Senior Product Designer · UX Research Lead
Eight years of designing and researching sales and ERP software: from orders by fax to an Outlook add-in and AI that answers business questions from ERP data.
“I want the same experience in our CRM as in my Outlook.”Customer workshop
Orders with 50–100 lines took 30–60 minutes to type into the ERP.Back-office interviews
“I don't want to file every mail.”Customer interview
“I was informed as the person responsible, then nothing came, or too late.”Customer interview
Four case studies from my current role, each with what changed and how we know.
Sales teams lived in Outlook, but the customer record lived in our software. Instead of rebuilding a mail client, I researched, designed and tested an Outlook sidebar that connects every mail to the right customer, quote and order. Six changes came straight from customer tests.
Read case study →An AI document import that reads customer orders in any layout and language and prepares the order in the ERP. 99 % of customers and 95 % of articles are recognised correctly, and a person only checks and confirms.
Read case study →A live MCP server that lets managers ask their ERP real business questions, from an early-warning list of slipping customers to a CEO briefing.
Read case study →Building a five-person UX team, and how customer interviews turned into the product roadmap, from the activities module to the journey of a field sales rep.
Read case study →I work where user value, business viability and technical feasibility meet. These habits show up in every case study above.
Interview findings get sorted into basic needs and delighters before anything is drawn. It keeps MVPs small and gives the roadmap a reason for every “later”.
Before designing, I walk through how established products solve the same job, flow by flow, and note what to copy, what to avoid and where the gap is.
Every major flow is tested as a prototype with real users. The findings go straight into the concept, not into a report.
I bring research to management and development as prioritised options with trade-offs. That is how interview findings ended up as roadmap items and user stories.
I design AI features (MCP, agents, assistants) and use AI daily: Claude for research synthesis, competitor and SWOT analyses, user flows, wireflows and journeys; Miro AI for whiteboarding. I also train colleagues to do the same.
I introduced a design system for one coherent product experience and treat it as a product whose users are designers and developers.
I lead people, not org charts. My job is to help each of them grow, so leading runs on a fixed rhythm: everyone knows when goals are set, where problems surface and where to ask. Decisions are explained, not just announced.
AI exchange: regular sessions where a different colleague hosts each time, shows the AI tools they are excited about and how they could fit into daily work.
Since 2021 I teach UX design and UX research at university level: module lead, author of the master's study scripts UX Design: Principles & Methods and UX Research, and supervisor of master's theses. As a mentor I guide students through complete case studies, from the first briefing to the finished portfolio piece.
A UX study for an app that connects expecting and new mothers with trusted care providers, from prenatal massage to lactation consultants, in one place. I mentored the student end to end: briefing, research plan, synthesis, concept and UI reviews, and the final portfolio presentation.
View the case study on Behance ↗A complete mobile app case study, from research and user flows to a high-fidelity interface built on Material Design. I mentored the student from the first briefing through every design iteration to the final portfolio.
View the Figma file ↗Shared with the students' permission. All design work belongs to the students.
Shorter pieces that show range: market analysis, research projects and the work that shaped how I design today.
For a new cloud middleware that lists products on marketplaces, I analysed Channable, ChannelEngine, Mirakl and Productsup: their shared flow (master catalogue, rules, AI enrichment, human approval, publish, monitoring) and their pricing. My recommendation, a price tied to merchandise volume instead of the number of marketplaces (“3 or 100 marketplaces must not matter”), was adopted.
Can a mobile provider's service app become a habit? Seven interviews and five user diaries, a service vision and a reward system, validated in three iteration interviews.
Read the case study →Project lead, product strategy and UI/UX for a platform where dealers and resellers order and communicate. Research showed orders still came by fax and printed catalogue. Competitive profiles, persona, customer experience map, card sorting, Crazy 8s, prototypes and think-aloud tests.
Emotion psychology and Norman's visceral, behavioural and reflective levels applied to an iPad app: recipes by time and budget, a shopping list that scales with the number of people, a guided cooking mode. Refined in three rounds of user testing.
About
I started as a trained media designer, studied UX/UI design in Munich and San Diego, and did a master's in design management. Since 2018 I've worked in B2B software: first as a UX/product designer, then building and leading a UX research team, and now leading UX research and AI solutions.
