Kanmon case workbench You are on LIVE My working material behind the case. The Present view (top right) is the intended walkthrough; these tabs are the notes and build I worked from.
16 sections · C limits · N notes
You are here: Start (overview and locked idea)
What I want to build

A pre-approval estimate and a shorter application

Work out roughly how much a business could borrow before they ask, show them that number where they already work, and use the same information to make applying take about a minute.

In the practice numbers, only about 4 in 10 people who start an application finish it. They had already shown interest by starting. They leave because the form asks for things they have to go and find, and because nothing tells them whether it is worth the effort. A real number answers that question, and the data behind it removes most of the form.

It is an estimate, not a promise. Underwriting still decides who gets approved.

I am not leading with “show financing at the perfect moment” (set aside for this case), changing who gets approved, or launching brand-new loan products first. · full reasoning ·

How I got here. The approach story walks from the prompt to this pick: what I sized, what I cut, and what I kept as a follow-on. Start there if you want the path in order.

Explore. Pre-approval estimate architecture, experience, and practice funnel lift, in this workbench.

Optional deep notes: architecture prose · partner copy · lift assumptions

Optional jump list navigation
1

Read the ask

Own the roadmap. Win = more businesses get funded through partners. Pick one thing to build and go deep.

2

Learn how it works

Partners show financing. Kanmon lends. Path: apply, get an offer, accept, and get paid. Examples make that real.

3

First AI idea, then set it aside

The assistant suggested “show financing at the right moment.” I set it aside: that pitch is common, and for this case I assume financing is already visible enough on live partners.

Market notes
4

Working constraints I am using

One product manager. Low volume. Little or no A/B testing. Credit and data teams own approval models. Texting is blocked for now.

5

Where to focus

Get more people through the steps on live partners, not hunt new logos first. Assume financing is already visible enough. Lead with a pre-approval estimate and a shorter application so more people finish applying; don’t try to own credit models.

Short list
6

Rough size of the problem

Example month: ~51 funded. ~288 people start and quit before submitting. Flow chart shows where they leave.

Numbers
7

Landed on the idea

A pre-approval estimate and a shorter application, so people who started can finish in the form more often, and good-fit customers get better invites. Recovery and preventable-decline fixes were seriously considered and ranked below.

Why
8

A pre-approval estimate and a shorter application

Partner-held facts can shorten the form and support a light fit check. The same direction also helps invite more of the right people into the funnel. Soft language only, not a hard loan promise.

Two payoffs
9

Still to do

Walk it end to end.” pre-approval estimate screens live under Experience; Screens page still shows the old “right moment” archive.

Success looks like

More small businesses get funded through a partner’s product, not more traffic to Kanmon’s website. Pick one idea and go deep on it.

Working assumptions
  • Small company, one product manager, low volume, little or no A/B testing
  • For this case: financing is already visible enough on live partners (reopen if that is wrong)
  • Credit and data teams own who gets approved; product can invite softly, not promise dollars
Companion files
Source of truth: Kanmon Product Manager Case Study. case brief (the brief). Quoted below. Don’t paraphrase the prompt on the call until you’ve internalized this. Format: live conversational session · up to 1 hour · no trick · no single correct answer
Prompt VERBATIM
If you owned Kanmon's product roadmap and your single success metric was originations driven through our partner platforms, how would you think about the biggest levers? Pick one feature you’d build for this and deep dive into that. PM_Case_Study_Prep_Guide.docx · Prompt
1 · Contract
Roadmap owned
Success = originations via partners
2 · Structure
Biggest levers
Frame the ambiguous problem
3 · Prioritize
Pick one feature
Can’t do everything
4 · Deep dive
Hypothesis + visuals
Bet → Design → Prove
What we’re looking for VERBATIM · the brief

We care about how you think, not whether you arrive at one specific answer. Specifically, we’re paying attention to:

Looking forWhere in this workbench
How you structure an ambiguous, open-ended problem Solve → IdeasFrame (lever equation before feature lock)
Your product judgment and instincts Solve → Bet (hyp + kill criteria + alternatives cut)
How you prioritize when you can’t do everything Frame levers ranked · Bet picks one · Constraints 4–6 wk MVP
How you reason about metrics and trade-offs Constraints + Prove (conversion ↔ risk; funded vs holdout)
How clearly you communicate your thinking Whole path: Case → Ground → Solve (Ideas→…) → Show; hour beats on Show
Any mockups, wireframes, prototypes to convey your thoughts on the feature Solve → Design (SMB · offer · partner admin)
Your AI workflow, what you prompted, where the model helped vs where you pushed back, what you threw away Show / AI → #effort process log (narrative beats)
On-the-fly: financing funnel (apply → offer → accept → fund) Ground → Space (#guide-funnel) · Journeys E2E · Prove funnel
On-the-fly: conversion vs risk tension Constraints (#conv-risk) · Prove trade-off cards
On-the-fly: distribute via partners, not only direct Plain (origination callout) · Space · Journeys (Cleo / PingPong / Cin7)

Also from guide (format): show thinking as mockups / Loom / prototype / wireframes · AI costs covered if needed.

