Process Redesign Patterns
Process redesign patterns are standard transformations that the analyst applies to the model of an existing process to derive a target process: remove a task, run two tasks in parallel, give one person the whole case, automate a routine task. The catalogue by Hajo Reijers and Selma Limam Mansar (2005) lists 29 of them. It sorts them by the process element they change: the customers, the tasks, their order, the organisation, the information, the technology and the external environment. Each pattern affects four criteria: time, cost, quality and flexibility. A gain on one is often paid for on another.
Goal
A redesign pattern is a proven type of change that the analyst applies to a process model to reach a quantified target: remove a task, merge two tasks, change their order. It answers a design question: which change to make to this process to reach the target. The analyst starts from a modelled process and a target, for example confirming an order on the day it arrives. The analyst works through the catalogue, keeps the patterns that apply, estimates the effect of each and combines the best into a target process.
Limam Mansar and Reijers present their framework as a set of ideas for thinking about a process while redesigning it. The pattern proposes a type of change. The analyst decides whether and where to apply it.
The deliverable is the target process model and a table of the selected patterns. Each row describes a change and its expected effect on time, cost, quality and flexibility. The layout of this table is our proposal: the sources prescribe none.
Process analysis locates the gaps in a process and their causes; redesign patterns supply the transformations that correct them.
Usage
When to use it
- Quantified target on an existing process: a lead time, processing cost or error rate to reach; produce the options that lead there.
- After a process analysis: the gaps and their causes are known; the transformations remain to be chosen.
- Before introducing a workflow tool: rework the process instead of automating the current one as it stands.
- Case that passes through several departments: each hand-off adds waiting time; the organisation patterns reduce the number of people involved.
- High volume of similar requests: a pattern's gain scales with the number of cases handled.
- Preparing a design workshop: the list of applicable patterns becomes the agenda.
When not to use it
- Process neither modelled nor measured: the analyst cannot estimate any effect; start with process modelling and a record of volumes and durations.
- Cause of the problem unknown: a pattern chosen without a diagnosis treats a symptom; look for the cause with a root cause analysis.
Description
A standard change
A pattern describes a change that does not depend on the domain. "Run tasks in parallel where possible" applies to a loan application as much as to an order. Reijers and Limam Mansar speak of best practices. The title of their 2005 article also calls them redesign heuristics.
The framework: the process element each pattern changes
Limam Mansar and Reijers sort the patterns by the process element they change:
- internal or external customers;
- the products or services that the process delivers;
- the process in two views: the operation view says how it runs (number, size and nature of the tasks), the behaviour view says when (task order, grouping, scheduling);
- the participants in two views: the structure (roles, groups, departments) and the population (the people);
- the information that the process uses or creates;
- the technology that it uses;
- the external environment other than customers.
The analyst reviews each element in turn. In their two case studies, Limam Mansar and Reijers found every element of the framework useful for redesign. No pattern in the catalogue addresses the products.
The catalogue
Limam Mansar and Reijers define 29 patterns. They rank the patterns used most often in the three simulated processes of their two case studies from 1 to 10. When two patterns tie or nearly tie, the authors break the tie by how often the literature cites each one. A survey of Dutch and British practitioners puts the same three patterns at the top but lifts parallelism to fourth place. According to the authors, their two cases did not lend themselves to it.
