Nobody in a restaurant group argues about whether the guest data is any good. They argue about whose fault it is, which is a longer conversation with no natural end.
A score does not fix anything. What it does is convert an opinion into a number that moves, which is enough to direct attention this quarter and to tell you next quarter whether the thing you did worked.
Four measures, all of them from a single export, and the whole exercise is an hour once you have the file.
The four measures, and why these four
Each of these blocks something specific, and you can compute all four without buying anything. There are more sophisticated measures of data quality. There is not much point owning them until these four are decent.
- Fill rate — how many guest records carry a usable contact detail. Decides whether you can reach anybody at all.
- Duplicate rate — how many records are the same person more than once. Decides whether any count you produce is true.
- Consent coverage — how many contactable guests you are actually permitted to contact, per channel. Decides what you may do with the rest.
- Recency — how much of the file describes your restaurant now rather than your restaurant in 2019. Decides whether any of it is worth acting on.
Computing each one
Start from one export — one location, all guests, all time — and resist the urge to clean it before you measure it. The mess is the measurement.
N = total guest records in the export FILL RATE = records with a valid email or phone ÷ N DUPLICATE RATE = (N − distinct people) ÷ N Estimate distinct people from distinct phone numbers and distinct emails, then read a sample of 100 records by hand and adjust. A hand count on a sample beats a confident number from nowhere. CONSENT COVER = records with a recorded consent for a channel ÷ records holding that channel's contact detail RECENCY = records with a visit in the last 24 months ÷ N
Turning four measures into one number
Score each measure out of twenty-five and add them. The weighting is deliberately equal: the moment you weight one higher you are asserting something about your business that you probably cannot defend to the person who disagrees.
The bands below are ours rather than an industry standard. There is no industry standard, and anybody who quotes you one has either invented it or is describing their own customer base generously.
For each measure, score against the rate you computed: 25 strong rate above 0.8 18 workable 0.6 to 0.8 10 weak 0.4 to 0.6 3 broken below 0.4 Duplicate rate is inverted — low is good: 25 below 0.05 18 0.05 to 0.10 10 0.10 to 0.20 3 above 0.20 HEALTH SCORE = fill + duplicates + consent + recency
What each score tells you to do
The point of splitting it into four is that each one hands the problem to a different person.
- Low fill rate is a training problem at the host stand and the terminal, not a software problem. No product invents a phone number that was never asked for.
- High duplicate rate is the one to fix before you measure anything else. It inflates guest counts, deflates visit counts, and turns each of your best regulars into three unremarkable strangers.
- Low consent coverage limits what you can send, and it is much better to know that before somebody sends anyway. Fixing it means asking — in the room, at the confirmation — rather than importing from somewhere.
- Poor recency often means the file is fine and the export was wrong. Check that you pulled the right period before concluding anything about your guests.
Re-run it quarterly, compare it only to yourself
Same export, same definitions, same afternoon of the quarter, written in the same place. The discipline matters more than the sophistication.
Do not benchmark it against anybody. We do not publish an average because we do not have one, and the figures circulating on the internet are largely one vendor's customer base described in a flattering light.
Four quarters of movement in the right direction is a genuine result you can defend in a board meeting. A single score is only a starting position, and it is supposed to be uncomfortable.
Where a score changes the conversation
The reason to have a number is that it makes an argument concrete. "Our data is a mess" gets nodded at and survives to the next meeting. "Our consent coverage on SMS is a third of what it is on email, and here is the export" gets a decision.
It is also the right thing to have open during a vendor demo. A vendor who reads your score and says "we can help with two of these four and not the other two" is being more useful than one who says all four are covered.