Shirley XueAll work
ablefyShipped2.5-year arc2023–2025

Designing where every deal is a negotiation

Problem
Sellers of high-ticket coaching and education products run sales teams on spreadsheets, duplicate products and manual invoices. The platform had checkout pages — but no answer to how a negotiated deal actually happens.
Solution
Deal and offer flows, multi-role commissions validated with working coded prototypes, payouts and analytics — every cycle started from recurring seller research.
Result
Shipped and in production: the segment runs its negotiated deals, commissions and payouts here instead of on spreadsheets, and the line kept growing after I handed it over.

Role & team

Sole product designer, start to handover

  • Early 2023 to the handover in October 2025.
  • A team of eight: one product manager, five engineers, one QA and me — plus the growth team who brought the business case.
  • Disagreements were settled on dated proposal pages, reviewed in the open; when two directions competed, both went on one page until one shipped.
  • Wrote the handover document, ran the walkthroughs, then mentored my successor.

Duration & scope

2.5 years, four surface families

  • MVP in production May 2023, after openly reviewed, versioned proposals.
  • Deal and offer flows, multi-role commissions, payouts, analytics.
  • Evidence base of 29 dated artifacts; eight interview transcripts in 2025 alone.

Key metrics

~11× annual deal volume, launch year → 2025 · monthly deals ~×20 by mid-2026

  • Processed volume ~12× over the same window.
  • Ratios only, absolutes withheld; the line's outcomes, not causal claims for the design.
Context

At ablefy, a creator-commerce platform, one segment of sellers sells differently from everyone else: high-ticket coaches and academies whose sales teams close each customer in a personal call. Nothing about that fits a standard checkout — every deal has negotiated products, negotiated installments, and a setter and a closer who each expect a cut.

Before this product line, sellers held all of that together with workarounds.

A deal has to behave like a sales call and like an invoice at the same time, inside a platform that already existed:

  1. A deal is not a checkout — the flow had to mirror how a sales call actually runs: pitch, negotiate terms, build an individual pricing plan, then send.
  2. Money must stay correct — commissions, VAT-compliant invoicing per installment, and payouts have no tolerance for design shortcuts.
  3. Ship inside a live platform — every surface reused the platform's components and order machinery, not a parallel system.
Problem

The workaround economy

The research kept surfacing the same picture: sellers weren't missing a feature, they were running a parallel operation to compensate for its absence.

  • Duplicate products as commission routing — one product copied per team member, in one case 1,200+ copies.
  • Manual invoices — setters spent hours per month writing invoices for closers instead of selling.
  • Spreadsheet commission math — sellers hired staff whose job was reconciling who earned what.
  • Legal friction — closers accepted terms and cancellation policies in the payer's name during calls, a dispute risk sellers worked around with signed paper contracts.
  • B2B buyers needed per-installment invoices; one invoice up front meant pre-financing the whole order's VAT.
Evidence

An evidence base, refreshed every cycle

Each design cycle started from data I collected and analyzed myself, and every artifact kept its date.

  1. 01

    Usage data

    Platform data analyses of who sells team-style, refreshed twice a year — the activation target list.

  2. 02

    Seller requests

    Request syntheses per cycle, deduplicated into problem statements rather than feature tickets.

  3. 03

    Interviews

    Recurring seller and team-member interviews — eight transcripts across 2025 alone, most of them run by me, some running two hours.

  4. 04

    Market sizing

    A TAM/SAM/SOM model for the segment: high-ticket telesales teams selling virtual products in the DACH market.

Users

What users taught me

The requests said “we need CRM features.” The interviews and the usage data said something more specific — three observations, each of which became a mechanic:

“We make a copy of the main product like ten times for ten people.”

Duplicate products weren't clutter — they were commission routing, built by hand. That reframed the requirement: not CRM features, but per-person commission mechanics on a single product.

When a payment fails, the payer calls their sales member — not support.

A member fielding that call needs the order's truth, not a CRM record of it. So deal states mirror order states — cancelled, closed, pending, charged back — and a deal can never disagree with its own order.

Closers were accepting terms in the payer's name during calls.

One legal observation in an interview split the model in two: a binding deal a closer finalizes in the call, or a non-binding offer the payer accepts themselves — terms, cancellation policy and all. Nothing is agreed in someone else's name.

Versions

Dated proposals, argued in the open

The deal flow was not decided in one pass: three dated proposal rounds, one of them an open review, then one page holding two competing directions side by side — a fuller build and a leaner one, “Shirley's Simplification”. A reduced scope followed, and the MVP was in production that May — four months from first proposal to shipped product.

  1. Jan 18, 2023

    V1

  2. Feb 7

    V2 — “Open Review”

  3. Mar 2

    V3

  4. One page, two directions

    The decision point

    “Version Mar15”

    the fuller build

    “Shirley’s Simplification”

    the leaner one

  5. Mar 15 → May 2023

    “Reduced scope”, then production

    Four months from the first proposal to the shipped MVP.

Dates and page titles are the design file’s own.
Mechanics

Deal mechanics, designed decision by decision

The core surfaces grew through openly reviewed, versioned Figma proposals into a deal flow that mirrors the call: pick products, build the individual pricing plan, capture the payer, send. Negotiated installments, future first-payment dates and add-ons ride on the platform's standard order machinery rather than beside it.

The screens here are a reconstruction I built: the product's real structure on synthetic data, so the mechanics can be clicked through. The shipped product has moved on since the handover.

