Distance creates an appetite for measurement, and the measurements most readily available are the ones worth least.
What not to measure
Hours logged. Keystrokes. Mouse movement. Screenshots at intervals. Application focus time.
Every one of these measures presence or activity, and none measures output. They are also gameable, cheaply and by anybody, which is why devices that simulate activity exist and sell.
We do not offer this kind of monitoring and we would advise against it if asked. A client who needs to watch a screen to know whether work is happening has a problem that watching will not solve, and installing it converts a management question into a surveillance relationship that is difficult to reverse.
What to measure instead
Throughput. Units of whatever the work produces, per week. Invoices reconciled, tickets closed, records validated, assets produced.
Error rate. Against your own check, on a sample rather than on everything. The sample is the point: reviewing all of it is the cost you were trying to remove.
Turnaround. Time from a request arriving to it being finished, which is what your own customers experience.
Rework. How much has to be done twice. It rises before quality problems become visible elsewhere and is the best early signal available.
Questions per week. High at first and falling. A flat line means the documentation has stopped improving; a sudden drop means somebody has stopped asking, which is worse.
Sampling rather than reviewing
Ten per cent, chosen at random, checked properly. It costs a fraction of full review, it produces a defensible error rate, and it is the only approach that scales past one person.
What happens when a measure becomes a target
It stops measuring. Throughput attached to a consequence produces throughput: work split into smaller units, easy items taken first, hard ones left. None of that requires anybody to be dishonest, and it is what any measure with a consequence attached does.
Use these numbers to see whether something is changing, and use a conversation to find out why. The moment the number itself is the thing being managed, you have lost the instrument.
What cannot be measured
Judgement, care, and the work that prevents problems rather than producing output. A person who notices an anomaly and asks about it has done the most valuable thing that week and it appears in no count.
Which is an argument for the monthly conversation rather than for a better metric.
Reviewing the numbers
Monthly, on the trend rather than the week. Weekly figures for a single person are noisy enough that a bad week means nothing, and reacting to noise teaches people that the numbers are arbitrary.
Show them the numbers. A person who can see their own throughput and error rate will manage them, and one who cannot will assume the worst about how they are being judged.
The number that matters most
Your own time. How many hours a week you spend on this person, tracked honestly, and whether it is falling.
If it is not falling by month three, something is wrong with the documentation or the role rather than with the person, and the entry on choosing the first role sets out how to tell which.
The dashboard question
Clients frequently ask for a dashboard. It is a reasonable request and it becomes a problem when it is populated with whatever is easy to collect rather than with the five things above.
Build it from the five, or do not build it. A dashboard of activity metrics is a monitoring system with a different name.
Comparing people
Between two people doing identical work, throughput comparisons are informative. Between people doing different work they are noise, and publishing them creates competition on the measured dimension at the expense of everything else.