Why score vendors at all
Most firms pick vendors by habit and drop them after one bad week. Both are memory, and memory is weighted toward the last call that went wrong. A scorecard gives you the whole quarter, the same measures for every vendor, and something specific to talk about when you need a vendor to change.
It also helps when an owner asks why you sent the vendor you did. "They close jobs fastest and have the fewest callbacks in our records" is a better answer than "we always use them".
What to score
Score only what you can count from records you already keep. A measure you have to guess is a measure that will be argued with.
| Measure | How to count it | Where it comes from |
|---|---|---|
| Response time | Assigned to first contact or scheduled, per job | Work order timestamps |
| Time to complete | Assigned to marked complete, per job | Work order timestamps |
| Callbacks | Jobs reopened, or a new order for the same issue soon after | Work orders, by unit and category |
| Invoice quality | Invoices sent back for a missing description, photo or price | Your AP notes |
| Communication | Jobs where you had to chase for an update | Work order notes, or your follow-up log |
| Paperwork | Insurance certificate and licence current: yes or no | Vendor file |
Leave price off the scorecard and look at it next to the scores instead. A cheap vendor with callbacks is not cheap, and you only see that when the two sit side by side.
The template
One row per vendor per quarter. Copy these columns into a sheet:
Vendor | Trade | Jobs this quarter | Median response (hours) | Median time to complete (days) | Callbacks | Invoices sent back | Jobs we had to chase | Insurance current | Licence current | Notes
Use the median, not the average. One job that waited three weeks for a part will drag an average out of shape and say nothing about the other forty.
Set your own thresholds for good, watch and act, by trade. A same-day standard that is fair for a plumber is not fair for a roofer in storm season. Write the thresholds down before you score anyone, so the line does not move to fit a vendor you like.
Do not score a vendor on fewer jobs than your firm decides is enough. Three jobs in a quarter is a story, not a pattern. Mark low-volume vendors "too few jobs" and look again next quarter.
Running it each quarter
One quarter, one scorecard
Thirty minutes a quarter is enough once the export is set up. The first quarter takes longer because you will find work orders without a vendor, completions never marked, and invoices filed against the wrong job. Fix the process that caused them, not just the rows.
Talking to a vendor about a score
Lead with the number and the job, not the label. Vendors respond to "these four jobs took more than a week to schedule" far better than "your score is low".
Hi [name], we review every vendor's jobs each quarter. Most of yours went well, and [something specific that went well]. One thing we want to fix together: [measure], for example [two or three work order numbers]. What would help on your side? If it is how we send jobs, tell us and we will change it.
Half the time the cause is on your side: a vague work order, no access details, a resident who never answered. The scorecard is how you find that out too. If the problem is updates, the vendor follow-up templates cover what to send at each stage.
Fill it with AI
Counting timestamps across a few hundred work orders is slow by hand and quick for AI. Export the quarter's work orders as a spreadsheet and put your thresholds in a ChatGPT or Claude Project. Then send:
Using the attached work order export, fill the vendor scorecard in this Project for each vendor: jobs, median response hours, median days to complete, and callbacks (same unit and category reopened or reordered within [days]). Count only from the dates in the file. List the work order numbers behind every callback. If a row has no vendor or no completion date, leave it out of the scores and list it at the end as DATA TO FIX. Do not rate or rank the vendors.Spot-check three vendors against the raw export before you use it in a conversation.
Before you act on a score
- The thresholds were set before scoring, by trade.
- The vendor had enough jobs in the quarter to judge.
- Every callback traces to real work order numbers.
- You checked whether the delay started on your side.
- Price is read next to the scores, not inside them.
Questions
What should a vendor scorecard measure?
Things you can count from your own records: response time, time to complete, callbacks, invoice quality, how often you had to chase, and whether insurance and licence are current.
How often should I score vendors?
Quarterly works for most firms. Monthly gives too few jobs per vendor to judge fairly.
Should price be part of the vendor score?
Keep it beside the scores rather than inside them, so a cheap vendor with callbacks is easy to spot.
Can AI build a vendor scorecard?
Yes, from a work order export. Have it count only from the dates in the file and list the order numbers behind every callback, so each number can be checked.