Professional Services Automation: What To Look For, and What To Ignore
Every PSA demo shows the same dashboard. The differences that matter are three levels below it.
Professional services automation is a crowded category whose products describe themselves almost identically: resource management, time and billing, project delivery, analytics. Sit through four demos and the feature lists blur. The differences that decide whether the thing works for your firm are structural, and they do not appear on a comparison grid.
One data model, or several products in a bundle
The first question is whether the opportunity, the engagement, the contract and the invoice are the same record or four records joined by a nightly sync. Bundled suites are common and the seams show exactly where the work crosses them: the scope that has to be re-keyed, the signed contract that lives somewhere the delivery team cannot see, the hours reconciled against tasks by hand.
Whether acceptance exists at all
Many PSA tools have no concept of engagement acceptance, because they were built for firms whose main constraint is utilisation rather than regulation. If your work needs conflict checks, independence, sanctions screening or a risk tier before mobilisation, a tool without that concept means running it in a spreadsheet beside the tool, which is where it will be forgotten.
What the client can see
Ask specifically how the client portal enforces its boundary. If the answer is a visibility flag and a filter, you are being told that correctness depends on every developer remembering. If the answer is a separate read path that cannot reach internal data, that is a structural guarantee.
Whether the numbers reconcile
The test that separates serious tools quickly: pick a number on the dashboard and ask what it is computed from. A margin figure that cannot be traced to the rate card, the recorded time and the approved change requests is a number nobody should present to a partner.
What to ignore
Ignore the count of integrations, which measures a partnerships team rather than a product. Ignore dashboard screenshots, which are the cheapest part of any system to build. Ignore artificial intelligence claims that do not say what the model is doing and what happens when it is wrong.