Anthony’s lens OPERATOR · NOT the brief

Your annotations on the what the brief lists list, use these while framing in Solve → Ideas. Do not overwrite the brief above.

  1. Structure ambiguous problem → grounding on the space, how the business works
  2. Product judgment → levers in UI/UX, visibility, incentives for different parties, credit risk models, constraints on capital Kanmon uses to fund, regulatory constraints
  3. Prioritize → high leverage × relative risk for impact, back into sizing; effort and complexity; sequencing and gradual unlocks + growth
  4. Metrics / trade-offs → value, effort + time, regulatory complexity (plus conversion↔risk from the brief)

Also: clear communication while presenting · mockups. Work the lens in .

Format BRIEF
  • Up to 1 hour, live, conversational working session
  • They share a short prompt; discuss your proposal on the call
  • Use AI freely (Claude / GPT / Gemini), they’ll cover cost if needed
  • Show thinking however it lands: mockups, Loom, prototype, wireframes
  • For a product role here, how you work with these tools is as interesting as the conclusions
The space we’ll work in the brief · ON-THE-FLY

It will help to be comfortable reasoning about:

  • Financing funnel: a business applies, receives an offer, accepts it, and gets funded
  • Conversion vs risk: approving the wrong borrower is expensive, growth and credit quality pull against each other
  • Partner distribution: embed on the partner’s platform, not selling financing only as a direct Kanmon channel

Ground these in Plain + Space (+ E2E companions). Constraints locks the trade-off for solutioning.

Context they gave the brief gist
  • Kanmon = embedded lending infrastructure for vertical SaaS, marketplaces, FIs
  • Products: term loans, LOC, invoice financing, purchase-order financing
  • Origination = SMB on a partner platform takes financing
  • More originations ≈ more partners live · better surface · more eligible converting · broader product set
Self-check before the call
  • I can structure levers without jumping to a feature
  • I can say why I cut logos / new products / ML day one
  • I can walk apply → offer → accept → fund inside a partner
  • I can name the conversion↔risk tension without Credit owning my UX
  • Wireframes + AI pushback + thrown-away are one click away
In plain English: Software companies want to offer loans to their small-business customers inside the app those businesses already use. Kanmon is the company that actually lends the money and owns the credit risk. Partners embed Kanmon; SMBs never have to become a bank’s customer first.
What Kanmon does

Working capital inside partner software

  • Partner (SaaS / marketplace / payments) keeps the relationship and brand
  • Kanmon underwrites, funds, complies, services, lender of record
  • SMB gets cash for invoices, suppliers, growth, without leaving the partner tool
  • Public live examples: PingPong · Cleo InvoicePay
  • Products: term · line of credit · invoice · AP / PO-style financing
What the case is asking you

Grow funded loans that start inside partners

Success metric = originations through partner platforms, not traffic to Kanmon.com.

  • Map the big levers that move that number
  • Pick one feature and go deep (judgment > laundry list)
  • Show how you’d measure it, and not blow up credit quality
  • Bring visuals + how you used AI to get there
Origination ≠ Kanmon.com traffic
Partner distribution: An origination counts when an SMB on a partner’s platform takes financing. Kanmon embeds in software SMBs already use, you are not optimizing a direct-to-SMB acquisition funnel as the main success metric.
  • Partner, wants revenue + retention, afraid of looking spammy
  • SMB, wants cash at the moment of need, low friction
  • You can’t optimize one and ignore the other

Concrete journeys: Journeys stage · E2E HTML (Cleo · PingPong · Cin7) · Ecosystem extension · Revenue + when lending happens.

Live idea locked for this case

A pre-approval estimate and a shorter application, makes finishing the application easier while people are still in the form, and helps invite more of the right people into the funnel. Soft ≠ a hard loan amount. “Right moment” ads are set aside for this case.