Open the clickable reconstruction ↗

Creating a deal in the reconstruction, four steps with the summary re-deriving: two products picked for 3.000,00 €, a six-payment plan whose first installment is edited to 1.000,00 € and the rest rebalance to 400,00 €, a business customer, then a setter and a closer assigned with their commission in euros on net, and the created dealView full size
A deal built the way a call runs · products, plan, customer, send — the summary re-derives at every step, and editing one installment rebalances the rest so the total never drifts
The deal surface, reconstructed — a closed deal with its sales member, commission on net and automatic payout state, the installment plan in plain language, the mirrored order status with payments 2 of 6, customer, payment method and a net, VAT and total summaryView full size
The deal surface · negotiated plan, sales member and commission on net, automatic payout state, and the order's own status beside the deal's
The deals list with pending, closed, cancelled and charged-back chips; a pending offer is opened, the test acceptance is sent, and the deal flips to closed while the order status beside it reads activeView full size
Deal states are order states, one for one · a non-binding offer stays pending until the payer accepts in their own name; the moment it is closed, the order agrees
Commission tracking, reconstructed — per-transfer rows with transfer amount and net side by side, transfer status, member and role, commission with its percentage underneath, bonus, total incentive, automatic payout state and dateView full size
Commission tracking · every transfer traceable to role, rate, base and payout state

Every column is a decision

Five money columns, and the decision behind each.

  1. Close-up of the commission table columns: member with role underneath, commission amount with the percentage underneath, bonus, total incentive

    Role · commission amount · bonus · total incentive

    Commission is a rate per role, per person, per product — the setter and the closer on the same deal each carry their own. This was the hardest thing on the line to get right, and the reason two working prototypes existed before any engineering did.

  2. Close-up of the transfer amount column: four hundred euros with three hundred thirty-six euros thirteen net underneath

    Transfer amount, with its net underneath

    The base the percentage was applied to sits on the row, not in a help article. A seller reconciling a payout should never have to reverse-engineer which number the commission came off.

  3. Close-up of the transfer status column: two transfers in an error state above a successful one

    Transfer status, including the failure

    Money surfaces have to show their failures. Every transfer carries its own status, so a failed one is visible on the row that caused it instead of turning up as a quiet gap in somebody's commission at the end of the month.

  4. Close-up of the automatic payout column: active, paused, active

    Automatic payout, as its own column

    Earned and paid are different states. Each member on a deal has their own payout eligibility, so “what did I earn” and “when does it reach me” stay two separate, separately answerable questions.

  5. Close-up of the deals list columns: deal status chips for pending, closed and charged back, next to price and plan

    Deal status, next to price and plan

    The states on the chip are order states, one for one. A deal that disagreed with its own order would put a salesperson in front of a customer with the wrong facts.

Close-ups from the reconstruction · synthetic data

Prototype

Prototype in code, test, then build

The hardest design question was multi-role commission setting — flexible enough for setter/closer/affiliate splits, simple enough to configure without a manual. Mockups couldn't answer it: the failure modes only show up when someone actually configures a team.

So in May 2025 I built two working coded prototypes — the first the company had — and ran an internal test with four colleagues configuring real scenarios. The test summary picked the direction.

In the reconstruction, the closer's rate on the incentive plan is stepped from ten to twelve percent and the preview line re-derives; a new deal's send step shows the new percentage on net; the commission tracking row for that deal carries itView full size
The rate a seller sets on the incentive plan is the rate the deal previews and the rate the transfer row carries — one definition, three places it shows

Build one, and watch the money follow

The reconstruction below is playable. It is the flow with the platform plumbing stripped out — four decisions, one running total: pick the products, change the payment plan, move a commission rate, choose how it goes out. The summary on the right re-derives as you go, the way a closer would see it mid-call.

Try it

Every deal is negotiated, so the products are picked per deal rather than taken from a fixed package.

Reconstruction, simplified for the story — synthetic products and people · the arithmetic is the real one (VAT 19%, commission on net) · the shipped product has moved on since I handed it over

Measure

The bottleneck the surveys found

I treated measurability as design scope: added the missing metrics to the line's adoption dashboards and ran recurring in-product team surveys from 2023 on. The last was a bilingual feature survey designed around awareness branching — sellers who didn't know the analytics existed were routed into them before being asked to judge them. Awareness, not capability, had become the bottleneck, a finding later research kept confirming.

Why

Why it worked

Two and a half years is long enough to see which decisions carried weight. Four did — and one thing the line never solved:

  • Reuse over reinvention. Deals ride on the platform's existing order, invoicing and payment machinery rather than a parallel system. Correct money was inherited rather than rebuilt.
  • Behaviour beat requests. Every cycle opened by asking who already sold team-style without a team setup, and designed for them.
  • One research detail reshaped the whole flow: the binding / non-binding split.
  • The expensive question was answered in working software, not in a review. Multi-role commission setting was the one design a mockup couldn't settle, so the direction was decided by four colleagues configuring real teams — months before the real build started.
  • And the one it never solved: capability shipped ahead of awareness. By the end the constraint was how many sellers knew what the line could do.
Handover

The handover as a design artifact

In late 2025 the line scaled beyond one designer and I handed it over. I treated the handover itself as a deliverable: a structured handover document, sessions walking my successor through every project's background and state, and onboarding her directly onto a live kickoff so the transition happened inside real work, not beside it.

The research archive went with it — as a queryable AI knowledge base built from 29 dated artifacts (usage analyses, interview transcripts, problem statements, test summaries), so the next designer could interrogate two and a half years of evidence instead of reading a folder.

The line kept growing after I left it.

Line chart of monthly deals through the product line from May 2023 to July 2026, indexed to the first full month after launch, rising from ×1 to roughly ×20 with design milestones marked and a dashed handover line in late 2025 after which the curve keeps climbingView full size
Monthly deals, indexed (first full month = ×1) · real platform data, absolute values withheld · milestones mark the design timeline