Outside of product work I co-founded a small natural-cosmetics brand that donates part of every sale, and I founded a non-profit that supports a school in Kathmandu, where I volunteered after my studies. Running both taught me what product, brand and sales look like when it's your own money on the line.
I write and speak German natively and work in English every day.
Experience
B2B CRM & sales software for ERP users. Outlook add-in, activities module, MCP server and AI roadmap; AI enablement for sales and marketing.
E-commerce & ERP software vendor, 200+ people. Built and led a five-person UX team; research goals and KPIs; reporting to management; mentoring researchers, designers and PMs.
Same company. Concept team for a cloud ERP assistance product; introduced Design Thinking, Lean UX and a design system.
SRH distance-learning university (module lead, study scripts, thesis supervision); CareerFoundry until 2026.
Macromedia University Munich; semester at National University, San Diego. Projects with BMW, Greenpeace Switzerland and a national newspaper.
Case study · 2025–2026
Sales teams in mid-sized B2B companies run their day in Outlook. Their customers, quotes and orders live in our CRM on top of the ERP. I led the research, concept and design of an Outlook add-in that connects the two.
Research lead and designer: interviews, workshops, competitive analysis, concept, prototype, usability tests, alignment with management and development.
Mails, customers, quotes and orders are connected in one place. Reps file a mail, create a task or a quote without leaving Outlook.
Concept tested with three customer companies; six changes went into the release. Live in 2026, beta with a free six-month trial for all customers.
Confidentiality: the findings, numbers and decisions in this case study are real. Screens and charts show the fictional product Belego CRM with fictional data.
In customer interviews and workshops the pattern was the same: every customer conversation starts in the inbox, but the context for it (open quotes, orders, complaints) sits in the CRM. Reps copied information back and forth, and mail history was invisible to colleagues.
“I want the same experience in our CRM as in my Outlook.”Customer workshop
12345Personal and shared mailboxes in one sidebar, each with its own sender address.
“Link all” files the mail to customer, contact and quote, or each item has its own link icon. Nothing is created automatically.
Open by default and pinned, because it holds what nobody may miss, such as changed payment terms.
The e-mail address matches a contact, so the customer appears without a search.
Customer, contact, last quote or order, recent activities. Users asked for exactly these. Everything else opens in Belego.
123Task, meeting, visit report, note and phone note are created right in the sidebar.
Quotes, baskets and orders need prices and conditions from the ERP, so they open in the full product.
Customer and mail are carried over, so the new record starts with its context.
12345The sender is unknown, and the sidebar says so instead of showing an empty card.
Customers, contacts, quotes and orders, including e-mail addresses in custom fields.
A domain in a custom field, a quote number in the mail, or only a similar name.
The best match is preselected, but the link only happens on click.
New customers are created in the ERP. The sidebar offers a task for the back office instead.
Redesigned as the fictional product Belego CRM · fictional data · findings real
Before building, I walked customers from three companies through the clickable prototype, recorded the sessions and synthesised the findings with management and development. Six changes went into the release:
1234Instead of a contact form, a task “Create customer in ERP”. This replaced the quick contact form after testing.
Sales passes it on, and the back office creates the customer the way their process requires.
The details carry the sender and company from the mail, so nobody starts from scratch.
The back office sees the original request.
1234It can’t be hidden or moved, whenever there is one.
Same logic as the customer view in Belego. Admins set the default for everyone.
Personal and shared mailboxes, each with its own sender address.
When the original mail is linked, replies and forwards follow automatically.
The add-in was released in 2026 as one of the top roadmap items and replaces the old mail-client integration. It is in beta, with a free six-month trial for all customers to drive the switch. In the feedback round, both test customers approved the strategy of moving into Outlook.
Next on the roadmap: automatic sync for recognised customers (it needs a background service with mailbox access, so security comes first), writing mails from the CRM, and the delighters from the Kano analysis such as turning a mail into an order.
Case study · 2026
Sales back-office teams receive orders as PDFs, scans and photos, in every layout and many languages. Typing an order with 50 to 100 lines into the ERP took 30 to 60 minutes. I rebuilt the AI document import so that it reads the order, matches customer and articles, and prepares the order. A person checks it and confirms with one click.
