Gap analysis

A gap needs two numbers. Most problems arrive with none.

Two numbers and a period

A gap is a subtraction. It needs a measure, a value the measure has now, a value it ought to have, and the stretch of time both were counted over. Four elements, and problems almost never arrive with more than one of them.

Rummler put the point at the level of the whole organization long before anyone was arguing about dashboards: a result is a property of the system that produces it, so a number without the system around it is not yet information (Rummler & Brache, 2013). The same holds one level down. A measure with no desired value beside it cannot tell you whether the system is working.

The period matters more than it seems. A single month is noise; a single year hides everything that happened inside it. Four quarters is usually the shortest span that separates a trend from a bad week, and it has the additional merit of being the span the organization already reports on.

Current performance is harder than it looks

The measure exists and somebody reports it, so the first number feels like a lookup. It is not. Four faults recur in it, and all four are found by asking one question: who computes this, and from what?

The fourth deserves emphasis because it is the one that quietly invalidates the whole analysis. Any measure that people know they are judged on will be managed, and the management shows up in the number before it shows up in the work. That is not dishonesty; it is what measures do. The remedy is not suspicion but triangulation: one independent trace is enough to establish whether the reported number and the actual one have drifted apart.

Where desired performance may come from

This is the weakest joint in most business cases and almost nobody inspects it. The first number can be looked up. The second one has to be argued for, and there are only four places it can honestly come from.

The first source is the gift this question receives from the next one. Number 07 establishes who already achieves the result and what they have that the others lack; the level they reliably reach is the desired level, and it arrives with its own proof attached. Where no such person exists, that absence is itself a finding: a level nobody has ever reached is a design problem rather than a benchmark, and it should be reported as one.

And a word about targets

There is a long-running argument in management about goals, whether stretching targets lift performance or distort it, and the desired level in an analysis sits close enough to it to be mistaken for one. It is worth separating them, because the two are not the same object.

Both sides of that argument have something. A specific, difficult target does concentrate effort, and organizations that over-prescribe targets reliably get narrowed attention, gaming and the occasional scandal (Ordóñez et al., 2009). Where the tipping point lies in ordinary work is not settled, and this paper does not need it to be, because the desired level in an analysis is doing a different job entirely.

Keeping those apart has a practical consequence that is easy to miss. A measurement point can be set at what a colleague already achieves, because nobody is being asked to promise it. The moment the same number becomes a target, the person it applies to acquires an interest in arguing it down, and the conversation stops being about evidence. Analyses that conflate the two lose the argument they should have won.

Four ways a desired level goes wrong

Each of these produces a number that looks authoritative and cannot be defended. Together they account for most of the desired levels we have met in practice. [practitioner observation, not a measured frequency]

Gilbert’s answer to all four is the first source in the previous section, and it is worth stating as a rule rather than a preference: the standard is what the exemplary performer accomplishes, because that is the only level anybody can prove is reachable in this organization (Gilbert, 1978).

That question is worth asking out loud and in the meeting. Not as a challenge but as a request: walk me from something we already know to this figure. A desired level drawn from source one or two takes about thirty seconds to derive. One drawn from source four cannot be derived at all, and the silence that follows is the finding.

The distribution of the gap and what it tells you

Here is the part of this question that most analyses never reach. The size of a gap tells you what is at stake. The distribution of it (how the shortfall spreads across the people, sites or products that make up the number) tells you where to look for the cause.

Take a unit of twenty people doing one job. The desired level is 90 and the unit averages 78, so the gap is twelve points. That single sentence is what almost every analysis reports, and three completely different situations produce it. In the first, all twenty sit between 77 and 79: everyone is short, nobody much more than anyone else. In the second, fifteen are at 86 and five are at 54. In the third, ten are at 86 and ten are at 70. Each of the three averages 78. Each is a different problem with a different fix, and the average conceals which one you have.

The practical difference is large. If the shortfall is uniform, whatever is chosen has to reach all twenty people, and it is worth paying for something that does. If it is concentrated, fifteen of the twenty need nothing at all, three-quarters of the unit, and a program bought for the whole unit spends three-quarters of its budget on people who are already at the level. The average is the same in both. It cannot tell you which of the two you are about to buy.

Reading the distribution, and the trap inside it

The three patterns point at three different kinds of cause, and the reasoning behind each is short enough to state in a sentence. The twenty people from the previous page are still the example.

This is also where the gap becomes a direction. Number 08 asks which level a cause sits at, number 09 what kind it is. The distribution answers neither, but it tells you which to open first, which on a four-week analysis is worth a great deal.

