Change Deployment Strategies
Change deployment strategies set how an organisation moves its users to a new system or a new way of working. The common forms are the direct cutover on a fixed date, parallel running of the old and new systems, the pilot deployment with a small group and the phased deployment. HERMES has the project choose the strategy in the deployment concept. Its traditional approach places that concept in the concept phase. The same concept covers user training, support during the transition and the emergency measures that apply if the go-live fails.
Goal
A deployment strategy is the way an organisation puts a new solution into use. It decides who moves to the new system, in what order and on what date. It plans what happens if the move fails.
The choice weighs the risk of a failed go-live against the cost and duration of the transition. It shapes the data migration, the training schedule and the support set-up. The project therefore makes the choice while these three plans can still follow it. The traditional HERMES approach places it in the concept phase. In the agile HERMES approach, the project draws up the concept in the execution phase. It then prepares and executes the deployment measures release by release.
The deliverable is the deployment concept. HERMES 2022 sets out its content:
- the background and the impact analysis;
- the deployment procedure: the chosen strategy;
- the deployment measures: organisation management, transition management, change management, emergency measures and the emergency organisation;
- the training concept;
- the deployment organisation and the deployment planning;
- the planning of preliminary acceptance and acceptance with their criteria;
- the go-live release criteria.
In HERMES, the user representative is responsible for this outcome. The project manager and the business analyst contribute to it. For the business analyst, the concept answers the transition requirements. BABOK defines them as the capabilities the solution must have and the conditions it must meet for the duration of the move from the current state to the future state. They cover data conversion, training and business continuity.
Usage
When to use
- Replacing a system in operation: decide how users leave the old system and on what date it stops.
- New process or new organisation: with or without a new tool, set the order in which teams or sites move and plan their training.
- Concept phase of a HERMES project under the traditional approach: the deployment concept is an outcome of the phase.
- A migration concept to write: the deployment strategy sets the data migration strategy.
- Business that cannot stop: billing, payroll or a service counter; plan the rollback before setting the date.
- Large, dispersed or multilingual user base: plan training by audience, by location and by wave.
When not to use
- Small change delivered as a single release or continuously into a live system: no group of users moves from one system to another; handle it through change control.
- Culture change with no cutover date: lead the transformation with Kotter's change process.
Description
The strategies
HERMES asks the project to choose between deployment on a fixed date and phased deployment. David T. Bourgeois's textbook Information Systems for Business and Beyond adds the pilot deployment and parallel running.
Direct cutover (also called big bang) stops the old system on a set date. From that day, all users work on the new one. Bourgeois rates it the fastest, the cheapest and the riskiest: a system defect or poorly prepared users hit the whole organisation on the same day.
Parallel running keeps the old and new systems in operation together for a limited period. If the new system fails, work continues on the old one. Bourgeois rates it the least risky and the most expensive: staff do the work twice and both systems need full support. The systems development guidance of the U.S. Department of Justice (DOJ) uses it to verify functionality and data integrity until the new system is certified.
Pilot deployment gives the new system to a pilot group before the rest of the organisation. A defect affects only that group. The support team concentrates on a few people. The rest of the organisation follows in one go or in waves.
Phased deployment puts the functions of the new system into service one after another. Each function of the old system stops when its replacement starts.
In HERMES terms, a pilot followed by waves and a deployment by function are two forms of phased deployment. Parallel running combines with a fixed-date cutover or with each phase. The pilot group can keep the old system in parallel for its first weeks.
| Strategy | What moves | Risk | Cost | Training | Rollback |
|---|---|---|---|---|---|
| Direct cutover | All users and all functions on one date | Highest | Lowest; shortest lead time | One wave that ends before the date | Full rollback plan: the old system no longer runs |
| Parallel running | All users; the old system stays in service for a limited period | Lowest | Highest: double data entry and support for both systems | One wave before the parallel run starts | Stop the new system; work continues on the old one |
| Pilot deployment | A pilot group, then the rest of the organisation | Limited to the pilot group | At least two deployments to organise | One wave for the pilot, then one per group | The pilot group returns to the old system, which still runs for the others |
| Phased deployment | One function after another | Limited to the function that moves | Temporary interfaces between the old and new systems | One wave per function | Function by function |
Software development teams use similar words for something else. In the canary release that Martin Fowler describes, the team first routes a small share of users to the new version of a piece of software, then adds more. If a problem appears, the team sends the users back to the old version. A canary release governs how a software version goes into production. A pilot governs how people move to a new way of working. It needs training, support and a decision before the organisation extends the change.
