When an enterprise decides it needs better decisions from its data, the conversation almost always collapses into two options: build it in-house, or buy a big platform and replace what you have. Both are expensive, both are slow, and both are often the wrong choice. There is a third path that rarely gets named.
Why build is harder than it looks
Building in-house sounds appealing — you get exactly what you want, and you own it. In practice, you also own the data engineering, the model maintenance, the integration work, and the retention risk of a small specialist team that could leave. Most internal builds underestimate the unglamorous 80% — the connectors, the data quality rules, the access controls — and overestimate how much the interesting 20% matters without it.
Why buy-and-replace is heavier than it sounds
Buying a large platform and migrating onto it promises a clean slate. What it delivers is a multi-year programme, a painful data migration, months of change management, and a long period during which the new system is live but not yet trusted. For many of the decisions you actually want to improve, you are ripping out a perfectly functional system of record to get at an intelligence layer that could have sat on top of it.
The third option: layer over what you run
Your ERP, LIMS, MES or HR system is usually fine at what it was built for — recording transactions, holding master data, running the process. What it was never designed to do is look forward, score a decision, or hold someone accountable for an outcome. That gap is where a layer lives.
Layering means reading from the systems you already have, adding the decisions they cannot make, and writing back only where that helps. Nothing gets ripped out. The first process can be live in weeks rather than after a discovery phase, because you are not rebuilding the foundation — you are adding a floor on top of it.
When each option is right
Build makes sense when
The decision is a genuine core differentiator, you have a durable specialist team, and no external system understands your domain. This is rarer than most companies believe.
Buy-and-replace makes sense when
Your system of record is genuinely failing — unsupported, unable to scale, or so fragmented that consolidation is the actual goal. Here the migration is the point, not a side effect.
Layer makes sense when
Your systems work but your decisions are late or wrong, you cannot afford a two-year programme, and you want a provable result this quarter. For the majority of "we need AI" conversations, this is the honest answer.
The question to ask first
Before choosing, name the decision you are getting wrong or getting late, and put a cost on it. If fixing that decision requires replacing your system of record, you have a buy-or-build conversation. If it requires reading your existing data and adding intelligence on top, you have a layering conversation — and it is almost always the faster, cheaper, lower-risk path to a number you can defend.