Concept, model evaluation, the extraction prompt on Microsoft Azure with Claude, pilot planning and the feedback loop with users.
According to customer interviews, a 50–100 line order took 30–60 minutes to enter. Now it takes about 1.5 minutes, or 5–10 with a thorough check.
Recognition rate in the pilot: 99 % for customers, 95 % for articles. Live since October 2026, with very positive user feedback.
Confidentiality: the findings, numbers and decisions in this case study are real. Screens and charts show the fictional product Belego CRM with fictional data.
Orders arrive in every format: purchase orders from large distributors, shopping-cart printouts, scans with handwritten corrections, documents in German, English, French, Dutch, Czech or Italian. Multi-page orders repeat letterhead, addresses and table headers on every page. The existing import technology was outdated and missed too much, so back-office staff typed long orders by hand.
Usage data showed which customers relied on the import most. That told us where the pain was highest and where to start.
I compared several AI models on real order documents of different types before deciding on Claude, running on Microsoft Azure. The core of the work was the extraction prompt: a specification of how to read an order like an experienced back-office colleague, written for a model.
// excerpt of the structured output · fictional example
"OrderNumber": { "Value": "4500123456", "Confidence": 0.99 },
"DeliveryDate": { "Value": "2026-08-28", "Confidence": 0.97 },
"Items": [
{ "Position": "0010", "CustomerArticleNumber": "4903041065",
"Quantity": { "Value": 24.0, "Confidence": 0.98 } },
{ "Position": "0020", "Title": "handwritten correction",
"Quantity": { "Value": 12.0, "Confidence": 0.68 } }
]
In the first week of the pilot, a large order went through without a single correction, and the feedback from the users was very positive. A first prospect is now planning a proof of concept with its own documents.
One general prompt is not enough. Every supplier's customers write their orders in their own way, from article number formats to how they label quantities. The remaining errors came from those conventions, not from the model. So the general system prompt needs a second layer: a customer-specific prompt that each installation can refine. I aligned this with development and submitted it as a change request.
Concept study · self-initiated · fictional data
The pilot kept the old review screen on purpose. But the engine now knows how sure it is about every field, and the screen doesn't show it. So people still re-check every line. I designed the review screen the engine deserves, inside the existing basket layout.
What the current screen gets wrong:
Foundation · every screen is tagged with the principles it serves
Based on Don Norman's four principles of human-centered design, built with Material Design 3.
The symptom is “people re-check every line”. The cause: they can't see what the AI is unsure about. The remaining errors come from customer number formats, so the fix is rules, not blame.
Back-office staff compare against the paper order, work with the keyboard and handle orders in several languages. The document stays next to them, and they move line by line.
The unit of design is the whole activity, from the order mail to the confirmed order: customer formats, ERP master data and prices, and what the next import learns from this one.
Wireframes in three versions, then high-fidelity screens in several review rounds. Next: think-aloud tests with 3–5 back-office users, measuring time to order and missed problem lines.
1234567Instead of a modal behind an icon. Flagged lines are highlighted in it, and it can be hidden for wide tables.
“All / To check” already exists. With “4” on it, the work looks as small as it is.
Check quantity, check article, no match, instead of one orange icon for everything.
An unclear quantity is confirmed inside the row with one click. No panel, no dialog.
“~36 min saved” (72 lines × ~30 s of typing). Minutes are what people feel.
When the AI is sure, the customer is applied and explained. One click to change, no blocking step.
“Order” stays disabled until every flagged line is checked. Today it can be pressed with €0.00 prices.
1234The sender block is highlighted in the document, so people compare against the source, not against memory.
Every suggestion names its evidence: same customer number, similar name, different region.
“Continue without” stays possible, but says what it means: no prices, no discounts.
Instead of €0.00, the price column says why it is empty.
1234Selecting line 0470 scrolls the document to page 2 and highlights it. No modal, no searching.
Today the panel promises “choose from the list below” and shows only a search. The list is now there.
The best match is preselected and explained by the customer's own order history.
The pilot's main learning, a customer-specific prompt layer, turned into a checkbox people understand.
123Plain sentences instead of prompt text. Under the hood each rule becomes part of the customer-specific prompt layer.
