How Many Visits Is Enough
Enough is the number of visits that lets the result carry the decision it will be used for, and that number differs for a reading of the whole network and a reading of one outlet.
A single visit to a single outlet records one member of staff, at one hour, on one day, handling one scenario. Any of the four can be unusual without anybody doing anything wrong: the experienced salesperson was on leave, a delivery had just arrived, the queue was the longest of the week. Repeat the visit and the accidents begin to average out; spread visits across outlets and sites can be compared with each other rather than with an impression.
One visit
An observationShows what happened on that occasion. Useful as a check that a standard exists at all, not as a measure of how the outlet normally performs.
Repeat visits at one outlet
An outlet scoreSeveral visits on different days, at different times and with different staff begin to separate a weak outlet from a weak morning.
Visits across many outlets
A network readingMany single visits, spread well, describe the network as a whole even though no individual outlet is measured reliably.
Ask one question before setting the count: will these results be used to compare outlets, or only to describe the network? The answer moves the number of visits more than any other choice.
Sampling by Outlet Count
The number of outlets changes the shape of the programme more than its size. A network of twenty or thirty outlets should be visited in full: sampling saves very little at that scale and gives up the comparison between sites, which is usually the reason for running the programme. Visiting every outlet several times produces a result where each site can be discussed on its own evidence.
Large networks have to sample, and the sample is built rather than drawn evenly. It has to represent the formats operated, the regions traded in, new and long-established outlets, and high and low trading volumes, because a sample that leans towards metropolitan flagships describes the flagships and nothing else. A simple way to keep it honest is to list the outlets by format and region first and then draw visits from each group in proportion to its share of the network.
Network size matters less than most buyers expect. The precision of a network figure depends on the number of visits, not on the number of outlets, once the network is more than a few dozen sites:
How precise a network figure is, by number of visits
For a yes-or-no question (for example, “was the customer greeted within a minute?”), visits chosen at random, 95% confidence. Standard arithmetic, not an industry rule.
100 visits
- Within about 10 points either way
- Shows only large problems
400 visits
- Within about 5 points either way
- A usable network reading for most decisions
1,000 visits
- Within about 3 points either way
- Little more to gain for the network figure
The same arithmetic holds whether the network has 300 outlets or 3,000. What grows with the network is the number of outlets that each need a reading of their own, and that is where the visit count climbs.
Frequency Against Depth
A fixed budget forces a choice between visiting more outlets once and fewer outlets more often. The two answer different questions, and a programme that does not choose ends up answering neither well.
- More outlets, one visit each, measures level: where the network stands as a whole, and which sites sit at the extremes. It cannot tell a poor site from a poor morning, because one observation carries both.
Answers: where do we stand? - Fewer outlets, repeat visits, measures change: whether a corrective action worked, and whether a bad result was an incident or a pattern. It says nothing about the outlets left out.
Answers: is it moving?
Most programmes need level in the first year and change after that. Establish where the estate stands, find the outliers at both ends, then concentrate repeat visits where the answer matters and release them from sites that have shown they are stable. Reviewing that allocation at the end of every cycle keeps the programme answering a current question rather than the one it was designed around.
Weighting by Risk and Value
Once a baseline exists, an even spread of visits is the least useful allocation available. Four weightings usually earn their place:
- New outlets and new managers. Performance in the opening months is the most variable and the most correctable, and habits formed early tend to stay. A store open six weeks and a store open six years should not get the same schedule.
- Outlets that scored badly. Revisit them, but time it: far enough after the corrective action to be fair, close enough to act on. A revisit the week after training measures the training, not the outlet.
- Outlets that carry the turnover. Where a few sites carry a large share of sales, a service or compliance failure at one of them costs more than the same failure at a marginal site, and the visit plan should reflect that exposure.
- Anything just changed. A new pricing rule, a new returns process or a product launch is adopted unevenly; extra visits for a cycle show where it has not landed.
Weighting moves visits; it should not remove an outlet from the programme. Keep a minimum for every site so that a quiet decline at a stable outlet is still seen.
What the Numbers Can and Cannot Support
A visit programme supports conclusions at two levels, and they need very different amounts of data. Network conclusions are the easier case: with visits spread over enough outlets, the overall score describes the network and its movement between cycles means something, even though no single outlet was visited often. Outlet conclusions are far more demanding, because a score for one site rests only on the visits that site received.
- Supported by many single visits: the network score, its change between cycles when the scorecard has not changed, and which questions fail most often across the estate.
- Supported only by repeat visits at an outlet: ranking that outlet against others, and saying it has improved or declined.
- Supported by no visit count: a judgement about a named person drawn from one visit.
Where an outlet has been visited once, the honest statement is that on one occasion these things were observed. A score stops being evidence at the point it is used for a decision the sample cannot carry, and a single-visit score in an individual's appraisal is the clearest example: it pins on one person a result that may belong to the hour, the queue or the day. Changing the scorecard between cycles has a similar effect on trends, because the old and new scores no longer measure the same thing.
Setting the Cycle
A defensible first-year plan establishes a baseline for the network and gives each outlet enough visits for its own result to mean something. On a fixed budget those two pull against each other, and the usual answer is to cover every outlet at modest depth in the first cycle rather than a subset in depth, because the first year's job is to find out where the network actually stands.
- Decide whether the results must describe the network, compare outlets, or both.
- List the outlets by format, region and trading volume, and set a minimum number of visits for every one.
- Spread the visits across weekdays and weekends, opening, peak and closing hours, since a cycle run on weekday mornings measures weekday mornings.
- After the baseline, move visits from consistently good outlets to those that scored badly or moved sharply.
- Review the allocation at the end of each cycle, keeping the core questions unchanged so the trend holds.
The number of visits is also what sets most of the fee; the drivers of mystery audit cost are explained separately.
Bringing in an outside team to design the cycle is worth it where the network is large, where formats differ enough that one scorecard will not serve, or where the results will feed incentives or appraisals. A mystery audit programme can then set the outlets, scenarios and visit plan around the decision the results must support.