| Element | Pattern | Definition | Rank |
|---|---|---|---|
| Customers | Control relocation | Move controls to the customer | |
| Contact reduction | Reduce the number of contacts with customers and third parties | ||
| Integration | Integrate the process with that of a customer or supplier | 8 | |
| Operation view | Order types | Check whether the tasks relate to the same type of order and create a separate process per type where needed | |
| Task elimination | Remove unnecessary tasks | 1 | |
| Order-based work | Remove batch processing and periodic activities | ||
| Triage | Split a general task into tasks specific to each case or merge alternative tasks into one general task | ||
| Task composition | Combine small tasks into a composite task and divide large ones into workable tasks | 2 | |
| Behaviour view | Resequencing | Move tasks to a more appropriate place in the process | 6 |
| Knock-out | Put the checks that take the least effort and reject most often first | ||
| Parallelism | Run tasks in parallel where possible | 9 | |
| Exception | Design the process for typical orders and take exceptional cases out of the normal flow | ||
| External environment | Trusted party | Use the result of a trusted party instead of producing the information yourself | |
| Outsourcing | Hand all or part of the process to an outside party | ||
| Interfacing | Set up a standard interface with customers and partners | ||
| Organisation: structure | Order assignment | Let one employee perform as many steps of an order as possible | 5 |
| Flexible assignment | Assign resources so as to keep the most room for manoeuvre later | ||
| Centralisation | Treat geographically dispersed resources as if they were centralised | ||
| Split responsibilities | Give responsibility for a task to a single department | ||
| Customer teams | Form cross-department teams that handle one type of order end to end | ||
| Numerical involvement | Reduce the number of departments, groups and people involved in the process | 10 | |
| Case manager | Appoint one person to handle each type of order | ||
| Organisation: population | Extra resources | Increase the number of resources when capacity falls short | |
| Specialist-generalist | Make resources more specialised or more versatile | 7 | |
| Empower | Give employees most of the decision-making authority and reduce middle management | 4 | |
| Information | Control addition | Check that inputs are complete and correct and check outputs before they go to the customer | |
| Buffering | Subscribe to updates from an external source instead of requesting the information again | ||
| Technology | Task automation | Automate tasks | |
| Integral technology | Remove the physical constraints of the process with new technology | 3 |
Each transformation a team raises in a workshop matches a named pattern. Removing hand-offs between departments falls under case manager, order assignment and split responsibilities. Running tasks side by side falls under parallelism. Automating routine work falls under task automation. Checking earlier falls under resequencing or knock-out.
The trade-offs of each pattern: the devil's quadrangle
Limam Mansar and Reijers assess each pattern on the four criteria of Brand and Van der Kolk's devil's quadrangle: time, cost, quality and flexibility. The name states the constraint: pulling one corner outward often pulls another inward. The analyst therefore looks for each pattern's trade-off before selecting it.
Two patterns applied to the same tasks can pull in opposite directions. Take an order that passes two manual checks. The warehouse checks stock in 20 minutes after 3 hours of waiting and rejects 2% of orders. Finance checks the customer's credit limit in 5 minutes after 4 hours of waiting and rejects 10% of orders. With stock first, the two checks take 7 h 25 min per order. For 100 orders they cost 100 × 20 + 98 × 5 = 2490 minutes of work.
Parallelism starts both checks at the same time. The order waits for the longer of the two: 4 h 05 min. Work rises to 100 × 20 + 100 × 5 = 2500 minutes, since finance also checks the orders that stock rejects. The warehouse also reserves stock for about ten orders that finance will refuse, then releases those reservations.
Knock-out puts the check that costs least and rejects most at the front. The credit check therefore comes before stock. Lead time stays at 7 h 25 min. The warehouse checks the 90 orders that finance accepts: 100 × 5 + 90 × 20 = 2300 minutes.
Parallelism saves 3 h 20 min per order over knock-out. It costs 200 more minutes per 100 orders and leaves reservations to release. The target decides: a time target selects parallelism, a cost target selects knock-out.
Improvement or radical redesign
Michael Hammer set out the principles of process reengineering in the Harvard Business Review in 1990. His title sums up the thesis: Don't Automate, Obliterate. Information technology should no longer speed up existing processes. The company wipes them out and redesigns them from the expected result.
At Ford in the early 1980s, the accounts payable department employed more than 500 people in North America. The first plan simplified procedures and installed new systems to cut headcount by 20%. Ford chose instead to change the rule of the process: the company no longer pays on receipt of the invoice but on receipt of the goods. Receiving compares the delivery with the order in a shared database. Payment goes out without an invoice. Hammer reports a 75% headcount reduction where the new process applies. He chose this case to argue his thesis.