Nobody configures this upfront. The list fills itself from corrections.
The AI may propose a rule from repeated corrections. It only applies after someone accepts it.
123A snackbar with “Open order” instead of a dialog. The person stays in the flow for the next import.
The new rule is listed before it shapes the next import from this customer.
Prices and conditions come from the ERP, never from the document.
Case study · 2025–2026
An ERP knows every order, quote and customer, but answering a business question still meant exports and pivot tables. I conceived and led an MCP server that lets AI work directly with that data, and defined which questions it should answer first.
Concept and project lead: AI framework, question catalogue, MCP server, use cases, internal AI enablement.
Managers ask in plain language which customers are slipping or where quotes get lost, and get an answer backed by their own figures.
Live with real customer data. First reaction of a managing director: “First feedback: aaaawesome!!!”
Confidentiality: the findings, numbers and decisions in this case study are real. Screens and charts show the fictional product Belego CRM with fictional data.
A normal ERP query answers “show me the orders of customer X”. Sales managers ask different questions: “Which of my important customers show first signs that we are losing business, and where should my team be active tomorrow?” That needs data from customers, quotes, orders and articles put into relation, not just found.
123A sales manager asks. Through the MCP server the AI reads customers, orders and articles from the ERP.
So the loss in the risk groups shows up behind the overall growth instead of disappearing in it.
The ten interventions that matter most, ranked by revenue at risk.
1234The answer names five unprocessed large quotes, with their value.
Figures from the ERP on one side, the AI’s reading of them on the other.
The AI says what it cannot know instead of inventing reasons.
The quotes worth following up, quantified in euros.
The MCP server is live and has been used with real customer data. The first reaction of a managing director after the launch was simply: “First feedback: aaaawesome!!!”
Inside the company I lead AI enablement: I defined the AI tooling for marketing and customer success, ran the potential analyses with both teams and have trained six colleagues so far.
Not every user knows how to connect an MCP server and work with it. So the next step brings AI into the software itself:
1234Orders, cancellations, revenue, activities and visit reports, without having to ask a question first.
Status changes, orders, cancellations, quotes and visit reports as they come in.
The AI suggests what to do next. Each suggestion becomes a task in one click, or is dismissed.
Order history, visit reports, analyses. People can check before they act.
12345Orders, quotes, visit reports, mails and activities in one question, without setting up an MCP connection.
Every statement carries a number that leads to its source.
The order lines, the visit report, the mail. One click to the original.
If the data holds no reason, the answer says so instead of guessing.
Answers only use Belego and ERP data the user is allowed to see.
Case study · 2018–2026
I joined as a UX/product designer when user research was not yet part of product development. Today customer interviews decide what goes on the roadmap and arrive in development as user stories.
Built and led a five-person UX team: two UX designers, a UX researcher, a UI designer and a product manager.
Research findings define the product roadmap for 2025 and 2026 and reach development as concrete user stories.
Activities phase 2 specified from interviews; a field sales journey and Kano synthesis from interviews at six customer companies and three partners.
Confidentiality: the findings, numbers and decisions in this case study are real. Screens and charts show the fictional product Belego CRM with fictional data.
Activities (tasks, notes, visit reports, opportunities) are the backbone of daily work in the product. I interviewed customers about how they use them and synthesised the findings into a phase-2 concept.
“Colleagues who are in the software all day are fine. If I'm only in briefly, we've become slower. I was informed as the person responsible, then nothing came, or too late.”Customer interview, translated
I aligned the concept with management and development, and wrote phase 2 as user stories in the development environment.
1234Users build their own views, such as “my opportunities” or “quote follow-up”.
The count grows, but nothing says which ones need you now.
Type and status stay visible and can be removed in one click.
Changes only show up when you open the record. People in the software all day keep up. People who drop in briefly miss things.
123Each phase holds its tasks, ticked off as the deal moves on.
Probability, expected revenue, phase and status.
Comments and mentions stay with the opportunity instead of in mail threads.
1234“For you” collects what is assigned to you, mentions and due tasks.
Like in Microsoft Teams, an @mention becomes a notification, not a comment someone has to find.
For each event: in Belego, as a mail digest, or off.
Field-level changes stay in the change log unless someone switches them on.