When there is nothing to put on either side

Everything above assumes a measure exists. Often none does, and the honest answer is neither to stall nor to invent one. It is to work out what could be established this month, and to say out loud how little is known meanwhile.

The situation is familiar. A director says service feels slower since the new till system. Chapter leads say volunteers are drifting away. Partners say projects are slipping. Founders say the engineers lack discipline. In each case nobody measures anything, the word at the center of the complaint has never been defined, and somebody has already chosen the training.

Refusing to proceed is not rigor, it is abdication: the request will simply be granted without you. What replaces the missing number is four moves, none of which requires data that does not exist.

And what not to do when the number is missing

Do not accept a measure you have been told to distrust. Regional managers who say their own audit scores are gamed are telling you something true and useful: number 15 opens on exactly this situation. The answer is not to use the scores anyway because they are the only ones there. It is to say which figures can be trusted, and to work from those.

Do not treat an absence as a finding. That nobody measures volunteer engagement does not establish that engagement has fallen. It establishes that nobody would know. Those are different sentences and only one of them is defensible.

Do not let the deadline manufacture the number. A board that wants “a figure we can show the regulator in sixty days” is asking for a figure, not for the truth. Whatever is produced under that instruction will be quoted for years by people who were not in the room.

The distributor whose target came from a brochure

An industrial parts distributor, 380,000 order lines a year, 41,000 stock items. The figures are illustrative and reconcile from the rates stated with them.

The stated problem was order-line fill rate: 92.0 percent against a target of 98. Six points short, and the proposal on the table was a company-wide stock increase costing $2.9 million in working capital. The board had asked twice why it was taking so long to approve.

Two questions were asked before it was. Where does 98 come from? It came from a competitor’s marketing material, three years old, with no definition of what counted as a line. What does the business actually need? The contracts with the three largest customers specified 95 percent, with service credits below it. Above roughly 96 the carrying cost of the extra stock exceeded the credits it avoided.

And then the distribution

The corrected gap was still a gap. What changed the decision was cutting it, and the cut that mattered had never been made.

Fill rate was reported by branch and by month. Nobody had reported it by product category. When somebody did, the 92.0 percent broke into two populations that had nothing to do with each other.

The proposal died on that table. A company-wide stock increase would have spent $2.9 million to raise a figure that was already sufficient across four-fifths of the volume. The finding pointed at 8,200 slow-moving items rather than 41,000, and at whatever it is about slow movers that makes them behave differently: a forecasting method, a reorder rule, a supplier lead time. Which of those it turned out to be is question four, and number 08 is where it gets answered.

What this does not claim

A method that produces numbers should be plain about what those numbers are not. Four limits.

• A gap is not a cause. It is a measured distance, and it says nothing whatever about why the distance exists. An analysis that stops here has produced a well-quantified complaint. Numbers 08 and 09 do the causal work, and they can fail.
• The distribution is a pointer, not a finding. A uniform shortfall suggests a shared condition; it does not establish one. Every pattern in section 05 is an instruction about where to look next, and each of them can be wrong.
• Every cut can be gerrymandered. Cut a population enough ways and some cut will look concentrated by chance. Decide which cuts are plausible before you make them, report the ones that showed nothing as well as the one that showed something, and treat a cut you went looking for with more suspicion than one that announced itself.
• The desired level is a claim about this organization. Set from what an exemplary performer achieves, it is defensible here and nowhere else. It is not a standard, not a benchmark for the sector, and it should not be quoted outside the analysis that produced it.

Both numbers, and the cut you have not made

Twenty minutes, on the measure you established in question one. Six lines. A blank is the finding, and line four is the one people cannot fill.

Two problems, one method, sixteen papers

Two models carry the method. Gilbert’s six conditions ask what kind of cause; Rummler and Brache’s three levels ask where it sits. Both are sound and this series uses both, and it leads with Gilbert, because his unit is the accomplishment, which is what an organization already counts, and because his six conditions are the categories an organization can actually change.

Where the method comes from

Two further pieces close the set and carry no number, because each points at all the numbered papers. Before the guide sits Arets L&D Business Models, the destination piece for the people who receive the requests: four models a learning function can run, and the route from HR-driven to business-driven. Last sits Solutions that cannot work, the field guide, ordered by the sentence that lands on your desk: seventeen forms a performance problem arrives in, each with an address in one of the numbered papers. Three orders in all: the destination you choose, the question you are on, and the form that landed on your desk.

Sixteen papers, six parts

Number 01 is the map and number 02 is the foundation. From number 05 on, every number is depth on exactly one step of the method. The filled block is the paper you are holding; the two unnumbered pieces close the set.

Where every claim in this paper comes from

The three books named earlier in this paper are listed in full in number 02, so that they do not recur in every reference list.