Spark is the assistant behind the sparkle button at the top right of the application. It answers questions against your live account data rather than searching help articles, so you can ask it about your own invoices, vendors and spend. This article covers what Spark is suited to, what it does not do, and how to check an answer before acting on it.

Verified 2026-08-26.

Spark has to be enabled on your account, and it is not switched on for every account. Where it is not enabled, opening the panel shows this message: Spark unavailable. It may not be enabled on this server, or the connection dropped. Close and reopen to retry.

Closing and reopening the panel is worth one attempt, because the same message appears when a connection drops. If the message persists, Spark is not enabled for your account, and no setting on your side changes that. Contact support to request it.

Every question Spark answers can also be answered from a screen. Which screen answers which question is the routing table.

Spark suits questions where you know what you want and not where it lives. Examples include which vendors have not sent a statement this month, how much uncategorized spend one shop is carrying, and whether a specific invoice has been matched.

It reads your data at the moment you ask, so an answer reflects the current state of the account rather than a cached report.

It is not a report builder. Spark answers questions. It does not produce a table you can save, schedule or export, and there is no query builder elsewhere in the product either. See what WickedFile does not do.

It is not a help system. Spark reads your account data rather than this documentation, so questions about how a feature works are answered here instead.

It is not a replacement for the screen that owns a number. A figure from Spark is a starting point. Where it reports that a shop is carrying a large amount of unplaced spend, the list of lines behind that figure sits in the AP Hub, which is where those lines are worked.

Where Spark can act rather than answer, it proposes the change and waits. You accept or decline, and declining leaves your data as it was.

The boundary is deliberate. A part code is the final state set by a person, so an assistant does not write one on its own. See part codes.

Spark shows what it based an answer on. Reviewing that is the quickest way to confirm it answered the question you asked rather than a similar one.

For a figure you plan to act on, and particularly a dollar amount you plan to quote to a vendor, confirm it on the screen that owns it first.