What to Demand From an ERP Data Warehouse

Ask: “When the contract ends, what do I keep, and where does it
live today?”

There are two models in this market. In one, the warehouse lives in the
vendor’s schema, on the vendor’s stack, behind the vendor’s subscription – your
data is a tenant in their building. In the other, the store is provisioned
inside your own boundary and stays with you. The difference
only shows up at renewal, which is exactly when it’s too late to discover it.
Demand the second model.

Ask: “If I leave, do I keep every version of every record, or
just a final snapshot?”

Half your real questions are time questions – what did this PO look like at
cutover, when did these two systems start disagreeing, prove the migration
moved everything. Those need retained version history, not a current-state
extract. A warehouse that only keeps “what’s true now” quietly throws away the
asset that compounds. Demand exit rights that include the history.

Ask: “Show me the proof from your last delivery.”

Most vendors sell visibility – another chart, another dashboard. Very few
sell attested correctness: numbers that tie out to the cent,
a month closed behind a checklist that was green when they said it was, an
evidence packet a change-control reviewer can actually file. A CFO can live
without one more dashboard. They cannot sign numbers nobody stands behind. If a
vendor can’t produce the evidence from real work, you’re buying a promise.

Ask: “What validates a report before it reaches me, and who
declares it production-ready?”

“We tested it” is not a control. Ask what is checked mechanically (does the
output parse, does it differ from baseline in exactly the intended way, did the
build touch anything it shouldn’t have) and what a human signs off. The honest
answer names both – and never lets a tool self-certify. Vague reassurance here
is the single most expensive thing to discover after go-live.

Ask: “When my ERP and my second system disagree, which one does
your warehouse believe – and how would I know?”

If you run Acumatica or one of its vertical editions alongside a shop-floor, CRM, e-commerce,
or field-service system, the expensive gap is the layer between them,
where no vendor sells the reconciled middle. Pulling both systems into one
warehouse is collection. Telling you this record, this field, this value
here, that value there, with a tolerance, is reconciliation. Demand the
second – with the divergence report as a deliverable.

Ask: “If I point an AI at this, does it answer with citations –
and does it refuse when it doesn’t know?”

Every buyer is now asking whether their data is ready for AI. A warehouse
built for last decade’s question – prettier dashboards – is not the same as one
a model can query and return a cited, deterministic answer from, abstaining
rather than guessing. Governance, provenance, and refusal-loud behavior are the
difference between AI that’s useful and AI that confidently invents a number.

Ask: “Price me the install, the steady state, and the exit –
all three.”

Per-seat BI subscriptions and template-license math look cheap on page one
and compound every year you can’t leave. A fixed-fee install plus a defined
monthly stewardship motion – feeds monitored, months closed with evidence,
drift reported – is a number you can plan around. And the exit cost should be
near zero, because in the model worth buying, you already own the store.