Open · ·

Assumptions: Many partners may be live but quiet on originations. From conversations with the team: one product manager, low volume, little or no A/B testing, underwriting and data science own credit models, factoring and buy-now-pay-later as later product lines. Burn rate is a planning hypothesis. Live idea = a pre-approval estimate and a shorter application (so finishing apply is easier in flow); “right moment” is set aside for this case.

Learn the space fast: embedded lending = partner distributes, Kanmon lends. Funnel has a conversion↔risk tension. Distilled from market research, not the full paper.
Embedded lending in one breath

The loan lives inside software the SMB already trusts. Partner is the storefront; Kanmon is the balance-sheet lender.

  • FACT Capital + credit risk on Kanmon’s books (their positioning)
  • FACT Partner monetizes via rev-share / bps pitch (LinkedIn; take rate unpublished), not a kanmon.com FAQ line
  • FACT Payments data helps underwriting when present, not required to start

Depth: Revenue + partner model · summary (who pays · when lending happens · Cleo path). Incentives: deep dive · summary (SMB · partner · gaps).

Financing funnel BRIEF

Guide’s on-the-fly comfort: a business applies → receives an offer → accepts → gets funded.

Apply Offer Accept Fund

Origination equation adds upstream discovery: Eligible → See / surface → then the guide funnel. Don’t confuse “clicked CTA” with funded.

Eligible See Apply Offer Accept Fund
Conversion ↔ risk: More apps / offers can grow originations, approving the wrong borrower is expensive. Growth and credit quality pull against each other (guide). Soft invite by default; Credit owns hard decisions.
Product set FACT + roadmap
  • Term, lump sum, fixed payback (growth / inventory)
  • LOC, ongoing draw for working capital
  • Invoice, get paid on AR sooner (Cleo InvoicePay)
  • AP / PO, pay suppliers / purchase orders, preserve cash
Money & risk (high level)
  • Gross yield from interest / fees on funded book
  • Credit losses + cost of funds sit on Kanmon, scale with originations
  • HYP Spiking apps without soft eligibility can look good on volume and bad on losses
  • Growth PM owns invite timing/UX; Credit owns who gets an offer
Competitors at a glance
  • Direct-ish: Parafin, YouLend, Liberis
  • Adjacent: Pipe, Capchase, Clearco, eCapital, Lendflow
  • Platform wallets: Stripe Capital, Amazon/Shopify lending, Brex, compete for SMB wallet
  • Kanmon angle (self-report): multi-product + licensed balance-sheet lender + no payments prerequisite
Partner vs SMB jobs · distribution
  • Partner: monetize + retain; brand-safe surfaces; insights for QBRs
  • SMB: cash without leaving the tool; clear offer; fast fund
  • Quiet partner problem: integration shipped, financing rarely seen → originations stall
  • BRIEF Distribute through partners rather than selling only direct

Incentive map + gaps: incentives summary · full. Walk a real path: Journeys stage · E2E summary (Cleo InvoicePay · PingPong · Cin7).

Feature landscape, initial 2026-07-29

Table stakes vs rare · source-check Moments

Table stakes

  • White-label embed + provider holds risk/capital/compliance
  • Pre-approved-style offers · marketing kits · category rev share (Kanmon bps = LinkedIn pitch)
  • Integration ladder · progressive limits / renewals

Rare / Kanmon-relevant FACT

  • Multi-product portfolio (term + LOC + invoice + AP)
  • No payments prerequisite underwriting path
  • Workflow invoice/AP surfaces (live partners)
  • Liberis leads publicly on contextual offer APIs (Adverts + webhooks). Kanmon site does not

Capital Moments narrative

  • COMMODITIZED Liberis, Parafin, Stripe, Defacto, Pipe all sell capital-in-platform / at the right moment
  • Don’t claim invention of contextual offers
  • Differentiate via vertical workflow objects + product routing and/or quiet-partner ops, after funnel diagnosis

Top research questions (P0)

  • See → start → fund for 2–3 partners; where quiet partners stall
  • Offer/trigger APIs already behind kanmon.dev?
  • Hole = placement vs GTM/CS vs credit?

