Your Training Partner
Techniques Toolbox
The path of Kepner and Tregoe's problem analysis. The team states the problem, then specifies it in the IS and IS NOT columns. The possible causes come from the team's experience or from distinctions and changes. A cause that contradicts a cell survives the test only at the price of an assumption that explains that cell another way. The most probable cause then goes on to confirmation. When no cause passes, the team tightens the specification.

Kepner-Tregoe Method

Charles Kepner and Benjamin Tregoe described the Kepner-Tregoe method in The Rational Manager (1965), then in The New Rational Manager. It is a set of processes for solving problems and making decisions. Its core is problem analysis. The team describes the deviation in four dimensions (what, where, when, extent), first where it occurs, then where it could occur and does not. The IS NOT column is the evidence: the team selects the cause that explains both columns. Decision analysis then chooses an action through musts and weighted wants. Potential problem analysis protects the chosen action.

Goal

Kepner and Tregoe call a problem a deviation between expected and actual performance whose cause is unknown. Their problem analysis finds that cause.

The IS NOT column carries the proof. A plausible cause that contradicts a single IS NOT cell stands only at the price of an assumption.

A problem asks why. Once the cause is known, the question becomes a choice. Decision analysis answers it. Problem analysis pays off when the team would act differently if it knew the cause.

Each process supports a decision: which cause to correct, which option to choose, which safeguards to fund. Problem analysis delivers the IS / IS NOT specification, the test of the possible causes and the confirmed cause.

Usage

When to use it

  • Deviation with an unknown cause: an expected performance is not reached and nobody knows why.
  • Several plausible causes under debate: the test against the specification decides between the candidates without a vote.
  • Deviation confined to one site, one version or one period: the neighbours it spares fill the IS NOT column with facts.
  • Recurring IT incident: the change log supplies the dated changes that the search needs.
  • Problem description and cause search in an 8D: the D2 grid and the test give D4 its cause to verify.
  • Experts who dominate the session: a facilitator from outside the subject asks the questions and the experts bring the facts.
  • Suspicion aimed at one person: the specification tests the accusation against the facts before any decision.
  • Distributed team: each member enters their facts in a shared specification before the session.
  • Choice between several options under constraints: decision analysis separates the mandatory from the desirable before it compares.
  • Go-live, migration or office move to protect: potential problem analysis prepares the countermeasures and their triggers.

When not to use it

  • No comparison possible: without facts or a spared object, the IS NOT column stays empty; brainstorm the causes with a fishbone diagram and measure.
  • A cause that would not change the action: an adaptive action is enough; treat the effect directly.
  • Direct cause known and a causal chain to trace back: follow the chain with the five whys.

Description

The processes of the method

Kepner and Tregoe separate their processes by the question each one handles. A fourth process comes before the other three: situation appraisal sorts the concerns of an organisation and routes each one to the process that fits it.

ProcessQuestionDeliverable
Problem analysisWhy does the deviation exist?The IS / IS NOT specification, the test of the causes and the confirmed cause
Decision analysisWhich action should the team choose and to what end?The grid of musts and weighted wants, the choice and its risks
Potential problem analysisWhat can go wrong and what should be done now?The plan of preventive and contingent actions, with their triggers
The three processes of the Kepner-Tregoe method: the question each one asks and what it delivers.

Stating the problem

The statement names an object and a defect: "parcels shipped without a tracking number". A vague statement such as "shipping problems" opens a search with no bounds.

If the team can explain the deviation at once, the statement sits too far downstream. "Parcels shipped without a tracking number" explains "customers call after-sales service". The analysis starts at "parcels shipped without a tracking number". Nobody can explain it yet.

One analysis handles one deviation. Two defects under one statement often have two causes. A shared specification isolates neither of them.

Specifying what is and what is not

The specification describes the deviation in four dimensions. What names the object and its defect. Where locates the object when the defect is observed, then the defect on the object. When gives the first observation, the pattern since then and the point in the life of the object at which the defect appears. Extent counts the objects affected, measures the size of one defect, counts the defects per object and describes the trend.

For each IS answer, the team looks for the closest comparison: what could show the defect under the same conditions and does not. It takes the neighbouring module, the adjacent line, the previous version or the month before. "All other parcels" compares too broadly. "Economy parcels from the same line" isolates a difference.

The team asks every question, even the one that seems pointless. Where a question does not apply, it writes "not applicable". Some rows have no logical comparison: the team records that.

