The analysis that produces the first findings is a data join and a set of charts. The purchase decision should come after it, not before.
What you already own
Warehouse management system reporting, which is usually configured for operational control and can be extended.
Telematics portals supplied with the fleet, frequently unused.
Environmental sensors, cheap and standalone.
A spreadsheet or a small database, which is entirely adequate for the first year's analysis in a single facility.
Deploy what you have and see what remains, which produces a much narrower requirement.
What a platform adds
Ingestion and joining across systems, maintained, which is the tedious part.
Handling of volume that a spreadsheet cannot.
Multi-site comparison and consolidation.
Real-time alerting, which a batch analysis cannot give.
Retention and access controls, which matter for the individual data stream.
Maintained connectors to systems you would otherwise integrate yourself.
The ingestion, the connectors and the access controls are the real purchase. The charts are not.
When building is defensible
A single site with a manageable data volume.
Existing data engineering capacity with a named owner for three years.
An unusual requirement no product meets.
Data that cannot leave the site, which is a real constraint in some sectors.
A short question: where does work wait. That is a query and a chart, not a platform.
When it is not
When nobody will own it. An unmaintained pipeline degrades quietly and produces confidently wrong numbers, which is worse than none.
Multiple sites, where the consolidation and comparison work is substantial.
When real-time alerting is the requirement.
When the individual data stream needs proper access control, which is harder to build correctly than it looks.
When the driver is avoiding a licence cost smaller than the engineering time.
Evaluating a platform
Trial on your own data, including the messy parts. A demonstration on clean sample data proves nothing about your scan discipline.
Ask what it does with data quality problems — silently exclude, flag, or produce wrong answers.
Ask about the identity model: can it enforce aggregation, or does it hold individual records by default with a reporting filter on top. This determines whether your policy is architecture or convention.
Ask what happens to your data on exit, and test the export during the trial.
Log your own hours during the trial, which is the cost never in the business case.
What to weight lightly
The dashboard, which is where demonstrations spend their time.
Benchmarking against other facilities, built on data of unknown comparability.
Savings estimates generated by the tool.
Connector counts, where you need six of them working properly.
Predictive features, which require more history and cleaner data than most facilities have.
The sequence
Join, clean, analyse, act, identify the specific gaps, then buy for those gaps.
A requirement written from evidence is narrower, cheaper and more likely to be met than one written from a vendor's capability list.
Trialling on messy data
A demonstration on clean sample data proves nothing about your facility.
Give the trial your real extract, including the batch scans, the open tasks and the stale locations.
Ask what it does with them: silently exclude, flag, or produce a wrong answer.
Silent exclusion is the dangerous behaviour, because the output looks complete.
Check whether the exclusion rules are visible and editable.
Ask how it reports data quality, which most products do not, and which determines how much you can trust anything it produces.
The three-year question
Asked before building, and it decides the answer.
Who maintains this when the author leaves? A named role with allocated time.
What happens when the management system is upgraded? Which it will be.
Who notices when an extract silently starts returning partial data? This is the failure that produces confidently wrong numbers.
What is the documented handover?
If any of these has no answer, buy, because a decayed internal pipeline is worse than none — people rely on it.