Peers: Parafin, Pipe, YouLend, Liberis, Stripe Capital, Defacto, finmid. Depth: kanmon-feature-landscape-summary-20260729.md

Stay grounded while solutioning. Labels: FACT INTERVIEW HYPOTHESIS
Stage & org INTERVIEW

Series A · ~21 · sole PM today

  • 4–6 week MVP = one wedge + one partner pattern.
Burn / runway HYPOTHESIS

~$360–660k/mo OpEx · base ~$500k

  • Fully loaded people + overhead @ ~21 FTE
  • Excludes cost of funds, credit losses, warehouse
  • Series A dollar amount not public, don’t invent
  • Bias to high-learning wedges; don’t growth-hack apps Credit can’t fund safely
Conversion vs risk BRIEF
Pull toward conversion
  • Surface at cash-need · lower start friction
  • More eligible SMBs entering apply
  • Partner intensity that actually gets seen
Pull toward credit quality
  • Wrong borrower is expensive (guide)
  • Soft invite; Credit owns offer/fund
  • Holdout + credit band on same scoreboard
  • Win originations inside an agreed credit-quality band
  • Refuse victory on clicks / starts alone. Prove uses incremental funded
Partner distribution
  • Partner controls intensity / brand feel
  • Moments need reliable partner events (invoice aging), data review gates rollout
  • Disclosures / marketing copy Legal-gated
Practical (4–6 wks)
  • 1 moment: invoice overdue → invoice financing
  • Inline CTA + partner intensity + soft eligibility rules
Aspirational / Phase 2
  • Full moment catalog · ML ranking · $-teases
  • Multi-vertical marketplace
  • Say: after proof + event volume + credit capacity

8 solutioning guardrails

1Kanmon is lender of record, no freelancing pricing / “approved for $X”FACT
2Soft invite / soft pull by defaultFACT
3Originations and credit-quality band on one scoreboardRULE
4Partner intensity configurable, no spammy forced modalsRULE
5One moment × one product × one pilot in 4–6 weeksSTAGE
6I own surface/start; Credit owns offer/fundBOUNDARY
7No rollout without partner event reviewTECH
8Copy/disclosures use Legal-approved templatesCOMPLIANCE
Tech / reg (distilled)
  • FACT CA CFL #60DBO-144925 cited; multi-state claimed; KYC/KYB/AML on Kanmon
  • ECOA still applies to business credit; UDAAP risk on in-product marketing claims
  • MVP stack: event → soft eligibility → CTA → existing apply → funnel events + holdout
  • Not realistic yet: ML ranking, cross-partner experimentation platform

Deep dive files: kanmon-market-research-20260729.md §§11–16 · kanmon-market-summary-20260729.md

The levers I weighed. The options I considered, and the one I chose. My full answer is under . The ranking and lead cards below are in plain language. Rougher working theories stay collapsed.
Earlier framing and pointers notes
Anthony raw notes

Organized notes: numerator/denominator, funnel drop-off, recovery, partner show-logic, pre-auth.

· full text

Money constraint: Good proposals touch partner funded volume × yield − losses − partner share − cost of funds, not vanity UX. Tag every idea Feature (existing shelf) vs Product line (e.g. Factoring / BNPL).
From the brief
  • Ambiguous, ground how the business works (partner distributes, Kanmon lends, funded = originations)
  • Judgment. UI/UX, visibility, incentives, credit models, capital, regulatory. PM feature vs Credit/Capital/Legal?
  • Prioritize, impact vs risk, sizing, effort, gradual unlocks and growth
  • Metrics, value, effort/time, regulatory complexity (plus conversion vs risk)

Working constraints: sole PM, low volume, little or no A/B, UW+DS own models; factoring/BNPL later, not default MVP. Case what the brief lists stays on Case.

Outcome card template

Not “we think this feature is good.” Fill: problem, lever, outcome metric, counter-metric/risk, effort, sequence, Feature or Product-line, money touch.

Low volume: save N potential funded per period, not “+X% at scale.”

Where I would focus first

Priority instinct: A pre-approval estimate and a shorter application so people who already started can finish more often. Next, improve decisions and plug preventable declines from bad or missing data, without trying to own credit risk models. Working assumptions for this case: every opportunity counts at low volume; focus on getting more funded paths through live partners (not hunting new logos first); assume financing is already visible enough (reopen “right moment” work if that is wrong). Depth: deep dive · summary · practice numbers ·
Practice volumes (simulation · not Kanmon’s real book)

In the simulation: about 51 funded per month; about 40% of starters submit (about 288 quit mid-form). A roughly ten-point finish-rate lift is on the order of fourteen extra funded / about $230–250k financed that month. · simulation

Live idea (see My pick for the full story)

A pre-approval estimate and a shorter application

Makes finishing the application easier while people are still in the form, and supports better invites for good-fit customers. Soft ≠ a hard loan amount. Recovery and preventable-decline work were seriously considered and ranked below.