12345Open and completed activities, calls, visits, mails and quotes, filterable, one click from the customer.
Calls are logged with a summary as a phone note, and the note stays editable.
Interviews asked for the whole history, not only what is still open.
Visit reports and activities can be shared with colleagues and with customers.
The customer sees title, date, notes and photos. Internal comments and prices stay hidden.
To decide what a mobile tool for field sales reps should do, I ran 30-minute interviews with reps, sales managers and managing directors at six customer companies and three business partners. I mapped every requested capability on a Kano diagram: what reps expect as a given (contacts, opportunities, phone notes, visit reports, access rights), what drives satisfaction (calendar, responsiveness, a customisable customer dashboard), and what would delight (analyses, pushed alerts about unusual customer behaviour).
| Scheduling the visit | Preparing the visit | Customer visit | Order | Visit follow-up | Delivery | |
|---|---|---|---|---|---|---|
| Data sources | CRMERPCustomer file (Excel)Visit reportsCustomer careMail, phone | CRMERPCustomer file (Excel)Visit reports | Saved or printed lists | ERP | CRMERPCustomer file (Excel)Visit reportsCustomer careMail, phone | ERP |
| Devices | LaptopWorkstationSmartphone | LaptopWorkstation | Laptop or tabletSmartphone | LaptopWorkstationPaper and pen | LaptopWorkstationSmartphone | LaptopWorkstation |
| Pains | The visit rhythm depends mostly on an ABC analysis (revenue, number of orders)Reps often plan their routes themselves | Relevant data has to be searched for by hand | Discount structures often missing from price listsStock and delivery times often unavailableNo push updates on current topics at the customer | Extra, sometimes slow communication delays the orderOrders are still written on paperAfter ordering, reps often cannot track their quotes | No software support for writing the visit report right after the visit. It happens on the home-office FridayThe back office often lacks information to set up new customers properly | No order tracking that warns about unexpected delays |
| Needs | ABC analysisOverview of regular appointmentsPositive and negative events at the customerEfficient route planning with appsStatus of target agreements | A customer dashboard with what matters at a glance: last visit report, contacts, assortment, targets, revenue, order history, notesA 360° view of the customer | Remote access to all relevant informationSearchable product catalogue with detailsInstant updates on the customer, “a status check 15 min before the visit”Suggestions for alternatives and new products360° view: “last call we talked about …” | Real-time access to all data for quotingClosing the deal digitallyCreate and sign quotes for standard products on site | A simple, standardised form for visit reportsPush updates to everyone involved when something changesProactive alerts about problems with orders | |
| Mobile potential | Office | Customer data on mobile: order history, payment behaviour, open issues OfficeHotelCar | Customer data on mobileAssortment info for cross- and upselling At the customer | Filling the basket digitally At the customerOffice via back office | Writing visit reports At the customerOfficeHotelCar | Office via back office |
| Roadmap potential | No potential, covered by the customer dashboard | Very high potential | More detail needed | High potential, details from user tests | Too little information | |
| Communication and collaboration with the customer and the back office | High potential, details from user tests | |||||
| Pains | Limited access to data | No real-time updates: changes to the order, incidents at the customer, promotions such as marketing campaigns | The back office often lacks information to set up new customers properly | |||
| Needs | Push updates as soon as there is news about my customers (today: mail, WhatsApp)Activity logReminders for myself, e.g. how do I structure the meeting? | Capture notes and tasks | Fast discount approvalSeamless communication between field and back office often closes the orderFeedback on delivery terms | Distribute tasks across the companyFollow up on quotesContact documentation: order, complaint, feedback, request | Notice about negative events to field sales, back office and customer | |
Alongside the interviews we mapped every role in a typical customer company: what they want, what the ERP gives them today and what the product can add. The map also made clear what we deliberately do not build, such as finance or warehouse planning.
Visits customers, sells, and keeps the relationship. Plans the week from an ABC analysis, prepares at the laptop, works at the customer with a tablet or phone and writes visit reports on the home-office Friday.
“A status check 15 minutes before the visit.”Field sales interview
Detailed above, and the customer journey is theirs. Every other role sees the same customer from a different angle.