DimensionISIS NOT: the closest comparison
What: objectWhich object shows the defect?Which similar object could show it and does not?
What: defectWhich defect is observed?Which other defect could be observed on this object?
Where: locationWhere is the object when the defect is observed?Where else could it be observed?
Where: on the objectWhere on the object does the defect appear?Where else on the object could it appear?
When: first observationWhen was the defect first observed?When could it have been observed earlier?
When: patternWhen has it been observed since and how often?Which times does the defect spare?
When: life cycleAt what point in the life of the object does the defect appear?At what other point could it appear?
Extent: objectsHow many objects show the defect?How many objects could show it?
Extent: sizeHow large is one defect?What other size could it have?
Extent: defects per objectHow many defects does one object carry?How many could it carry?
Extent: trendHow is the number changing?What other trend could be observed?
The specification questions: for each dimension, the question on the IS side and the question on the IS NOT side.

Finding the possible causes

Possible causes come from two sources. The first is the experience of the team: its knowledge, the similar cases it has seen, a brainstorm. The team writes each cause as a mechanism that produces the effect. "The printer driver" cannot be tested. "The new driver truncates the tracking number field" can.

The team turns to the second source when experience yields too few causes, too many or only unlikely ones. For each row, the team asks what sets the IS side apart from the IS NOT side. A distinction is a piece of information that the specification does not yet contain and that holds only on the IS side. "Priority parcels use a different label template" is one. "The parcels are different" is not. One or more dimensions often yield none.

For each distinction, the team then asks what has changed in, on, around or about it. A distinction becomes a lead when a change touches it before the first observation. The team then states the cause: how could this change have produced the deviation? Two changes can together produce a deviation that neither produces alone. A distinction that no change touches can also yield a possible cause. That cause goes through the test like the others.

The list of every change in an operation is too long to search. The distinctions narrow it to the changes that touch the IS side.

Testing each cause against the specification

For each possible cause, the team asks the same question on every row: if this is the true cause, how does it explain this IS answer and this IS NOT answer? The true cause explains every cell in both columns. A cause that contradicts a cell survives only at the price of an assumption that explains that cell another way.

A cause that holds only at the price of extra assumptions ranks behind a cause that needs fewer. The team writes each assumption out in full next to the cause. It judges the test on these assumptions: how many there are and how plausible they are.

Selecting and confirming the most probable cause

When several causes pass the test, the team compares their assumptions. The most probable cause needs the fewest, most plausible and simplest assumptions.

The team still has to prove this cause. It confirms it through an action outside the analysis: check the assumptions against the facts, observe the phenomenon, run a trial or apply a correction and watch its effect. It chooses the safest, most conclusive, cheapest, simplest and fastest action. Often the team reverses the suspect change and checks that the deviation disappears.

When no cause passes the test, the team goes back to the specification and tightens it. It does the same when the selected cause fails confirmation. The method works on the facts that the team holds. It produces none.

Statementan object and a defect
SpecificationISIS NOTwhat · where · when · extent
Team experienceknowledge, similar cases
Distinctions, then changeswhat sets the IS side apart and what changed it
Possible causeseach written as a mechanism
Testdoes the cause explain every IS and IS NOT cell?
contradicts a cell
ruled out
Most probable causeassumptions: number and plausibility
Confirmationone of: check the assumptions against the facts, observe the phenomenon, run a trial or correct and watch the effect
Confirmed cause
no cause passes: tighten the specification
confirmation fails
The path of Kepner and Tregoe's problem analysis. The team states the problem, then specifies it in the IS and IS NOT columns. The possible causes come from the team's experience or from distinctions and changes. A cause that contradicts a cell survives the test only at the price of an assumption that explains that cell another way. The most probable cause then goes on to confirmation. When no cause passes, the team tightens the specification.

The deviation present from the start

Some objects have never reached the expected performance. The authors call this a Day One Deviation. No change precedes it. The search therefore rests entirely on the distinctions. The team first examines the expected performance itself: is it realistic, has anyone ever reached it, who set it?

expected performance actual performancedeviation change
pastpresent
expected performance actual performancedeviation day one no change: search the distinctions
pastpresent
Two deviations between expected and actual performance. On the left, performance drops after a change: the team looks for that change. On the right, the deviation has existed since day one: the team searches the distinctions and examines the expected performance itself.

Common mistakes

Specifying with too few facts. The authors put this mistake first. A half-empty IS NOT column lets almost every cause through.