1 · Lead Feature

A pre-approval estimate and a shorter application

Shorten the form with partner-held data; light fit check. Makes finishing easier in flow; better invites for good-fit customers. Never hard “pre-approved.” Recovery stays secondary.

2 · Outcome of the lead Job

Easier to finish while still in the form

Raise complete submits. Pre-approval estimate and shorter application is the feature that does this job. Fix UX and handoff where diagnosis points.

3 · Later Feature

Preventable decline plugs

Cut declines from bad or missing data, not by loosening credit policy. Underwriting and data science own the box.

4 · Later Feature

Decision experience (no model change)

Status, time-to-decision, and offer clarity, surface outputs with data science; don’t change model weights.

5 · Parallel ops Feature / ops

Simple ledger: completes, preventable vs true-risk declines, funded in-box

Judgment plus before/after under low volume, not multi-cell A/B theater.

Out of scope

Risk model / credit box changes

Underwriting and data science own models. Product partners on surfacing, don’t claim underwriting.

Set aside for this case

Partner show-logic / “right moment” ads

Reopen if financing turns out barely visible on live partners.

What I would do first, and what waits

In one line: convert the demand that already shows up before spending to create more of it. Most of the loss is people who start an application and never finish. Fixing that is cheaper and faster than buying more traffic. So the first three things all sit at the bottom of the funnel, and the top of the funnel work comes after. The first one does double duty. The same data that produces the estimate is what makes the application short.
OrderWhat it isWhere in the funnelImpactEffortRisk
P0 Pre-approval estimate, and an application short enough to finish
the one I chose. Also helps bring people in.
Start to submitHighMediumMedium
P1 Fix declines caused by bad or missing information
people who finished, then got turned down for a fixable reason
Submit to approveMedium to highLow to mediumLow
P2 Bring back people who dropped out mid application
save their progress, invite them back
Start to exitMediumLowLow

Why this order: P0 moves the biggest number and also helps the top of the funnel. P1 rescues people who did all the work and were turned away for something fixable, so they are closest to funded. P2 helps too, but those people are further away and less committed.

The assumption this rests on not yet confirmed

P0 only works if partners already hold enough information to produce a useful estimate, things like business identity, revenue and a bank connection. I am assuming that, not confirming it. It is the first thing I would check with a real partner. If the data is thinner than I think, the estimate gets vaguer and the win gets smaller.

The same thing, said plainly

Kanmon makes money when a small business borrows through a partner's software. Plenty of owners start an application and give up, or finish and get turned down for something that was fixable. Before spending to get more people to start, it is worth making sure the ones already trying get through.

  • First: tell them up front roughly how much they could get, and use what the partner already knows so the form is short.
  • Second: stop losing people over a wrong field or a missing document.
  • Third: let people pick up where they left off.
Where AI does the work

For P0, reading partner data and documents to work out the estimate and fill the form. For P1, sorting declines into real credit calls versus fixable data problems, and flagging the fixable ones before the final answer. For P2, working out who is worth inviting back and when. In all three, AI helps produce and present the information. Underwriting and data science still decide who gets approved.

Working theories (rough, earlier) archived
T1 · Distribution working

Partner-only GTM; originations = funded via partners

Evidence: partner-vs-direct brief; no public D2C · Case: don’t invent mix %; don’t pitch D2C

T2 · Diagnose fork working

Capability vs incentive, diagnose first

Evidence: incentives §4–6; SPIFs absent-public · Case: see→start→fund before SPIFs or louder Moments

T3 · Placement working

Portal ≈ downstream CVR; leverage often upstream

Evidence: ecosystem writeup; Cleo Finance-on-invoice · Case: don’t assume features live on *.kanmon.com

T4 · Three constraints working

Funnel × credit box × capital

Evidence: Anthony thinking + Kanmon assumes risk · Case: PM owns funnel inside box; widening = Credit/Capital

T5 · Misalignment working

Biggest fight: volume vs risk vs partner brand

Evidence: incentives deep dive §4 · Case: Prove = incremental funded inside credit band

T6 · Product mix source-check

Invoice/AP = embedded-native core

Evidence: source-library FAQ density · preferred mix HYP · Case: bet moments on cash objects

T7 · Capital Moments Feature one hyp among many demoted · A2 source-check

Upstream object CTAs can move see→start, falsifiable; not default under A2

Default: funnel throughput first · Rise if A2 dies · Kill if: surface healthy or stall ≠ placement

T15 · Funnel math Anthony notes working

Originations = move numerator or denominator

Denom: top-of-funnel (see / candidates) · Num: funded via partners = UX + recovery + credit · Notes

T16 · Surface logic Anthony notes working

Partner show-logic, if I don’t need funds, do I see Kanmon?

