use AI to automate emailing spreadsheets and wondering why your "digital transformation" keeps failing.

Conclusions

As promised, we will bring this back to healthcare.  We don’t get a lot of spreadsheet emailing in healthcare, but we DO get a lot of linear, workflow thinking.  Emailing spreadsheets is a symptom of having inefficient, outdated processes, just like workflow thinking.

As detailed in “The Theory of Constraints” Eli Goldratt describes several scenarios where a manufacturing plant has assigned one worker to one step in a process.  This is fragile because that worker only does that step.  If anything goes wrong, the downstream workers are idled until that problem is fixed.  Further, if one worker is slower than another, s/he is a bottleneck, and the production is only as fast as the slowest link.

Eli’s solution is to train several workers to each run several machines and eliminate the bottlenecks and mitigate problems.  “But how does that apply to medicine?” I can hear you asking.  Let’s say we have a floor with 20 rooms and five nurses.  This also works for practices as well, but we need an example.  That is one nurse for four rooms.  Sometimes, these rooms are easy cases and sometimes not.  This leads to one nurse sitting around or gossiping or “charting,” while others are busting their hump and missing critical work because it keeps falling through the cracks.  This leads to patients not getting the care they deserve while you are paying people to sit around.

Once more for the cheap seats: Workflow thinking causes patients in the waiting room for hours, patients on gurneys in the hallway and staff doing nothing while “waiting” on whatever it is.  

If we ask Eli how to solve this, he would say, assign those 20 rooms to a queue and automate the tasks to perform in them, both on-demand (I need to use the bathroom) and scheduled (Patients need to be looked in on every two hours).  Then assign the entire staff to the room queue.  The oldest/most urgent task gets assigned to the longest idle nurse.

Apply this to the entire organization.

Now, not only do we have a better use of resources with no bottlenecks, we have, for the first time, a way to judge the efficacy of each nurse with hard  productivity numbers.

“But the patients have to be familiar with the staff!”  No, no they do not and in most cases they couldn't care less and won't remember the name scribbled on the whiteboard next to the TV.

“But the nurses have to be familiar with each case individually!”  No, no they do not.  That is why we have doctors.  The doctor will have their own queue to document efficacy and outcomes and not let tasks go uncompleted.  A patient can even be assigned to several doctors, but for the most part the nurses shouldn't be giving care that isn‘t either routine or specifically ordered by the doctor.

Now that we have put the fallacy of workflow to the torch, we can implement the queuing system, and keep everyone busy, it will be too soon if I ever hear ‘workflow’ again.  Workflow is wrong and stop using it.

The sharpest among you will notice that we haven’t mentioned AI in the solution/conclusions.

This process management system is the fourth plank in Sentia’s four plank platform called “Saving Healthcare.”  You can read a high level overview of all four planks (data driven EMR, Integrated coverage inside the MER, integrated health and wellness to mitigate chronic disease, and of course the ERP style practice/hospital management system, complete with queueing system)  here.  You can see a demo of this system here.  

We have shown a way to regulate every financial transaction a practice or a hospital enters into, the Practice/Hospital Management System (PHM).  Included with the PHM is a workflow elimination tool that extracts more and better work from employees and streamlines and automates every facet of patient care.  All that increases revenue and decreases costs.

We have shown a way to incentivize healthy living in a population and decrease chronic disease and therefore decrease costs for us all in a streamlined and automated manner.  This alone has the potential to save $1.34 trillion or about 25% of healthcare spending in the US

We have shown a way to revolutionize the way medical records are thought of, executed, used and searched.  This eliminates Epic, all the legacy EMR vendors and makes research a simple pick and click operation, saving millions of lives.

We have shown a way to integrate health coverage into the EMR.  The practice or hospital gets paid as the practitioner documents patient care.  That eliminates medical coding, verification, adjudication, pre-authorization, denials, delays, insurance networks, rate negotiations, sales/brokers/agents, money for a third-party EMR, skyscrapers in every major city, hundreds of thousands of employees, all the insurance monkey business and reduces cost by about half. 

It also eliminates Epic/Cerner AND the legacy insurers.

It also makes your facility leaner faster, more efficient and more profitable.

This system includes the automation of the health insurance industry completely, eliminating more than half the costs with Sentia as the coverage company, employer based captive or TPA or by direct payments to doctors and practices.

Here are additional points detailing the costs incurred by the legacy insurance companies that you pay currently, in addition to wasting about half your premium, according to Grand View Research and current as of 2023 and that Sentia would eliminate completely:

Medical Records:

• The average practitioner spends $35,925 annually on electronic medical records
• The average patient spends $106 annually on electronic medical records
• The average patient encounter or visit cost for electronic medical records alone is $32

Medical Coding:

• The average practitioner spends $20,286 annually on medical coding
• The average patient spends $60 annually on medical coding
• The average patient encounter or visit cost for medical coding alone is $18

Compliance and Efficacy Reporting:

• The average practitioner spends $17,165 annually on compliance and efficacy reporting
• The average patient spends $51 annually on compliance and efficacy reporting
• The average patient encounter or visit cost for compliance and efficacy reporting alone is $15

Totals:

• The average practitioner spends $73,376 annually on completely avoidable costs
• The average patient spends $217 annually on completely avoidable costs
• The average patient encounter or visit cost for completely avoidable costs alone is $66

Yes, you read that correctly: $66 per visit. That is probably more than the practice makes on the average encounter.  There must be a better way. There is a better way and Sentia has it.

Remember also that these costs are over and above the 50%+ your insurance company wastes or shoves into their pockets.

Implementing this system should be fairly simple and will completely revolutionize the way healthcare is delivered and paid for, saving countless lives. We have shown a way to use this system to make the best healthcare system in the world also the most efficacious and the most affordable.

If you liked what you read contact us here, on our site, SentiaHealth.com, our parent company SentiaSystems.com, or send us an email to info@sentiasystems.com or info@sentiahealth.com.