Describing vaguely. "The system is slow" rules out no cause. "Searches take more than 8 seconds from the Sion site between 8:00 and 9:00" rules out several.

Taking the opposite for the comparison. On the row "What: defect", the answer "parcel with a tracking number" restates the norm. The closest comparison names another defect that the same parcel could show: "illegible label".

Skipping the distinctions. The team jumps to the list of changes in the period. Nothing then sorts that list.

Letting assumptions skew the test. The favourite cause survives thanks to one more assumption on each row. The more assumptions a cause needs, the less likely it is to pass confirmation.

Debating causes before specifying. A team that opens the session on causes defends its ideas. It then fills in the specification to confirm them.

Dressing up a conclusion reached in advance. Someone picks a cause before the analysis, then presents it in the vocabulary of the method. The authors call this veneer process trinkets. Whoever leads the questioning steers the recommendation. The business analyst who asks the questions therefore carries this risk.

Abbreviated use and teamwork

Most analyses never reach paper. The manager asks the specification questions in a meeting, asks what sets the case apart and what has changed, then tests aloud each cause put forward. The authors consider this abbreviated use legitimate when time is short or facts are scarce. One abbreviated form studies a single dimension with its distinctions and its changes. That dimension is often when.

In a team, the specification gathers the facts of each member in a common format. The authors report that many of their clients give the questioning to a person with no expertise in the subject. The experts bring the facts. The facilitator stops them from drowning the group in technical detail. According to these clients, the facilitator also keeps the group from overrating the assumptions that come from expertise. The business analyst can take this role: the analyst asks the questions without defending any cause.

Decision analysis

Decision analysis chooses one action among several. It sets the objectives before it looks for options. The authors present it as the reverse of a common practice: choose an action, then build the case that justifies it.

The decision statement names a choice, an action and the intended result. It also sets the level of the decision. "Choose an electronic signature tool" assumes that the decision to sign online has been taken.

The team then classifies the objectives. A must is mandatory, measurable and realistic. It sets a minimum that every option has to reach. It does not necessarily outrank a want in importance. To test a must, the team asks which benefit makes it mandatory. If another means delivers the same benefit, the objective becomes a want. The musts decide which options stay in the running. The wants decide which one wins. The team can also restate a must as a want to rank the options that meet it: "at most CHF 12'000 a year" becomes "lowest annual cost".

The team weights the wants. The want that matters most gets 10. The team gives every other want a weight from 10 to 1 relative to the first. The authors list warning signs: too many high weights, too many low weights, too many objectives from a single stakeholder and objectives tailored to one option.

The team then lists the options. The current course of action is one of them. The options first go through the screen of the musts. An option that fails a must leaves the analysis. The team records why. The team compares the remaining options want by want. On each want, the best option scores 10 and the others get a relative score. The score multiplied by the weight gives the weighted score. The sum of the weighted scores ranks the options. According to the authors, there is nothing magic about these numbers. They make the judgement visible.

The option in the lead is a tentative choice. The team then looks for adverse consequences. It starts with the leading option and examines it without comparing it to the others. It then examines the next options the same way. It rates each consequence for probability and seriousness on a word scale (high, medium, low) or from 10 to 1. With numbers, it does not multiply the two ratings. The authors report that their first book did so and that those totals proved useless. A serious risk can lead the team to choose the option ranked second. The final choice weighs the score and the risks together.

ObjectiveWeightTool ATool BTool C
Must: data hosted in Switzerland–YesYesNo: hosted in the European Union. Option dropped
Must: annual cost at most CHF 12'000–Yes: CHF 9'600Yes: CHF 11'400–
Want: integration with the HR software without development1010 × 10 = 100. Standard connector4 × 10 = 40. File import–
Want: lowest annual cost710 × 7 = 708 × 7 = 56–
Want: signing from a mobile phone56 × 5 = 30. App to install10 × 5 = 50. Browser–
Weighted total200146–
Decision analysis for an electronic signature tool for employment contracts. Tool C fails the screen of the musts. The weighted wants rank tool A ahead of tool B.

Examined on its own, tool A shows one adverse consequence: its connector depends on the version of the HR software. The team rates it medium in probability and high in seriousness. Examined in turn, tool B sits close to the CHF 12'000 limit: a price increase of 6% takes it out. The team chooses A and protects its rollout.

Potential problem analysis

Potential problem analysis protects an action. It often protects the implementation of a decision already taken. The adverse consequences of decision analysis help the team choose between options. This process builds an action plan for the chosen option.