Hammer draws seven design principles from his cases. Several overlap with patterns in the catalogue.
| Hammer's principle | Closest pattern in the catalogue |
|---|---|
| Organize around outcomes, not tasks | Order assignment, case manager |
| Have those who use the output of the process perform the process | No direct equivalent |
| Subsume information-processing work into the real work that produces the information | Task composition |
| Treat geographically dispersed resources as though they were centralized | Centralisation (almost identical wording) |
| Link parallel activities instead of integrating their results | Parallelism; Hammer adds coordination during the work |
| Put the decision point where the work is performed, and build control into the process | Empower |
| Capture information once and at the source | Task elimination (re-keying), interfacing |
The two approaches share these moves. They differ in scope and in risk. A pattern changes part of an existing process. The analyst estimates its effect before implementation. Radical redesign starts from the expected result and challenges the implicit rules of the process (at Ford, "we pay when we receive the invoice"). It changes jobs, organisation and management systems at the same time. Hammer concedes that reengineering unsettles people and overturns everyone's habits. A redesign by patterns also carries an organisational risk. A Dutch municipality studied by Limam Mansar and Reijers decided not to implement the patterns it had selected. It judged this risk too high for the gains.
Running a redesign with patterns
No source sets out a numbered procedure. These steps are our synthesis of the approach Limam Mansar and Reijers followed in their case studies.
- Set the target and its criterion. Choose time, cost, quality or flexibility and give it a target value. In the case studies the municipality aimed at labour cost and the public works department at average lead time. The two organisations did not select the same patterns. The analyst justifies each selected pattern by its effect on the target criterion.
- Start from the current model. Record the process as it runs: volumes, task durations, rejection rates of the checks and resources. The analyst draws the model with process modelling, for example in BPMN.
- Review the catalogue element by element. For each pattern, note whether it applies and where. Limam Mansar and Reijers present their results in a three-column grid per process: applicable, significant effect, selected.
- Estimate the effect of each pattern on its own. The authors built one simulation model per pattern. They fed it with the real data of the process and with expert estimates of the pattern's local effect. For a lower-volume process, two hand calculations are enough. Lead time adds up the work and the waits along the path; of two parallel branches, only the longer counts. The expected effort of a sequence of checks adds up the effort of each check multiplied by the share of requests that reach it.
- Combine the patterns. Each row of the table names the tasks it changes. Two rows on the same tasks flag patterns that may exclude each other; the target decides between them. Keep the subset that gives the best combined result. The authors chose it in two workshops with users, managers and IT staff.
- Check the four criteria. Record the effect of each selected pattern on each criterion. Enter each accepted loss on a criterion in the table with the name of the person who accepts it.
- Consider radical redesign if no combination reaches the target. Start from the trigger and the expected result. Name the implicit rules of the process and ask of each one why it exists. Hammer wants a team that keeps asking "Why?" and "What if?".
Pitfalls
Automating the current process. Putting a workflow tool on the current process applies only one pattern: integral technology. For the invoices of the public works department, the tool alone cut the expected average lead time by 25%. The redesign with the other selected patterns cut it by almost 80%. Three months after go-live, the measured performance matched the estimate almost exactly.
Forgetting the organisational cost. The case manager and empower patterns change jobs, job classifications and reporting lines. At Mutual Benefit Life, the job evaluation scheme could not classify the case manager position. The position carried a lot of responsibility without a single subordinate. The effect estimate does not see this cost. The workshop puts a figure on it or at least names it.
Removing a check required elsewhere. A check that looks useless in the flow sometimes serves the internal control system (ICS) or a legal obligation. The task elimination and empower patterns then remove a four-eyes principle. The order assignment and case manager patterns concentrate the steps on one person and collide with segregation of duties. A UK practitioner interviewed by Limam Mansar and Reijers rules out order assignment for this reason. Before removing a check, the analyst asks its owner why it exists.
Transferring the figures from another case. Ford's 75% and the 80% of the public works department describe one process, one organisation and one period. Two public administrations and three simulated processes do not make a general law. The analyst re-estimates the effect on the process at hand.
AI considerations
A language model receives the list of 29 patterns and the description of the current process in text or in BPMN. It works through the list and proposes a place to apply each pattern and a reason. This first pass gives the workshop its agenda. The analyst checks each proposal against what the model does not know: legal controls, collective labour agreements, the limits of the existing system. The analyst discards the proposals that conflict with them.