The steps
- Analyse the impact. Establish whom the change affects, in which processes, at which sites and at what volumes. The stakeholder list is the starting point.
- Choose the strategy. Weigh the deployment risks, the complexity of the systems and their importance, then select one strategy or a combination.
- Align the migration. Make the migration concept follow the split and the schedule of the deployment.
- Organise support. Define the roles that support the deployment: superusers, on-call support and the contact person in operations.
- Design the training. Identify the training needs of users and operators. Write the training concept.
- Set the criteria and the emergency measures. Write the acceptance criteria, the go-live release criteria and the rollback plan.
- Coordinate and plan. Have the stakeholders validate the concept. Integrate the deployment planning into the project management plan.
- Implement the measures. Produce the training materials, train the superusers and put the emergency measures in place. The preliminary acceptance decision checks them.
- Execute. Train the users. Have the sponsor decide on the go-live. Move the pilot group, the first phase or the whole organisation. Check the result. In a pilot or a parallel run, then decide on the rollout to the rest of the organisation or on the end of the parallel run.
In the traditional HERMES approach, steps 1 to 7 fall in the concept phase, step 8 in the implementation phase and step 9 in the deployment phase.
Choosing the strategy
HERMES asks the project to draw up the deployment concept with the deployment risks in view. Bourgeois makes the choice depend on the complexity and importance of the old and new systems. The analyst turns these two criteria into questions for the user representative and for operations.
- What interruption can the business tolerate? If it tolerates none, a direct cutover requires a proven rollback. Without one, parallel running is the only option.
- Can the old and new systems run together? Parallel running and phased deployment need consistent data across both systems. Temporary interfaces often keep them in sync.
- Can the functions be separated? A deployment by function assumes that a migrated function does not depend on a function left in the old system.
- Is there a representative pilot group? It must handle the same cases, the same volumes and the same exceptions as the rest of the organisation.
- Can the organisation do the work twice? Parallel running doubles data entry for its whole duration.
- How many people must be trained by how many trainers? Training capacity sets the size of the waves.
- Is a date imposed from outside? A law coming into force, the end of the old system's contract or the start of a financial year push towards a direct cutover.
A survey by Johansson and Sudzina of Danish, Slovak and Slovenian companies that introduced an ERP system finds no significant link between the method chosen and company size, growth, country, the existence of an IT strategy or a CIO on the management board. The authors group the companies by size, by country, by whether they have an IT strategy and by whether they have a CIO. In each of these groups, between 44% and 56% of the companies chose direct cutover.
When the organisation buys the solution, the project and the supplier draw up the concept after the purchase. The supplier brings experience from similar projects. The contract already sets the acceptance criteria.
The pilot chosen because it volunteered. The most motivated and best-equipped team volunteers. Its results say nothing about the rest of the organisation. The project chooses the pilot because it is representative. Before the pilot starts, the project writes down what it must demonstrate.
The parallel run with no end date. Users keep treating the old system as the reference. The new one never takes over that role. The concept sets the end date, the criterion that must be met before the parallel run ends and the person who settles a discrepancy between the two systems.
The direct cutover with no prepared rollback. On go-live day, no old system runs alongside. An improvised rollback that day runs into the data entered since the cutover: the old system does not hold them.
The strategy chosen as the go-live approaches. By then the migration, the training and the operations contracts are already planned for another strategy. That is why the traditional HERMES approach places the choice in the concept phase.
Aligning the data migration
According to HERMES, the strategy set in the deployment concept determines the migration strategy: a phased deployment calls for a phased migration. The migration concept in turn checks its impact on the deployment concept. If the data cannot be split by group or by function, the migration happens in one go. The deployment then follows on a fixed date.
A pilot migrates only the data of its own group. The project team then decides which system handles the cases that the pilot group shares with the others. In a parallel run, the same data exist in two databases. During the parallel cycle, the DOJ implementation plan asks the team to check that the two systems hold consistent data. Data mapping establishes the field correspondence that these checks need.