Case: diagnose need-gated vs always-on surface before louder CTAs

T17 · Portal ownership Anthony notes working

Kanmon owns customer from *.kanmonhq.com land

Case: recovery / apply UX / drop-off = Kanmon-owned after portal land

T18 · C1 scarcity working

Every opportunity counts (low volume)

Case: save N funded paths · recover before net-new · judgment + before/after

T19 · A1 assumption source-check

Goal = funnel throughput, not new logos

Kill: live inventory can’t originate → partners-live returns

T20 · A2 assumption source-check

Visibility assumed solid (partners incentivized)

Kill: buried tabs / no impressions / won’t push → Moments or incentives

Open questions archived
  • Where does each quiet partner stall: see / start / offer / fund?
  • Is *.kanmon.com portal eligibility-gated or open educate/apply?
  • Partner admin intensity / rev-share, felt or pitch-only?
  • Capital capacity vs credit-box tightness, which binds first?
  • Offer/trigger APIs already behind kanmon.dev?
  • What particular project is planned for a new hire? Factoring/BNPL timing vs live-partner originations?
  • Impression / CTA / click-to-convert data from partners? Recovery campaigns for abandoned apps?
  • Non-risk declines, holes to plug? Multi-product eligibility + cross-partner apply awareness?
  • Soft pre-auth / partner-data share / pre-fill?
  • Share of stalls = identity / KYB data quality? Entity-resolution vendor today?

Lead cards

The two lead cards below are the proposed stack behind My pick. “Right moment” and other demotions sit under Thrown away.

Lead · a pre-approval estimate and a shorter application lead bet Feature

A pre-approval estimate and a shorter application

Makes finishing and submitting easier while people are still in the form, and helps invite more of the right people. Soft ≠ hard approval amount. Full card below · · ·

Same feature · Two payoffs Feature

Smoother submit, and better invites for good-fit customers

· partner types

Lead write-up · a pre-approval estimate and a shorter application settling Feature

A pre-approval estimate and a shorter application so more people finish while still in the form

Why this lead: for this case I focus on funnel throughput on live partners, assume financing is visible enough, and put completions first. Mid-apply quit is the largest pool of people who already showed intent. In the simulation, only about four in ten starters finish, so preventing quit beats chasing people after they leave. Full logic: drive-down doc · sizing ·

Feature: A pre-approval estimate and a shorter application from partner-held data (progress, handoff, bank and business-identity help, after diagnosing top quit reasons). Secondary: email and one-time code to resume for people who still leave.

Cut for now: “right moment” ads, recover-first, preventable-declines-first, and owning risk models. Stop leading here if almost everyone already finishes, or if most quitters were never eligible.

Two-payoffs write-up · a pre-approval estimate and a shorter application Feature

One capability that helps finishers finish and invites the right starters

First payoff: partner-shared data supports a soft fit check and prefill, so the form is shorter and fewer people quit mid-apply. Second payoff: honest copy on time and effort, plus selective invites for good candidates (“easy to quickly check if you can get funding”).

Partner data: start with richer types (invoice, wallet, ERP or purchase order, payments, staffing); defer thin SaaS; expect new data pipelines. partner types × data

Fits the lead: A pre-approval estimate and a shorter application is the feature; easier finish-in-flow is the job it does. Soft language only, never hard “approved for $X.” Credit and legal gate stronger claims; underwriting and data science own the box.

Explore: two payoffs notes · if a few-week ship is too heavy, ship UX-only finish-apply improvements first, then pre-approval estimate

Thrown away / demoted not the bet

“Show financing at the right moment” (Capital Moments). Set aside for this case under the visibility-is-solid-enough assumption, and because a generic right-moment pitch is common. Reopen only if financing is barely seen on live partners.

Kept in the archive so the trail of what I explored stays visible.

Other feature areas archived
Quiet-partner tooling Feature

Raise attach on live partners with health checks, QBRs, and playbooks.

CS recommend-capital Feature

Assisted attach. Watch adverse selection.

Rev-share dashboard Feature

Make partner incentive felt. Don’t invent take-rate percentages.

Soft prequal teasers Feature

Raise start quality. Credit and legal gated. Never hard pre-approved.

