Redesign a product people already use in slices: record what users rely on, test with current users, release in stages with a way back and a written pass rule. To redesign a live SaaS product without losing users, change it in slices rather than in one switch.
To redesign a live SaaS product without losing users, change it in slices rather than in one switch. List what current users have learned and built around the product, decide per capability whether to keep, adapt, replace or leave it alone, test with current users as well as new ones, then release to a small group first with a way back and a written pass rule.
The pressure arrives from two sides at once. Sales says the product looks dated in demos, support warns that long-time customers will be furious if anything moves, and both can be right. Selective change is how you satisfy the first without proving the second.
This guide is about software people use to do their jobs and that you intend to keep running while you change it. It is not about a first version of a product, and it is not about a rebuild on a new foundation. For the second case, start with preserve, redesign, replatform or retire, which decides what happens to each capability before anyone picks a technical route.
Why existing users are the risk, not the design
A redesign is judged by people who already know how the old product works. That is a central difficulty.
Two pieces of research describe the cost to them. In a study of new system acceptance in MIS Quarterly, Polites and Karahanna reported in 2012 that habitual use of the incumbent system, perceived transition costs and perceived sunk costs build inertia. That inertia lowered how easy and how advantageous people judged the new system to be, and it lowered their intention to use it. A 2022 laboratory study from HEC Montréal, Wang and colleagues, had current users of a mobile banking app do tasks on the existing and an updated version. Where the update disrupted familiar steps, they showed greater cognitive load, perceived effort and task time, and judged navigation worse for familiar tasks.
Treat that second study as a neighbouring result, not direct proof. It is one lab experiment, in consumer banking, reported as a conference paper. Our inference is that the same mechanism applies, and likely more strongly, to a work product where customers have also built saved views, reports, integrations, shortcuts and internal procedures around the current behaviour. We have not seen a study that measures that directly.
The practical consequence is a change of question. Do not ask whether the new design is better in a vacuum. Ask whether it is better by enough to be worth what each existing user must relearn, capability by capability.
Expect a reaction even to a good change. Google researchers Aaron Sedley and Hendrik Müller, writing about the move from Google Docs List to Google Drive, describe change aversion as "a natural response, which technology often exacerbates". A loud first week is neither proof that the design is wrong nor a reason to ignore what people say; set it beside what they do.
Decide how deep the redesign goes
"Redesign" covers very different amounts of change. Match the depth to what is actually wrong, and leave alone what is not.
| What you observe | Depth of change | What it changes | What it does not mean |
|---|---|---|---|
| The structure is sound but hard to see | Repair | Bounded interaction and interface fixes | The product model has changed |
| Tasks and exceptions do not match how people work | Workflow redesign | Sequence, roles, states | That rearranging screens fixes an authority gap |
| The objects and relationships are out of date | Product-model redesign | Meaning, data, interaction | That the change can stay in design files |
| The current foundations cannot support the intended behaviour | Progressive modernisation or selective rebuild | Behaviour and foundations, in stages | That the whole product must be replaced |
| The platform itself is the constraint | Replatforming | Behaviour and migration | That platform defaults should decide the product |
| An area is stable and low-risk | Leave unchanged | Nothing | That every area needs a visible redesign |
This is the scale we at Tcules use on our product redesign page. The last row matters for retention. Every screen you do not touch is a screen no user has to relearn.
A way to find the right row is to follow the work before redrawing anything. Repeated data entry can come from disconnected records. A crowded dashboard can be compensating for unclear task status. A spreadsheet export can be standing in for a comparison the product cannot show. If the cause is structural, a visual refresh will keep the fragmented behaviour and add relearning on top.
A visual refresh, meaning colour, type, spacing and icons, sits below the first row of the table. It fits when people can do their work and the problem is first impressions in a sales demo. It is also the easiest depth to approve, which is its risk: applied across the whole product, it changes the least of what daily users care about. The signs that point to a workflow redesign instead are support tickets asking where something went, onboarding calls that walk every new customer round the same obstacle, and sales demos that skip a module.
Then pick the first slice by consequence. Start where customers make poor choices or lose the most time, and where other areas depend on the same patterns, so the rest can follow the rules the first slice sets. Before designing it, record task time and errors for existing users separately from new ones, support contacts by topic and use of the area. Without that starting point, every later number is an opinion.
Write down what users have built around the product
Before design starts, record what existing users depend on, one capability at a time. At Tcules, we call this a continuity ledger. In our view, a plain spreadsheet is enough. For each capability, write down five things: what existing users have learned, what is failing for them today, what will change, how people and data will move from one to the other, and what evidence must be in before the next rollout stage.
