Five Prompt Patterns
Claude training for store and regional decisions
At Car-Mart the decisions that matter are almost always about one specific car or one specific account. These five patterns are how operators get Claude Desktop to stay inside those decisions instead of drifting into general suggestions. Each pattern is anchored to a real judgment that happens on the floor or in the payment book.
Be specific
Vague questions produce answers that ignore the actual constraints on a given lot. When you name the vehicle, the days on the floor, the exact repair items, and what similar units have actually sold for at that store, the response has to work with real numbers instead of general ideas about "maximizing turns."
The Pine Bluff store has a 2015 Chevrolet Silverado that has been on the floor 48 days. The reconditioning manager has a written quote for $1,650 to replace the water pump and do the front brake job. The store manager must decide whether to approve the spend so the truck can be priced at $11,900 or send it to the next auction while half-ton trucks are still moving in that market.
- • Name the exact year, model, and current days on the floor
- • List the specific repair items and the written quote amount
- • Include what comparable trucks actually sold for at that store in the last 60 days
- • State the current traffic pattern for trucks in that price range
- • Name the alternative: what the auction is currently netting on similar trucks after fees
Give it a role
The same payment history can support very different decisions depending on whose judgment is being applied. When you tell Claude it is acting as the person who has actually spoken with the customer before and who knows the store's current back-lot count, the answer respects the real risk tolerance instead of defaulting to either maximum leniency or maximum firmness.
A customer at the Jonesboro store has made 11 on-time payments on a 2018 Nissan Altima but missed the last two. The file shows one prior extension granted six months ago. The customer works at the local poultry plant and says overtime was cut. The store needs to decide whether to approve a 10-day extension or begin the first repossession contact.
- • State the role as the account manager who has spoken with this customer on the phone three times before
- • Include the store's current number of vehicles already in the back lot
- • Direct it to weigh the customer's actual payment pattern against how many other accounts are in the same window
- • Ask it to consider what has happened with similar extensions at that store in the last 90 days
Show your work
When a recommendation arrives without the reasoning, you still have to do the work of validating it against what you know about each store's customer base. Asking for the assumptions and the data first lets you see immediately where the logic does or does not match the actual traffic patterns at Springdale versus Harrison.
The regional manager is considering moving a 2019 Honda Civic and a 2017 Ford Escape from the Springdale store to the Harrison store. Springdale has been turning smaller cars faster than the regional average. Harrison has stronger demand for anything under $10,000 but limited front-row space and slower overall volume.
- • First ask for the assumptions it is making about demand in Harrison versus Springdale
- • Then ask it to show the recent turn-rate data it would use for each vehicle at each location
- • Require it to state the trade-offs (space at Harrison, current pricing pressure at Springdale)
- • Only then ask for the actual recommendation on which cars should move
Iterate
The first answer almost never accounts for every constraint that exists after delivery. The second and third passes usually move the recommendation once you add the actual service department findings and what the customer said on the phone. This is how you stay inside both policy and the relationship on the Conway lot.
A customer at the Conway store took delivery of a 2014 Toyota Camry three weeks ago. During a follow-up visit they requested, the service department found a leaking radiator. The original condition report listed the cooling system as "fair." The customer is current on payments but upset. The store manager must decide what repair or adjustment to offer while staying inside company guidelines on post-sale adjustments.
- • Start with the basic facts of the sale date, vehicle, and reported issue
- • Run once, then add the exact language from the original condition report
- • Add the service department findings and the customer's actual words from the call
- • Add any prior service history on the car and recent similar post-sale issues at that store
- • Compare what changed between passes before using any part of the response
Save the prompt
Certain decisions repeat on a fixed schedule. The first time you get an answer that a store manager can actually act on, the real work is capturing the structure so the next time you only have to drop in the new vehicle list and current sales data. This is how the regional team keeps the 60-day review consistent across five stores without rebuilding the logic every Monday.
Every Monday the regional ops team reviews vehicles that have been on any of their five stores' lots for more than 60 days. For each vehicle they decide whether to reduce the price, invest in additional reconditioning, transfer it to another store in the group, or send it to auction. The categories of information stay the same even though the specific cars change every week.
- • After you have worked through one full Monday review and the output is usable, save the prompt
- •