Product router Feature

Match need to existing products. Not a new product line.

Fee calculator parked

Weak unless tied to start or accept.

SPIFs / promos parked

After “works but unused.” Never first.

Direct SMB parked

Off the case metric (partner originations only).

Partner onboard + ops Feature

Raise partners live and KYB throughput. Diagnose before louder ads.

Factoring Product line parked

Roadmap, not the default short MVP.

Entity lookup / D&B DUNS adjacent Feature

Identity resolution to raise complete KYB and cut preventable identity rejects

Match or issue DUNS-class entity data without changing risk models. Not a full KYB/AML substitute. Don’t claim Kanmon uses D&B. 1-pager

Incentive map (seed)

SMB pays fees for speed / soft pull / stay-in-workflow / invoice·AP control. Partner gets rev-share pitch (% unpublished) + retention / no credit risk. Kanmon gets yield − losses − share; assumes credit risk.

priorities · framing · incentives · exploration park

Education: KYB = Know Your Business vs KYC. Soft pull + bank + KYB leads to a decision.

Written before I built anything

My raw notes

These are my working notes on the problem, straight from when I first read the prompt. I only added the section headings and put like thoughts together. The words, and the half-formed bits, are mine. Everything in the case traces back to something here.


What the interview is assessing

  • How you structure an ambiguous, open-ended problem – grounding on the space, how does the business work
  • Your product judgment and instincts – there are levers in UI/UX, there are levers in visibility, there are incentives for different parties, there are credit risk models, there are maybe constraints on the capital that Kanmon uses to fund, regulatory constraints
  • How you prioritize when you can’t do everything – identify high leverage x relative risk for impact – back into sizing; consider effort and complexity, consider sequencing and gradual unlocks + growth
  • How you reason about metrics and trade-offs – value, effort + time, regulatory complexity
  • How clearly you communicate your thinking
  • Any mockups, wireframes, prototypes to convey your thoughts on the feature

What Kanmon is / how the business works

Kanmon offers embedded lending - basically they enable businesses that benefit from their customers having funds, to extend funds to those customers for liquidity

  • I.e. a business needing to purchase stock or supplies ahead of the funds they have on hand
  • The existing product set spans term loans, lines of credit, invoice financing, and purchase-order financing.

The prompt (single success metric = originations)

If you owned Kanmon's product roadmap and your single success metric was originations driven through our partner platforms, how would you think about the biggest levers?

"An origination happens when an SMB on a partner's platform takes financing. More originations come from some mix of: more partners live, partners surfacing financing more effectively, more eligible SMBs converting, and a broader product set that fits more needs.”

  • The existing product set spans term loans, lines of credit, invoice financing, and purchase-order financing.
  • An origination happens when an SMB on a partner's platform takes financing. More originations come from some mix of: more partners live, partners surfacing financing more effectively, more eligible SMBs converting, and a broader product set that fits more needs.

Framing: numerator / denominator

How I understand this is that we need to understand what levers we have to net an increase in originations

We can change the numerator or the denominator-

Denominator is top of funnel (this can be defined as candidate customers for application, candidate customers that clicked through - or wider, all customers that use the partner platforms) - this is a function of partner customer volume; customers that see the opportunity; customers that are good candidates + have the opportunity/need to apply

Presumably some people click but abandon (due to UX friction - [patterns, clear copy], was just curious, time commitment, don’t have necessary info on hand)

Numerator is any SMB sourced from a partner that is issued funds - this is really a function of [UX + application recovery] + the credit risk model

  • Really interested in recovery campaigns; you didn’t finish (abandon cart); as well as any tooling that can streamline the application experience
  • Really interested in application drop off
  • Really interested in the logic partners have for showing Kanmon - if I don’t need funds, do I see Kanmon?

For exercise purposes I will consider Kanmon the owner of the customer from that point that the customer lands on [*partner specific portal].[kanmonhq.com](http://kanmonhq.com)


Funnel drop off

Funnel drop off

See -> Click - anyone that doesn’t click

Click -> Abandon before starting - anyone that bounces

Start application -> exit - anyone that didn’t complete for some reason

[unsure of pickup where you left off recovery campaigns]

Application submission -> rejection - I assume people get denied for various reasons but fr any reason thats not risk related, thats an opportunity to plug a hole

Also, need to consider that the funnels may be shaped slightly differently by product line - and customers may have optionality on product they choose (more than one product to choose from)

Also, curious about eligibility for multiple product lines at once - and then applying through different partners for the same or different product lines - obv Kanmon can detect internally but the customer may not know early in the funnel


Areas of opportunity

So there are a few areas of opportunity that make sense to explore

What leverage we have within the partners - this can be funding a campaign, this can be

From a low hanging fruit standpoint - who’s attempting and not converting? Do we receive success reporting from the partner? Do we know how many people land? Do we have impressions of CTA views?