Then assign each capability one of five dispositions: preserve, adapt, replace, retire or leave unchanged. These are finer-grained than the four user outcomes in the legacy software decision framework: adapt and replace are two depths of redesign, and leave unchanged is a preserve that needs no work. Use this checklist to find what belongs in the ledger.
| Product element | Where to find it | Decision to make before implementation |
|---|---|---|
| Saved views, links and drafts | Product data | Which remain valid, and what happens to work that cannot carry over |
| Familiar names and daily actions | Support tickets, sales calls, watching people work | What stays recognisable, and why a rename is worth the cost |
| Roles, permissions and approvals | Account administrators, product configuration | Which authority and access rules must survive unchanged |
| Exports, reports and integrations | Logs, account administrators | Which outputs other systems or customers' own procedures read |
| Keyboard shortcuts and bulk actions | Your heaviest users, usage data | Which power-user paths survive, move or go |
| Training guides and procedures customers wrote | Account administrators | Which screenshots and steps the change will make wrong, and who tells them |
| New workflow or state | Your own design | How it is introduced, explained and supported |
Not every habit deserves protecting. Some exist only to work around the product, such as checking a status in three places because no screen shows it plainly, and those are what the redesign should remove.
Two rules keep the ledger honest. Anything marked "replace" needs a reason that a customer would accept, not only a reason the team prefers. And the person who owns the ledger should be someone who talks to support, because support sees the exceptions that design files omit.
Test with the people who stand to lose something
Test changed tasks with two groups, because they answer different questions. Existing users show what the change disrupts: the habit that no longer works, the saved view that no longer opens. People new to the product show whether the new structure is understandable without learned workarounds. A test with only one group tells you only part of what you need.
Ask existing users to do the same task twice. A slowdown that fades on the second attempt is relearning; one that persists is a design problem.
Add a third source: the people who handle failures. Support and customer-success teams know the recovery paths and the exceptions, and a redesign that handles the normal route well can still strand someone on the abnormal one.
For a B2B product, there is a stronger version of the test, which is a pilot with real customers using the new design in their real work. Slack's team described this for its 2020 redesign. Slack designers worked with about 100 users from its champion network, across dozens of organisations, in a shared channel, and could ship a revised version within a day of hearing feedback. The pilot and a separate benchmarking study with people who had never used Slack changed the design. In that benchmark, participants ignored a compose button placed at the bottom of the sidebar, so it moved to the top. The team had tucked channel member counts behind a click, then found that people use them to judge how to behave in a channel. Right-click actions were made more comprehensive after testing and pilot feedback. Views such as the people directory that started collapsed were opened by default after people unfamiliar with the product missed them.
These findings came from people using working prototypes, not from reviewing screens: pilot customers over weeks and months, and first-time users doing set tasks.
Release in stages with a way back
A redesign that is finished and tested still carries some risk that only live use reveals. Reduce it by controlling who sees the change and by deciding in advance what would stop it.
Start small and compare. Google's Site Reliability Workbook defines canarying as a partial, time-limited deployment of a change together with its evaluation, against a control that does not get the change. Two of its points transfer to a redesign. The canary has to be large enough and last long enough to be representative. And metrics should be broken out by version, because a serious problem that affects a small share of traffic can be diluted in a system-wide average. For a redesign, the same logic says to compare the group with the new design to the group without it, not to watch a single total.
Widen in rings, with a pass rule for each. A 2019 Microsoft case study describes what its authors call controlled rollout, which treats each ring of a phased rollout as a measured experiment and defines the criteria to pass before the rollout widens. Write the rule before launch, so the decision to widen is not made by whoever is most excited or most anxious on the day.
Read early numbers carefully. Effects in a controlled test are not always stable over time. Sadeghi and colleagues at Microsoft describe novelty, an initial interest in something new that fades, and primacy, a growing engagement as people adopt it. For a redesign, our inference is that a first-week dip or lift is not a verdict. An opt-in group can also differ from your whole customer base, so a good result in a beta does not carry over by itself to customers who did not choose to try it.
Give people a way back, and learn from who takes it. When Figma released its redesigned interface, it ran an open beta first, gave users a way to revert, and tracked feature adoption, performance and feedback. The team reversed one of its own design choices before full release: floating panels, which it found cramped the canvas and slowed people down, became fixed panels that can still be resized. Our inference is that a revert option can show where the design fails, as long as you read it as evidence and not as a loyalty vote.