Training the users
HERMES asks the project to identify the training needs of users and operators and to record the training measures in the deployment concept. There the training concept covers the requirements, the training procedure, the training materials and the training infrastructure. The DOJ training plan details what each part must settle.
| Part | What it sets |
|---|---|
| Requirements | The audiences by role and by location, the tasks each audience must perform on day one and the skills these tasks require. |
| Training procedure | The method per audience, the schedule per wave and per location, then for each course: duration, group size, audience, objectives, content and prerequisites. |
| Training materials | Course materials, trainer and participant guides, exercises and quick-reference cards for the workstation. |
| Training infrastructure | Rooms, workstations, the training environment and its fictitious database with the practice cases. |
| Evaluation | The questionnaire for participants, the measurement of effectiveness and the revision of the courses. |
| Follow-up | Course updates, refresher training and the procedure for requesting additional training. |
HERMES asks the project to check the effectiveness of the measures carried out. Quality assurance uses the participants' course evaluations as evidence.
The DOJ guidance treats training users before their first use of the new system as good practice. In its view, a lack of budget most often prevents the team from carrying out the training plan. The training budget therefore goes into the deployment concept from the moment the team starts writing it.
The strategy sets the rhythm of the training. A direct cutover trains everyone before a single date and concentrates the trainers' workload in a few weeks. A pilot or a phased deployment trains in waves. The project schedules each session close to the cutover of its audience, because users forget what they learned when their first real use comes late.
At an Israeli financial institution, Kaplan-Mor, Glezer and Zviran compared lecture-based training, an online simulator, mentoring and a hybrid method during the go-live of a system. They measured the calls to the help desk after each session. Lecture and simulator show no significant difference. The hybrid method gives the best result.
At a Dutch social insurance agency that was introducing a process management system, de Waal and Batenburg surveyed 143 users and conducted 49 interviews. User attitude significantly moderates the link between training and implementation success. Training does not make up for a rejection of the new system. The communication plan prepares the ground before the sessions.
HERMES cites superusers as an example of the deployment organisation. The project trains them in the implementation phase. They support their colleagues during deployment. In the pilot group they meet the real cases before they help the following waves.
Training the project team belongs to the resource planning of the project.
Support, go-live and rollback
The deployment concept defines the deployment organisation: who answers users, how a problem escalates to operations, who decides on a rollback.
In HERMES, the sponsor decides on the go-live. The decision rests on preliminary acceptance, migration acceptance, the deployment measures carried out and the release criteria set in the concept. Before the decision, the project evaluates these criteria and presents the deployment risks. The acceptance criteria of the solution feed into this evaluation.
In the DOJ implementation plan, the rollback plan (back-off plan) sets the moment of the go/no-go decision and the factors it rests on. This plan also lists the actions that return the site to its original state. The implementation plan describes how to verify whether the implementation succeeded. If it did not, the rollback plan applies.
In a direct cutover, the old system has stopped. The data entered since the cutover exist only in the new one. For this case, the business analyst sets a point of no return in the concept: the date after which fixing the new system costs less than returning to the old one. The risk register holds this risk and its response.
AI considerations
A language model speeds up the writing of training materials. From process descriptions and use cases, it drafts first versions of quick-reference cards, exercises per role and an FAQ for the superusers. It generates the fictitious cases of the training database without real customer data. It translates a document into the languages that the organisation uses. A colleague from the business reviews each version in their own language.
During a pilot, it groups support tickets by theme and flags recurring problems. The committee that decides on the rollout to the rest of the organisation receives a weekly report of problems by theme. The model also proposes a first list of rollback actions from the operations manual.
It does not choose the strategy. The choice rests on how much interruption the business can tolerate, on the real state of the data and on the confidence of the business side. No document holds all of this. A named person judges whether a pilot group is representative and takes the go/no-go decision. Support tickets and training data may contain personal data within the meaning of the Federal Act on Data Protection (FADP). Before handing them to a hosted service, the organisation checks the data processing agreement and the country where the data are processed.
