AI should be the last layer you add
Thomas Van Duüren
Tuesday 27 January 2026
Artificial intelligence has become a commodity at an unprecedented pace.
Models are powerful, tooling is cheap and experimentation (which we encourage of course) feels almost frictionless.
As a result, many organisations start their digital roadmap with the same question: “What can we do with AI?”
We believe this is the wrong starting point.
Not because AI isn’t valuable but because AI is a multiplier, not a fix.
When you multiply a weak foundation, you don’t get intelligence but accelerated chaos!
AI is a multiplier
A simple mental model explains why so many AI initiatives disappoint:
AI impact = Model Quality × Data Quality × Process ClarityIf any of these elements is weak, the entire outcome collapses.
-
A strong model trained on unreliable data produces confident nonsense
-
Automation layered on unclear processes scales confusion
-
AI applied to disputed numbers erodes trust faster than any manual error ever could
AI does not hide structural issues. It exposes them brutally and immediately.
What a solid core looks like
A solid core means three very concrete things:
A single operational source of truth
Your organisation must agree on where reality lives. If sales, finance, operations and management each work from different numbers, AI will only accelerate the disagreement. For most companies, this is an ERP system. Whether that is Odoo or another platform, provided it is:
correctly implemented
broadly adopted
aligned with how the business actually operates
Data you trust, not just data you store
A data warehouse is not a dumping ground for exports and logs. If people still ask “which report is correct?”, AI has no stable ground to stand on. Before AI becomes meaningful, data must be:
structured
reconciled across systems
historically consistent
explainable to humans
Clear ownership of processes and decisions
AI forces uncomfortable questions. If responsibility is unclear, AI doesn't solve the problem but automates it. Keep in mind that AI works best when humans remain accountable and systems are designed to support that accountability.
Who owns this data?
Who corrects errors?
Who approves decisions?
Where AI belongs in the architecture
AI never replaces the core but consumes it.
ERP systems remain deterministic and auditable.
AI operates above them:
-
interpreting
-
predicting
-
recommending
-
orchestrating
But never becoming the system of record.
The most dangerous anti-pattern: “We’ll clean the data later”
Few statements are more costly than: “Let’s add AI now and fix the data afterwards.”
In practice, this leads to:
-
models learning inconsistencies
-
decisions that can’t be explained
-
outputs that can’t be trusted
-
systems that are hard to roll back
Once AI is embedded into daily decision-making, fixing the foundation becomes exponentially harder.
A practical readiness check
Before investing heavily in AI, ask these questions honestly:
-
Can sales and finance reconcile without discussion?
-
Do reports change depending on who exports them?
-
Is Excel still the integration layer between systems?
-
Is data ownership clearly defined?
-
Do teams trust the numbers they work with?
If the answer is “no” more than once, the priority is clear: fix the core first.
Book a meeting
Discuss your idea directly with one of our experts. Pick a slot that fits your schedule!
Keep reading
Why most AI demos don’t scale
Every AI demo works perfectly. It has a smooth interface, fast responses, seamless and predictable outcomes. But what happens after the demo?
Thomas Van Duüren
The developer who hits 'Merge' is still responsible, even in the AI era
At Endare, we fully embrace AI-assisted coding but we also believe that accountability, context and human judgment are more important than ever.
Thomas Van Duüren
What if the AI agent makes a mistake? Designing for safe failure
No matter how well an AI agent is trained or configured, it will occasionally get things wrong. The solution isn’t to stop automation altogether. The solution is to design for safe failure.
Thomas Van Duüren
