TL;DR
- Evaluate on your own data and your own cluster by reproducing a plan your team already accepted, never on the vendor's demo project.
- The three long-term cost drivers are model runtime, import and export fidelity, and licence terms for occasional users.
- Ask what happens to your project data if you stop paying. The answer is the main long-term lock-in mechanism in this software category.
What should you test in a trial?
Import your own data, on your own cluster, and reproduce a plan your team has already accepted. That single test answers more than any feature list. If the tool cannot reproduce last year's accepted plan, either the tool or your data is wrong, and you need to know which before you sign.
Then test the three things that decide long-term cost: model runtime on a realistic cluster, import and export fidelity for the formats you actually exchange with partners, and the workflow for a routine parameter change. A tool that takes six hours to run a prediction will not be used for iteration, no matter how accurate it is. A tool that loses antenna tilt on export will corrupt every partner exchange.
| Evaluation item | How to test it | Pass criterion |
|---|---|---|
| Model runtime | Predict a 150-site cluster at the resolution you will actually use | Fast enough that an engineer iterates within a working day |
| Data import fidelity | Import your site database and count discrepancies field by field | Zero silent corrections; every change logged |
| Export fidelity | Export and re-import; compare parameter tables | Round-trip identical, including antenna model references |
| Automation | Run AFP or ACP on a real cluster with your constraints | Produces a reviewable result, not just a finished run |
| Reporting | Produce an acceptance report in your customer's format | No manual post-editing in a spreadsheet |
| Licence flexibility | Ask for the price of five occasional users | Occasional access is affordable, not a full seat each |
Which capabilities are usually over-specified?
Technology breadth is the classic one. A tool supporting every standard from GSM to the newest release is valuable to a vendor's marketing and largely irrelevant to an operator planning two technologies. You pay for the breadth in licence cost and in interface complexity that your team will never fully learn.
Highest-resolution ray tracing is the second. It produces beautiful predictions and is genuinely useful in dense urban and indoor work, but for a suburban macro cluster a calibrated empirical model at coarser resolution is faster and, after calibration, often more accurate. Buy ray tracing for the projects that need it, not for the whole organisation.
How do licence models differ, and why does it matter?
Three models dominate: perpetual licences with a maintenance percentage, subscription per named user, and node-locked or dongle-based seats. The differences matter far more than the headline price because planning teams are bursty. A team of four engineers with twelve occasional users of the output is badly served by twelve full seats and well served by four seats plus a viewer licence.
Also check what counts as a seat for automation. If running a batch of predictions on a server requires an additional licence per concurrent process, the cost of automation scales with the thing you most want to do. That clause is worth negotiating before signature, because it is rarely negotiable afterwards.
What does migration actually cost?
Three components, and the licence is the smallest. Data migration is the first: site databases, antenna pattern libraries, clutter and vector data all have to be validated in the new tool, and validation is manual work by an experienced engineer. Model recalibration is the second: a new propagation implementation needs its own calibration against your drive tests, even if the underlying model is nominally the same.
The third and largest is the workflow relearning. A planning team's productivity lives in habits — where the buttons are, how the reports are built, which export feeds which downstream process. That cost is real, it lands on your best people, and it is the reason a marginally better tool can still be the wrong purchase.