Pricing Agent

Unlike a traditional tool, Pricing Agent makes decisions on behalf of the user, setting price ceilings on work orders automatically from past invoices and market data. The design challenge is trust in a system that acts without being asked.

The experience was invisible. Users could not see when the agent was working or why, so they had no way to build trust in it. The design work was making its decisions legible and giving people a way to steer it.

The Problem

Users can't see whether the agent set the price, why, or how it compares to their usual rates. Without that transparency and control, they have no reason to rely on it. The redesign gives them both.

Decisions

Every price shows who set it. The agent, the user, or the default, at the moment of creation, in the work order list, and on the invoice. I considered a badge on agent-priced work orders alone and rejected it, because the comparison is the point: a user judges the agent's price against what they would have set.

The agent is tuned, not toggled. The old setting was a strategy and a savings level. The new one lets a manager choose how aggressive the agent is, which work types and vendors it touches, and what it optimizes for. Someone uneasy about the agent narrows its scope instead of switching it off.

The money sits next to the switch. Settings show the savings so far when the agent is on, and the projected saving when it is off. Running an agent is a financial decision, so the number belongs where the decision is made, not in a report.

Checks are listed, not scored. The invoice recommendation shows every check it ran and which ones failed. A single confidence score would have been cleaner, but Invoice Agent had already taught us that a score gets misread. A list is longer and gets trusted.

Impact

The agent lowers what customers pay on invoices. The algorithm was already powerful. The design problem was making it something people trusted enough to rely on.