Examples
A health insurer in French-speaking Switzerland is replacing its application for processing invoices from healthcare providers. Its 140 claims handlers work at three sites. The concept selects a pilot followed by one wave per site. The deployment plan sets a criterion that each wave must meet before the next one starts.
| Wave | Audience | Training | Cutover | Criterion for the next wave | Rollback |
|---|---|---|---|---|---|
| Pilot | Lausanne, team B: 12 claims handlers who process all invoice types. Four of them become superusers. | Week 9: 2 classroom days on the training database | Week 10 | Average processing time at most 110% of the old one; no invoice blocked for more than 5 days; payment errors at most 0.5% | Open invoices taken back into the old application, which still runs for the other teams |
| 1 | Lausanne, other teams: 58 claims handlers | Weeks 14 and 15: 2 classroom days co-led by the superusers | Week 16 | Same criteria; at most 2 help-desk calls per handler per day | Each team returns to the old application as in the pilot |
| 2 | Sion: 40 claims handlers | Week 20: 1 classroom day and 1 day at the workstation with a superuser | Week 21 | Same criteria | Each team returns to the old application as in the pilot |
| 3 | Neuchâtel: 30 claims handlers | Week 25: 1 classroom day and 1 day at the workstation with a superuser | Week 26 | Same criteria at all three sites: decision to decommission the old application | Full rollback plan until decommissioning in week 30 |
The last column shows where the rollback changes in nature. As long as a site still works in the old application, a rollback affects only one team. After wave 3 nobody uses it any more: a rollback becomes a cutover in reverse. The insurer therefore keeps the old application in service for four weeks after the last wave. It decommissions the application once all three sites meet the criteria.
Visualisations
In front of the steering committee, the analyst compares the candidate strategies on a shared time axis. Each band shows the share of the work handled in the new system and the lifetime of the old one. The deployment plan takes the form of a table with one row per wave. The steering committee updates it from one wave to the next.
Cost
| Dimension | Level | Rationale |
|---|---|---|
| Preparation | Medium | Impact analysis, selection questions put to the business and to operations, alignment with the migration concept. |
| Execution | High | Training for all users, extra support during each cutover, double data entry during a parallel run. |
| Documentation | Medium | Deployment concept, training concept and rollback plan. The team updates them after each wave. |
Tooling
HERMES publishes a Word template that follows the content of the deployment concept. A spreadsheet is enough for the deployment plan and for tracking the wave criteria. A project planning tool can place the waves in a Gantt chart beside the rest of the schedule.
An e-learning platform manages registrations per wave, remote modules and evaluation questionnaires. The training environment is a copy of the new system fed by a fictitious database. The team prepares it together with the test environment and resets it before each session. During the pilot and the waves, the help-desk ticketing tool supplies the measurements for the wave criteria if each ticket records the wave and the site.
Sources
- Swiss Confederation, Federal Chancellery, HERMES 2022: the task Design deployment concept, for the choice between a fixed date and phases, the roles and the activities; the outcome Deployment concept, for its content and the training concept; the task Design migration concept, for the link between deployment strategy and migration strategy; the task Realize deployment measures, for superusers and emergency measures; the outcome Deployment measures carried out, for the course evaluations; the task Decide on launch of operation, for the sponsor's decision and its criteria.
- Bourgeois, D. T., Information Systems for Business and Beyond, Saylor Foundation, section 10.10 Project Implementation (LibreTexts copy, CC BY-SA licence): the four methods, the risk and cost of direct cutover and parallel running, then the selection criterion.
- U.S. Department of Justice, Systems Development Life Cycle Guidance Document (2003): chapter 10, for parallel operation and training before use; appendix C-20, for the parallel cycle, the back-off plan and the post-implementation verification; appendix C-24, for the training plan.
- IIBA, A Guide to the Business Analysis Body of Knowledge (BABOK Guide) v3, §2.3: the definition of transition requirements.
- Kaplan-Mor, N., Glezer, C., Zviran, M. (2011), "A comparative analysis of end-user training methods", Journal of Systems and Information Technology, 13(1), 25-42: the comparison of four training methods.
- de Waal, B., Batenburg, R. (2012), "What makes end-user training successful?", International Journal of Knowledge and Learning, 8(1/2), 166-183: the role of user attitude.
- Johansson, B., Sudzina, F. (2008), "ERP Systems Implementation: Factors Influencing Selection of a Specific Approach?", 3gERP workshop, Copenhagen Business School: the factors linked to how a company chooses to introduce an ERP system.
- Fowler, M., Canary Release, martinfowler.com: the progressive rollout of a software version.