The team states the action to protect. It lists the potential problems where the action is vulnerable: an activity carried out for the first time, overlapping responsibilities, a tight deadline, remote execution. The team names each problem precisely. "IT" names an area. "The connector rejects contracts after the HR software update" names a problem. The team rates each problem for probability and seriousness.

For each problem, the team looks for the likely causes. A preventive action acts on a cause to avoid the problem. A contingent action reduces the effect of a problem that no prevention has avoided. A trigger sets off the contingent action. The best trigger often sets off the action without anyone stepping in. Some triggers call for judgement: someone decides that the threshold has been reached. Every action has a cost. The team spends in proportion to probability and seriousness. When a serious problem admits no feasible preventive or contingent action, the team accepts the risk or goes back to the decision.

Potential problemLikely causePreventive actionContingent actionTrigger
Contract created in the HR software missing from the signature tool. Medium probability and high seriousnessAn update of the HR software changes the connector interfaceThe vendor tests the connector on each new version before it goes into productionHR sends the contract by file importThe daily check flags a contract with no signature request for 24 hours
Contract signed by a person without signing authority. Low probability and high seriousnessSignatory list carried over wrongly into the toolManagement approves the signatory list before the rolloutHR cancels the contract and has it signed againThe weekly check flags a signatory missing from the approved list
Potential problem analysis for the rollout of tool A. Each problem gets a preventive action against its cause, a contingent action against its effect and a trigger.

Boundaries with neighbouring techniques

Root cause analysis groups the techniques that trace an effect back to its cause. Its verification already tests a cause where the effect should appear and where it should not. The Kepner-Tregoe method makes this comparison its main instrument. It generates the possible causes from distinctions and changes, then eliminates them with the IS NOT column.

The fishbone diagram sorts possible causes by category. It delivers candidates that still wait for data. These candidates can enter problem analysis as causes drawn from experience. The test against the specification decides between them. The distinctions add a second source of candidates that the diagram does not use.

The five whys follow a single chain of causes. A fact supports each link. Problem analysis tests a cause against a set of paired facts. The cause that survives the test can open a five whys chain that traces it further back.

8D problem solving reuses the IS / IS NOT grid in D2 and compares its columns in D4. In D5 it splits its selection criteria into musts and wants. Decision analysis makes the same split. 8D makes containment a discipline (D3) ahead of the cause search. Kepner and Tregoe name the interim action without making it a step. 8D also extends prevention to neighbouring products and processes.

AI considerations

An assistant prepares the specification. From a ticket export, it groups the incidents by object, location, date and volume, then proposes the IS cells. It also proposes comparisons for the IS NOT column: the neighbouring module, the previous version, the site that was spared. The team checks every cell against the source data. Without data, an assistant invents a plausible comparison. A wrong IS NOT cell eliminates the right cause.

It cross-checks the distinctions against the change log. For each distinction, it lists the dated changes from before the first observation. It drafts possible causes as mechanisms and flags the cells that each one contradicts. The team keeps the judgement on the assumptions. Only the team knows which ones are reasonable.

In decision analysis, it proposes objectives and spots those that favour one option. The weights stay with the decision makers: they express their trade-off. In potential problem analysis, it lists the problems met in similar implementations.

An assistant rearranges the facts it is given. It observes none. Tickets and logs often contain personal data about customers or staff. The Federal Act on Data Protection (FADP) and internal rules decide which tool may receive them.

Examples

The case is fictitious. A canton forwards to its municipalities the building permit applications filed on its online portal. Since 4 February, some municipalities have been receiving applications without plans.

