CPH/May 1, 2026
01 — The Thesis

Two kinds of software now sit inside every modern company.
One records what is true. One reasons about what to do.

The mistake is treating them as competitors. The opportunity: use each for what it does best, and draw a clean line between them.

DETERMINISTIC

System of Record

Microsoft Dynamics & the legacy stack

Logic
Rules a human wrote, in advance.
Input
Structured fields — SKU, quantity, price, date.
Output
Same input → same output, every time.
Question
“What is true right now?”
Strength
Audit. Compliance. Money movement.
Failure
Silent — it just does not fire.
To add a case
IT ticket. Six weeks.

“The customer’s credit limit is $50k because that's what we typed in last year.”

NON-DETERMINISTIC

System of Judgment

The AI-native layer that surrounds it

Logic
Patterns inferred from data & context.
Input
Anything — email, PDF, voice, image, free text.
Output
Probably right, with a confidence and a citation.
Question
“What does this mean, and what should we do?”
Strength
Judgment. Ambiguity. Language. Exceptions.
Failure
Loud — it tells you when it is unsure.
To add a case
An example. An afternoon.

“The customer's risk profile rose this week on local-currency volatility — suggest a $35k limit.”

Dynamics forces the world into a rigid spreadsheet.
The AI-native layer thrives in the chaos of global chemical trading.
CPH needs both.

Comparison : Deterministic vs. AI-Native at CPH

Business Challenge
Deterministic (Legacy)
AI-Native (Non-deterministic)
1. Port Congestion (Far East)
Wait for a manual email from the shipping agent saying the ship is late.
Predicts delay 2 weeks in advance based on satellite weather and port AIS data.
2. “Open Book” Reporting
Send a PDF or Excel sheet every Friday. Producer has to find the insights.
Producer chats with a live portal : “Why did my sales drop in Germany this week?”
3. Trade Finance Risk
If buyer misses one payment, stop shipping (too late).
Flags credit risk before a payment is missed, based on the buyer's industry news signals.
4. Inventory (20+ Hubs)
Re-order when stock hits X units (static safety stock).
Dynamically adjusts stock levels daily, based on local demand probability curves.
5. Chemical Compliance
Hire a technical clerk to manually read REACH / SDS docs for errors.
AI autonomously scans thousands of pages to flag non-compliant hazards instantly.
6. Last-Mile Logistics
Route trucks based on fixed postal-code zones.
Optimises routes in real time on live traffic, weather, and warehouse pick-speeds.
7. Lead Generation
Sales reps search Google / LinkedIn manually for new factories.
AI scans news & industrial permits to find new factories before they open.
Why this matters

Two scaling curves.
Only one of them is affordable.

The deepest reason to go AI-native isn't productivity. It's the shape of the curve when CPH's order book doubles.

DETERMINISTIC SCALING

2× orders  →  ≈1.8× headcount

Manual entry, phone calls, reconciliations — all linear. Every extra deal is another email to translate, PDF to retype, margin to recompute by hand.

Growth becomes a cost centre. The bigger CPH gets, the slower it moves.

NON-DETERMINISTIC SCALING

2× orders  →  ↑ API spend

The unstructured mess — emails, delays, risk shifts, foreign documents — is absorbed by the AI layer. Humans focus on the high-stakes, high-judgment moments. One more deal ≈ a few more API calls.

Growth becomes a margin. The bigger CPH gets, the faster it compounds.

Stylised. The point is the divergence.
See the three layers  →