I once picked up a multi-vendor environment where on-time delivery was running under 20%.
Not 20% of work delivered late. 20% delivered on time. Four out of five commitments missed, month after month, and the quarterly business review deck was full of green squares.
We got it above 65% without renegotiating a contract, adding a penalty clause, or replacing a supplier. What changed was measurement.
Nobody was tracking the date that mattered
There were three dates floating around for every piece of work, and everyone used the one that flattered them.
- The date the business asked for.
- The date the vendor committed to.
- The date the vendor most recently re-committed to.
Performance was being reported against the third. Which meant a task rescheduled four times was on time, because it hit the fourth date. The vendor wasn’t lying — they were reporting exactly what we’d asked them to report.
The fix was to freeze the first committed date and measure everything against it, forever, no matter how many times it moved afterwards. Slippage became visible as slippage rather than disappearing into a new baseline.
That single change made the number drop through the floor, which was the point. You cannot fix a number you can’t see.
Make it their number, not yours
The second change was who produced the report.
When you build the scorecard yourself, every review turns into a debate about your methodology. When the vendor produces it against a definition you agreed in advance, the review is about performance instead. Same data, completely different conversation.
What that requires up front:
- A definition of “done” that isn’t ambiguous. Delivered means in production and accepted, not “handed to the customer for testing.”
- A single agreed source for the raw data. Usually their ticket system, which is fine, as long as you can see into it.
- A cadence that’s frequent enough to correct. Monthly. A quarterly review discovers a problem a quarter late.
Escalate early, and about the trend
The instinct is to escalate when something specific goes wrong. That produces a conversation about one ticket, and one ticket always has an explanation.
Escalating on the trend is harder to deflect. “Six of your last ten commitments slipped” is not a conversation about a ticket. It’s a conversation about capacity, and capacity is a thing account managers can actually go and fix.
None of this is adversarial. Most vendors want to hit their dates; they’re optimising against whatever you inspect. If you inspect the re-committed date, they’ll get very good at re-committing.