Marije Rottier.
← All work

Case 03 / 07 · SaaS dashboard · In progress

Client Portal

A portal for small business owners on a monthly website plan: is my site OK, what was done, and what do I get for my money. Research, flows and a clickable prototype, with AI that only drafts.

Role
Research, product design, prototype
Type
Dashboard, roles, notifications
Timeline
October 2026, ongoing
Status
In progress
A tilted collage of the portal prototype: overview, change request, plan and the studio review screen
Screens from prototype v1, with a fictional client. The interface is in Dutch because the owners are Dutch.
Problem

Owners pay monthly for website care, but good maintenance is invisible, so they can't see what they pay for.

What I did

A scan of six tools, proto personas, jobs to be done, the IA, three flows, notification rules and a clickable prototype with an AI review screen.

Outcome

Prototype v1, built on clearly marked assumptions. Interviews with 3 to 5 owners come next, then a usability test.

01

Context

The challenge

Paying every month without seeing what for

Small business owners pay a monthly plan so they don't have to think about their website. But when nothing breaks, it looks like nothing happened. The portal has to show how the site is doing and what the owner gets for their money, without becoming another tool to learn.

It's a portal for my own web studio, so it has two roles: the owner, and the studio person who reviews every AI draft.

Make invisible care visible.

Research

Six tools, built for the agency

  • Assembly, SuiteDash, PlutioAll-round portals around the agency's workflow: files, CRM, invoices. Not around the owner's question.
  • Duda, WebflowClient access and billing, but tied to their own platform.
  • ManageWP client reportsMake maintenance visible, but agencies report that uptime numbers alone don't convince clients.

Insight. Existing portals answer “how does the agency manage clients?” The owner asks “how is my website doing, and what do I get for my money?”

Hypotheses for the interviews. Owners want to know that everything works more than they want details. An email summary matters more than logging in. Not knowing whether a request is being handled frustrates more than the wait itself.

02

Process

Define

Jobs to be done and a small MVP

Written before the interviews, on marked assumptions. Two fictional proto personas: a salon owner who reads email between clients, and a small B2B shop owner who checks things in the evening and worries about jargon.

Jobs to be done

  • When I pay monthly for my website, I want to see quickly that it works and what was done, so I feel it's worth it.
  • When something needs to change, I want to ask it in one go and know when it's done.
  • When my business needs more, I want honest advice on what helps.

MVP

  • In: overview, change requests, monthly report, plan and invoices, and the studio review screen
  • Out for v1: chat, file manager, editing the site, analytics, staff accounts, payments in the portal

Information architecture

Four places for the owner, one for the studio

Each client page answers one question. The studio side exists so that no AI text reaches a client without a person checking it.

Client side · the owner

OverviewRequestsReportsPlan
  • Is my site online, does anything need me
  • Ask for a change, see what happens next
  • Read the monthly report in plain language
  • See what my plan includes, invoices

Studio side · me

Review report
  • AI draft from visible data sources
  • Checklist of things to verify
  • Required personal note
  • Nothing is sent without approval

Notification rules

Quiet by default

Owners are busy running a business, so every message has to earn its place. Nothing at weekends except an outage.

TriggerChannelWhen
Request receivedEmailImmediately
Request doneEmailImmediately
Site down over 10 minutesEmailImmediately, one "back online" mail after
Monthly reportEmailFirst week, weekday morning
Upgrade suggestionPortal and report onlyAt most once a quarter
Marketing or sales mailNoneNever

Design decisions

What changed, and why

  1. An honest upgrade

    Only suggested when a rule is met, like change time used up three months in a row. Shown side by side with equal weight, and it can be undone in the first month.

    WhyNo fake urgency, no pre-ticked options, no hidden cancellation. Trust is the product.

  2. Plan usage in plain sight

    “20 of 30 minutes used” on the overview, with the request button right next to it.

    WhyNone of the scanned tools shows plan usage to the client. Seeing it removes the guesswork about what a request costs.

  3. AI drafts, a person decides

    The monthly report is drafted from visible data sources and can't be sent without a personal note and a checklist.

    WhyIt makes human oversight concrete and testable, instead of a promise in a footer.

03

Outcome

Prototype v1

Clickable, in Dutch, with demo data

Built in HTML with a fictional client, so no real business data is involved.

Status

Where the project is now

PhaseWhatStatus
1Discovery: scan, insights, hypotheses, interview guideDone
1bInterviews with 3 to 5 business ownersNext
2Define: personas, jobs, user stories, MVPDone, on assumptions
3Structure: IA, three flows, notification rulesDone, on assumptions
4Prototype v1Done, to revise
5Usability test with 5 usersPlanned
04

Reflection

What I'm watching for

This case is still in progress, so instead of hindsight, the risks I'm keeping an eye on.

  1. Designing ahead of the interviews

    Designing before talking to owners keeps momentum, but it can anchor me to my own assumptions. That's why v1 stays as it is: the interviews get compared against it, not quietly folded in.

  2. Not building what doesn't need building

    The same value can be delivered without code: a monthly report email, a status page, a request form and invoices from the accounting software. A real portal only makes sense with more clients and proof they want to log in.

  3. Keeping the AI honest

    The review screen only works if checking the draft is faster than writing it. That's the first thing the usability test needs to show.