What to watch. A starting set for each ring:
| Signal | Why it matters |
|---|---|
| Time and attempts to finish core tasks, for existing users separately | Shows relearning cost that a completion rate hides |
| Support contacts by topic | Shows confusion that does not appear as an error |
| Reverts to the old design, and the stated reasons | Shows what the new design lost |
| Errors, failed saves and abandoned drafts | Shows breakage, not preference |
| Accessibility checks on the changed screens | A redesign can regress them; check against the current W3C WCAG 2.2 Recommendation (December 2024), whose conformance also satisfies 2.1 and 2.0 |
In a B2B product, also decide who controls the timing. Giving account administrators notice and a choice of when their workspace switches is our recommendation, not a finding from the sources above, but it follows from the ledger: they are the people who own the saved views, training material and procedures.
Tell customers what changed, and what did not. Before a stage reaches an account, tell its administrator what will move, what stays where it was and when the switch happens. Inside the product, a short notice at the changed area can say the same, with a link back to the old route while one exists. Saying that their saved views, exports and reports still work matters as much as announcing anything new. Keep price changes and feature removals out of the same release, or the redesign will take the blame for both.
How long all this takes depends on how many areas change and how long each stage needs to give a reliable reading. Plan the first stage as one area, its starting measures, the shared rules it sets and one compared release, and size the next stage from what that one shows.
One design that changed one branch
Selective change is easier to see in a small case. In Dealpath, institutional real estate software, adding a listing to the pipeline opened a new-deal flow before checking whether the deal already existed. In our Dealpath work, that order was reversed in the proposed design: the user searches existing records first and creates a new deal only when there is no suitable match. The other listing actions, which are viewing details, marking comparables and passing, were left substantially intact, because the structural problem sat in the pipeline path. Dense listing information also got a clearer reading order.
The case shows the shape of the argument in this guide: one structural fix, with the rest of what users knew left alone. It is not evidence of a rollout method or of a business result, and the case page claims neither and does not confirm which version shipped.
When a redesign is the wrong tool
Hold the redesign back when any of these is true.
- The foundations block the change. If the behaviour you want cannot exist on the current data model or platform, a redesign is likely to stall in development. Decide the modernisation question first.
- There are no established users yet. Without existing habits to protect, this is a product design problem, and the continuity work above has nothing to record.
- Stakeholders disagree about the cause. Settle whether the problem is the screen, the workflow or the product model before choosing a depth. A written diagnosis from an audit is one way to settle that before committing to a redesign.
- The pain is small and the area is stable. Leaving it unchanged is a legitimate result of the ledger.
What to ask a design partner before you hire one
Buyers may also ask which agencies to hire for a product redesign. A list of names will not tell you whether a partner can protect your existing users. These questions will.
- What did you leave alone on a comparable project, and why? A partner who redesigned everything may not have made the preserve-or-replace decision.
- How do you test with existing users, not only new ones? Look for a plan that includes both groups and support staff.
- What goes in the continuity record, and who keeps it? There should be an artefact, with an owner on your side.
- Who will do the work day to day, and who owns implementation and rollout? Design only, design carried through to code, designers who work inside your team's planning for an agreed share of their time and see decisions through rollout, and an agreed software scope are different engagements. The answer should be settled before the redesign reaches production.
- What would make you advise against a redesign? A credible partner has an answer, such as the cases in the section above.
- What will we have at each decision point? You want evidence for the next commitment, not a deliverable list.
At Tcules, we work on this kind of change for established B2B SaaS products. The product redesign page sets out the depth options, the continuity ledger and the rollout reasoning in more detail, and a good first step is to bring one part of the product customers have built work around, one part that now creates sales, support or operating cost, and the change the business needs to make. A senior designer leads that work and does it, and we bring in specialists when a problem calls for them.
About the authors
Yash Gidwani
Product Designer
More from YashYash Gidwani is a Product Designer at Tcules who writes about getting people into a SaaS product and keeping them there. Yash's guides cover SaaS onboarding strategies, onboarding and engagement software, and redesigning a live product without upsetting the people who use it. Other articles go into specific workflows: lending and loan origination apps, insurance quoting and claims, staffing and recruitment software, and a search-first create flow that stops duplicate records. Yash has also co-written the SaaS UX audit checklist and a 20-question checklist for legacy modernisation readiness.

Devagya Sharma
Product Designer
More from DevagyaDevagya Sharma is a Product Designer at Tcules who writes about the point where design meets code. Devagya's articles cover when to prototype in code instead of Figma, how a UI/UX designer differs from a design engineer, and how to get an app built with AI tools ready for real users, from security to clean-up. Other pieces deal with keeping a live product steady: silent interface regressions, legacy software modernisation, rebuilding versus modernising, and auditing a design system. Most are written for product leads deciding what to fix first.