Process mining reconstructs the actual process from the event logs of the ERP or the workflow tool. It supplies the figures the estimates need: the duration of each task, the wait between two tasks, the rejection rate of each check. Without them the analyst can assess neither knock-out nor parallelism.
AI does not set the target and does not settle the trade-offs. Accepting a loss of flexibility to gain time is a management decision. Event logs often name the person who performed each task. These are personal data under the FADP (art. 5). Handing them to an external AI service makes that service a processor (art. 9). If the service is hosted outside Switzerland, sending the logs also falls under the rules on disclosure of personal data abroad (art. 16). Pseudonymised identifiers are enough to estimate durations.
Examples
A sanitary-ware wholesaler confirms an order in a day and a half on average. It wants to confirm each order on the day it arrives. Each criterion column shows what the pattern gains or costs.
| Pattern | Change | Time | Cost | Quality | Flexibility |
|---|---|---|---|---|---|
| Task elimination | The order that the customer enters on the portal arrives in the ERP. The re-keying in the back office disappears. | + | + | + fewer keying errors | = |
| Task automation | The ERP checks the credit limit. | + | + | = | − a negotiated condition needs a new parameter |
| Knock-out | The automatic credit check comes before stock. | = the check takes a few seconds | + no stock check on a refused order | = | = |
| Exception | Only orders over the credit limit go to finance. | + normal flow − orders over the limit | = | = | = |
| Case manager | One person follows the order from receipt to confirmation and answers the customer. | + no hand-off in the normal flow | − training in credit and exceptions | + a single contact | − absences to cover |
Parallelism is missing from the table. Once the credit check is automatic, putting it first no longer adds time. Parallelism would gain a few seconds at most and would leave stock reservations to release for each refused order. Task automation has removed knock-out's effect on time.
The before-and-after model places the same patterns on the order flow.
Visualisations
The table of selected patterns has one row per pattern and one column per criterion of the devil's quadrangle. The before-and-after model shows the current process and the target process one above the other. Each change carries the name of its pattern. The quadrangle sets the four criteria on four axes. It shows how two patterns applied to the same tasks deform the figure in opposite directions.
Cost
| Phase | Level | Justification |
|---|---|---|
| Preparation | Medium | Build or verify the current model. Record volumes, durations and rejection rates. |
| Execution | Medium | Catalogue review and estimate of each applicable pattern in one or two workshops. High if each pattern is simulated. |
| Documentation | Low | One target model and one table of selected patterns. |
Tooling
A spreadsheet holds the catalogue review grid and the table of selected patterns. A BPMN modelling tool draws the current process and the target process. Bizagi Modeler and SAP Signavio also simulate a model from durations, resources and branching probabilities. A process mining tool such as Celonis, Disco or ProM extracts durations and rejection rates from the ERP logs. For a first pass in a workshop, the team sorts one printed card per pattern into three piles: applies, to discuss, out of scope.
Sources
- Hajo A. Reijers and Selma Limam Mansar, Best practices in business process redesign: an overview and qualitative evaluation of successful redesign heuristics, Omega 33(4), 2005, pp. 283-306: the origin of the catalogue of 29 patterns and their assessment on time, cost, quality and flexibility.
- Selma Limam Mansar and Hajo A. Reijers, Best practices in business process redesign: validation of a redesign framework, Computers in Industry 56, 2005, pp. 457-471: the classification framework, the names and definitions of the 29 patterns (table 1), the devil's quadrangle taken from Brand and Van der Kolk, the case studies of two Dutch public administrations, the ranking of the ten most frequent patterns and the survey of Dutch and British practitioners.
- Michael Hammer, Reengineering Work: Don't Automate, Obliterate, Harvard Business Review 68(4), July-August 1990: the reengineering thesis, the Ford and Mutual Benefit Life cases and the seven principles.
- Swiss Confederation, Federal Act on Data Protection (FADP), SR 235.1: the definition of personal data (art. 5), processing by a processor (art. 9) and disclosure of personal data abroad (art. 16).