By Hiran de Silva
This morning, Paul Barnhurst made an interesting post about AI and financial modelling.
The broad sentiment was something along the lines of:
AI is brilliant…
but humans are still better.
Now before going any further, let me make clear that I am not disagreeing with the importance of human thinking. Quite the opposite.
But as I often do in discussions like this, I asked a very simple question:
Better at what?
Because one of the recurring problems in discussions around Excel, AI, financial modelling, automation, Power Query, dashboards, analytics, and enterprise systems…
is that people frequently use the same words to mean completely different things.
“Financial model.”
“Automation.”
“AI.”
“Excel solution.”
“Enterprise.”
These terms sound obvious until you ask ten people to define them.
And that matters.
Because unless the terminology is anchored to real requirements and real use cases, conversations become vague very quickly.
While writing my response to Paul’s post, I realised something important.
For months now, I have already been running what are effectively AI benchmark projects.
Not theoretical ones.
Real-world enterprise benchmark models.
The reconciliation challenge.
The REG Call Handler.
The Budget Review Benchmark.
The Beatles Jam model.
The collaborative drilldown review models.
The GET/PUT architecture demonstrations.
The global consolidation models.
All of them are really exploring the same underlying question:
How can AI help solve real enterprise problems?
Not social media demo problems.
Not “write me a formula” problems.
Not “generate some VBA” problems.
But actual collaborative business problems involving:
- multiple users
- continuous processes
- governance
- auditability
- scalability
- live data
- operational review
- drilldown analysis
- workflow integration
- process connectivity
- distributed responsibility
Because these are the problems organisations deal with every month.
And here is where something fascinating happens.
The internet constantly tells us:
- AI is revolutionary
- Excel is powerful
- automation is the future
- Power Query is modern
- Copilot changes everything
And yet…
the moment you present ordinary enterprise requirements…
many people suddenly respond with:
“Well… Excel isn’t really the right tool for that.”
That is where the discussion becomes extremely interesting.
Because my response is:
Excel already solved it.
Not theoretically.
Not hypothetically.
Not as an experimental concept.
But in real implementations.
Used by real organisations.
At scale.
Collaboratively.
Under pressure.
With deadlines.
With management oversight.
With governance requirements.
With real operational consequences.
And that creates a very uncomfortable tension.
Because if Excel already solved these problems…
why do so many intelligent people still believe it cannot?
I believe the answer sits between three things.
Technology.
AI.
And business reality.
Imagine three circles.
The first is technology.
User-accessible technology.
Excel.
Access.
SQL.
Desktop tools.
Shared databases.
Connected systems.
The second circle is AI.
ChatGPT.
Claude.
Copilot.
OpenAI.
Generative AI systems.
The third circle is the business challenge itself.
Reconciliation.
Budget review.
Consolidation.
Collaborative reporting.
Continuous operational review.
Enterprise workflow management.
Now here is the strange thing.
All three circles are powerful.
All three circles already exist.
All three circles are immensely valuable.
And yet…
the solution still often fails to appear.
Why?
Because there is a missing link.
Architectural thinking.
That is the thing sitting in the middle.
Most AI demonstrations today still assume the existing workflow is fundamentally correct.
That is the hidden trap.
So AI becomes:
- faster formulas
- faster Power Query
- faster spreadsheet construction
- faster transformations
- faster manual processes
But still trapped inside the same underlying mental model.
Excel as document.
Instead of Excel as connected architecture.
And according to my own experience over the last thirty years, that distinction changes everything.
Because the moment you stop thinking of Excel as an isolated workbook…
and start thinking of it as a client sitting on top of connected business processes…
the entire problem changes.
The reconciliation challenge changes.
The budgeting challenge changes.
The collaboration challenge changes.
The scalability challenge changes.
The governance challenge changes.
Because now you are not merely designing a spreadsheet.
You are designing a process.
This is precisely why I have been creating these benchmark models publicly.
Not to “win arguments.”
But to expose the underlying assumptions people are bringing into the discussion.
The benchmark models are intentionally public and intentionally broad.
They cover use cases relevant to:
- finance leaders
- industry leaders
- operational management
- ERP vendors
- the Excel Replacement Industry
- trainers
- students
- academics
- consultants
- citizen developers
- social media Excel educators
Why?
Because these are not niche problems.
These are common enterprise realities.
And the benchmark itself is simple.
Here is the requirement.
Here are the constraints.
Here are the non-negotiable business conditions.
Now:
How do you solve it?
How can AI help solve it?
Can conventional Excel thinking solve it?
Can Power Query solve it?
Can AI solve it while still treating Excel as a document?
Or does the entire approach fundamentally change once Excel is understood as part of a connected client-server architecture?
That is the real question underneath all of this.
And this is where the discussion becomes far more important than merely “AI versus humans.”
Because I do not believe the missing ingredient is simply coding skill.
Or prompt engineering.
Or formula knowledge.
The missing ingredient is architectural intelligence.
The ability to understand:
- where data should live
- how processes connect
- how collaboration operates
- how governance functions
- how review systems scale
- how workflows integrate
- how users interact with live information
- how operational reality differs from social media demonstrations
AI can accelerate implementation enormously.
I know this because I have used AI to help implement many of these benchmark demonstrations myself.
But AI alone does not automatically produce architectural thinking.
And that, I believe, is why many people are currently struggling to achieve transformational enterprise outcomes with AI despite the technology itself being extraordinary.
Because they are still approaching the problem with the wrong foundational assumption.
Excel as document.
Instead of Excel as infrastructure.
And perhaps that is the real miracle of Excel.
Not the spreadsheet itself.
But the fact that ordinary users were quietly handed access to enterprise-grade architectural capability decades ago…
without most of the world ever fully realising it.
Add comment