AI and delivery time promises you can actually keep

6 days ago
3 min read

Updated: 3 days ago

Introduction

The date you promise is a commercial decision disguised as an operational one. Promise too long and you lose work to a competitor who promised sooner. Promise too short and you win the job and then generate a complaint, a chase, an apology and a customer who does not return. Most businesses do the second, and treat the resulting complaints as a service problem rather than a promising problem.

What makes this tractable is that you already know how long things actually take. The distribution of your past lead times, by job type, is the honest basis for a promise, and comparing it against what you have been telling customers is usually an uncomfortable but decisive exercise.

1. AI and delivery time promises should come from your recorded distribution

Not from the best case.

Measure actual lead times per job type over a year and look at the whole spread, not the average. The promise belongs somewhere in the upper part of that spread, not at the fastest observed time.

2. Promise a percentile, and decide which one

The key choice.

Promising the median means missing half your dates. Promising the eightieth percentile means keeping four in five. Making this an explicit decision, rather than an accident, is most of the improvement available.

3. Make the promise depend on current load

The dynamic part.

Your realistic lead time in a quiet month and a full one are different, and a single published figure is therefore wrong most of the time. Tying the quoted date to current committed work is where the analysis adds something.

4. Include the dependencies you do not control

The common omission.

Supplier lead times, subcontractor availability, access dates, customer approvals, permits. A promise that assumes all of these behave is a promise you will break, and the variability of these is often larger than your own.

5. Communicate a window, then narrow it

Two-stage promising.

An initial window at order, then a confirmed date once the work is scheduled and materials are secured. Customers accept this readily and it removes the pressure to commit precisely at the point you know least.

6. Measure delivery-to-promise, not lead time

The metric that matters.

The proportion of jobs delivered by the date you gave, reported weekly. A business with long lead times and reliable dates is preferred over a faster and erratic one by most customers, and this number is what tracks it.

7. Tell the customer early when a date will slip

The failure that causes complaints.

Customers absorb a delay they know about in advance and react badly to discovering it on the due date. Detecting the slip early enough to communicate it is a scheduling capability, and it is worth more than shaving days off the lead time.

8. Look at where the time actually goes

Usually not where you think.

Most lead time in small businesses is waiting rather than working: waiting for approval, for materials, for a slot, for a decision. Attacking queue time is far more effective than trying to work faster.

9. Use reliability as a selling point

The commercial return.

Once your delivery-to-promise performance is genuinely high, say so and quantify it. Reliability is difficult for competitors to match quickly and it justifies a price premium in a way speed alone does not.

Be careful about promising differently to different customers for the same work. Where an expedited date is available for a fee that is a published option; where it depends on who asked, it becomes a problem when they compare.

Conclusion

Base the promise on your recorded lead-time distribution rather than on your best case.

Choose deliberately which percentile you are promising, tie the quoted date to current committed load, include supplier, subcontractor and approval dependencies in the calculation, give a window at order and confirm a date once scheduled, report delivery-to-promise weekly instead of lead time, tell customers about a slip as soon as you know rather than on the due date, attack the waiting time rather than the working time, and use a genuinely high reliability figure as a selling point.