DimensionISIS NOTDistinction on the IS sideChange before 4 February
What: objectArchitect's plans attached to the applicationForm, land register extract and photos from the same applicationThe plans are the largest documents in the application–
What: defectDocument missing from the application received, with no error messageDocument rejected at filing with a message to the applicantThe portal accepted the document–
Where: locationApplications that the municipalities receive from the online portalApplications filed on paper at the municipalityOnly online applications pass through the exchange gatewayVersion 4.0 of the gateway installed on the evening of 3 February
Where: on the objectPlan filesDocument list: it names the plansThe portal produces the document list. The files pass through the gatewayGateway 4.0
When: first observationApplication filed on 4 February at 8:40Applications filed up to 3 FebruaryApplications forwarded by gateway 4.0Gateway 4.0
When: patternEvery day since 4 February, at any hourOnly certain hours or certain daysNone–
When: life cycleBetween filing and receipt by the municipalityAt filing: the applicant sees the plans in their account after sendingThe loss occurs during transmissionGateway 4.0
Extent: objects63 applications out of 410 filed online from 4 to 14 February, in 21 municipalitiesThe other 347 online applicationsEach of the 63 contains at least one plan larger than 25 MB. None of the 347 doesVersion 4.0 limits a document to 25 MB by default. Version 3 accepted 100 MB
Extent: sizeWhole file missingFile truncated or unreadableNone–
Extent: defects per objectEvery plan in the application larger than 25 MBPlans smaller than 25 MB in the same applicationOnly the files larger than 25 MB are missing25 MB limit
Extent: trendStable: 5 to 8 applications a dayRising or fallingNone–
IS / IS NOT specification: plans missing from the building permit applications that the online portal forwards. The last two columns record the distinctions on the IS side and the changes that touch them before 4 February.

Three rows yield no distinction. The distinctions on the other rows point to the gateway and to file size. The gateway moved to version 4.0 on the evening of 3 February. That version lowers the size limit. From this the team derives cause C in the next table.

Possible causeSourceCells contradictedAssumptions neededVerdict
A. The browser version released at the end of January interrupts the upload of large filesExperienceWhen, first observation: complete applications up to 3 February. When, life cycle: the applicant sees the plans after sendingThe portal would display files it never received. Every applicant would have switched browser on 4 FebruaryRuled out
B. The PDF/A format check required since 1 January rejects non-compliant plansExperienceWhat, defect: the check rejects at filing with a message. When, first observation: complete applications from 1 January to 3 FebruaryThe check would have changed its behaviour on 4 February without any updateRuled out
C. Gateway 4.0 drops any document larger than 25 MB without a messageDistinctions and changesNoneThe production configuration keeps the default limitMost probable cause
D. The storage space of a municipality is fullExperienceExtent, objects: 21 municipalities affected. Extent, defects per object: the small plans from the same application arrive. What, object: the form and photos from the same application arriveTwenty-one municipalities full on the same day. A full storage space that still accepts small filesRuled out
Test of the possible causes against the specification. Three causes drawn from experience each fail on an IS NOT cell. The cause drawn from distinctions and changes explains every cell.

The team confirms C in a test environment. A 30 MB plan disappears with the default limit and arrives once the limit is raised to 100 MB.

Visualisations

The flow diagram shows the two sources of possible causes and the test of each cause against both columns. Two loops lead back to the specification: when no cause passes the test and when confirmation fails. A second diagram compares two deviations. The first follows a change that the search targets. The second has existed since day one. There the search rests on the distinctions.

Cost

DimensionLevelJustification
PreparationMediumGather the facts about the deviation and find the closest comparison for each IS answer. AHRQ judges the grid easy to learn but slow to fill in.
ExecutionMediumOne session for the distinctions, the changes and the test. Then a confirmation action, often a trial.
DocumentationLowOne page for the specification, the test and the confirmation. Decision analysis and potential problem analysis each add one table.

Tooling

A spreadsheet or a whiteboard holds the specification: one row per question and one column each for IS, IS NOT, distinctions and changes. A second tab holds the test of the causes. The same spreadsheet computes the weighted scores of decision analysis.

In an IT department, the problem record in the IT service management tool holds the specification. The change log of that tool supplies the dated changes. Application logs and monitoring data fill the fact columns.

A distributed team gathers its facts before the session in a shared document or an online whiteboard.

Sources

  • Kepner, Charles H. and Tregoe, Benjamin B., The New Rational Manager, updated edition, Princeton Research Press, ISBN 978-0-9715627-1-4: ch. 2 for problem analysis (pp. 21-47), ch. 3 for abbreviated use, the Day One Deviation and teamwork (pp. 50-66), ch. 4 for decision analysis (pp. 77-100), ch. 6 for potential problem analysis (pp. 139-166), ch. 7 for the entry criterion and the cause with no effect on action (pp. 176, 184-185). The first version of the book appeared in 1965 under the title The Rational Manager. The method is the two authors' own. Their company teaches it.
  • Agency for Healthcare Research and Quality (AHRQ), Kepner-Tregoe Matrix, Workflow Assessment for Health IT Toolkit: the IS / IS NOT grid applied to health IT and the effort it demands.
Kano Analysis
All techniques
Kick-off meeting