The same pattern shaped the whole roadmap for 2025 and 2026. Research findings became these priorities:
Nothing is more honest than real life. Research and user tests show, again and again and without a filter, what really happens at our customers: visit reports written on the home-office Friday, orders on paper, a quick-create form that would have broken one customer’s ERP process. No internal discussion gets that close.
Case study · Master thesis 2017
Most people open their mobile provider's app only when something goes wrong. In cooperation with a leading digital agency, I investigated whether a service app can become a habit, and designed a service vision and reward system to support customer retention.
How can habit-forming conditions be applied to digital service experiences to support customer retention, using the example of a mobile provider's service app?
Secondary research on habits, persuasion, gamification and reward systems, then seven in-depth interviews and five user diaries, synthesised with affinity diagramming into personas and findings.
“I just want everything to work. As long as I have reception, I don't care about my provider.”Interviewee · translated from German
Three iteration interviews indicated that the concept would raise customers' intention to stay. The thesis names field tests with real usage data as the necessary next step.
Retention is the same question in on-demand commerce: why does a creator come back to a platform after the first order? The thinking from this project shapes how I approach repeat journeys today.
Legal notice
Dominik Dippner
Rainer Str. 40
86609 Donauwörth
Deutschland
E-Mail: dominik.dippner@gmail.com
Dominik Dippner, Anschrift wie oben
Diese Website enthält Links zu externen Websites Dritter (z. B. LinkedIn, Behance, Figma). Auf deren Inhalte habe ich keinen Einfluss; für sie ist stets der jeweilige Anbieter verantwortlich. Bei Bekanntwerden von Rechtsverletzungen entferne ich entsprechende Links umgehend.
Texte, Grafiken und Gestaltung dieser Website stehen unter meinem Urheberrecht, soweit nicht anders gekennzeichnet. Die gezeigten studentischen Projekte gehören den jeweiligen Studierenden und werden mit deren Erlaubnis verlinkt. Kunden- und Firmendaten in Screenshots sind Platzhalter, Demodaten oder unkenntlich gemacht.
Privacy policy
Dominik Dippner, Rainer Str. 40, 86609 Donauwörth, Deutschland · E-Mail: dominik.dippner@gmail.com
Diese Website ist ein persönliches Portfolio. Sie verwendet keine Cookies, keine Analyse- oder Tracking-Werkzeuge, keine Kontaktformulare und lädt keine Inhalte von Drittanbietern. Schriften und Bilder sind direkt in die Seite eingebettet.
Die Website wird bei Netlify, Inc., 512 2nd Street, Suite 200, San Francisco, CA 94107, USA gehostet. Beim Aufruf der Seite verarbeitet der Hosting-Anbieter technisch notwendige Daten wie IP-Adresse, Datum und Uhrzeit des Abrufs, aufgerufene Seite, übertragene Datenmenge, Browsertyp und Betriebssystem. Dies dient der sicheren und stabilen Auslieferung der Website. Rechtsgrundlage ist Art. 6 Abs. 1 lit. f DSGVO; mein berechtigtes Interesse liegt im sicheren Betrieb der Website. Die Übermittlung in die USA erfolgt auf Grundlage des EU-U.S. Data Privacy Framework bzw. von Standardvertragsklauseln. Weitere Informationen: netlify.com/privacy.
Wenn Sie mir eine E-Mail schreiben, verarbeite ich Ihre Angaben, um Ihre Anfrage zu beantworten (Art. 6 Abs. 1 lit. b und f DSGVO). Die Daten werden gelöscht, sobald sie dafür nicht mehr erforderlich sind und keine gesetzlichen Aufbewahrungspflichten bestehen.
Links zu LinkedIn, Behance oder Figma öffnen die Websites dieser Anbieter. Erst dann gelten deren Datenschutzbestimmungen; beim bloßen Besuch dieser Website werden keine Daten an diese Anbieter übertragen.
Sie haben das Recht auf Auskunft, Berichtigung, Löschung, Einschränkung der Verarbeitung, Datenübertragbarkeit sowie Widerspruch gegen die Verarbeitung (Art. 15–21 DSGVO). Außerdem können Sie sich bei einer Datenschutz-Aufsichtsbehörde beschweren, zum Beispiel beim Bayerischen Landesamt für Datenschutzaufsicht.
Stand: November 2026