  • Do we have data on the business segments that are clicking -> converting and can we see who’s not converting?
  • When a partner wants a customer to be successful, how do they leverage Kanmon? And are there any post engagement follow up from partners to their customers proposing Kanmon or lending as a solution?
  • Can SMBs independently engage Kanmon with attribution from a partner?

Can we pre-authorize SMBs based on their relationship with the partner? Can we roll out a pre-check program that enable customers to authorize partners to share their info to Kanmon for pre-auth? For lending, like a soft check or at least pre-fill applications

Fuller exploration and later theories: .
Full partner E2E examples inlined below · Mermaid via CDN · works on file://
Open standalone HTML ↗
You are here: Big picture (steps before the pick)
More funded loans come from a chain of steps. Partners must be live. People must see financing. People must finish applying. Then offer, accept, and funding. For this case I assume seeing financing is already solid enough, so I lead with a pre-approval estimate and a shorter application to make finishing easier in flow. “Show it at the right moment” is set aside unless that visibility assumption fails.
After the big picture, open my pick

A pre-approval estimate and a shorter application

Makes finishing the application easier while people are still in the form, and supports better invites for good-fit customers. Soft ≠ a hard loan amount.

Now
1. Big picture
Map the steps
Next
2. Cut
What not to do first
Then
3. Pick one
pre-approval estimate and shorter application
Deep
4. Numbers
Size it on the flow chart
Partners live × See financing × Finish apply × Offer × Accept × Funded
Slow
More partners
Sales and setup. Real, but not my first build.
Assume OK
Seeing financing
Assumed fine for this case. Reopen if wrong.
Focus
pre-approval estimate and shorter application
Shorter form, light fit check, fewer quitters.
Credit
Offer to fund
Credit team owns approvals. Soft invite only.
Slow
New products
New loan types later. Not this first ship.
When you cannot do everything

Cut first: hunting logos, launching a brand-new product on day one, treating “right moment” as the whole answer, or leading with email chase-back.

Wait on changing who gets approved, and on fixing every decline reason with one tool.

Do now: a pre-approval estimate and a shorter application so people who already started finish the form more often, and good-fit customers get clearer invites.

Check before building

Finish rate by partner, and the top reasons people quit mid-form.

Which customer fields partners already have for prefill.

If almost everyone already finishes, this is the wrong lead.

·

You are here: My pick (the thing I would build first)
My pick

A pre-approval estimate, and an application short enough to finish

Work out roughly how much a business could borrow before they ask, show them that number where they already work, and use the same information to make applying take about a minute.

The problem

In the simulation, only about four in ten people who start an application finish it. These are people who already showed interest. They leave because the form asks for things they have to go and find, and because nothing tells them whether it is worth the effort.

Why this fixes it

A real number answers "is this worth my time". The data behind that number then removes most of the form. Two problems, one piece of work.

The line I will not cross

It is an estimate, not a promise. Underwriting and data science decide who gets approved and on what terms. A firm offer of credit carries extra obligations, so the wording stays honest: "you could be eligible for up to", never "you are approved for".

See it in more detail.

Optional deep notes: architecture prose · partner copy · lift assumptions

How I got to this pick

Success is more funded originations through partner platforms. For this case I assume financing is already visible enough on live partners, so I set aside “show it at the right moment” as the main bet. I am also treating the company as small-PM and low-volume: little or no A/B testing, and underwriting plus data science own credit models.

In the simulation, about 51 loans fund in a month, and about 288 people start then quit before submitting. That mid-apply quit pool is the largest product-owned group of people who already expressed interest. The flow chart on the Numbers page shows those drop-offs by partner and product.

Full reasoning write-up · Opportunity sizing ·

Why this is the lead

A pre-approval estimate and a shorter application is the feature I would ship first. People who start have already raised their hand; shortening the form and a light fit check is the strongest product move under the visibility assumption.

In the simulation, if the share who finish after starting rises by about ten percentage points, that is on the order of thirteen to fourteen extra funded loans in a month, or about two hundred thirty to two hundred fifty thousand dollars more financed volume that month. That is a medium-size estimate, not a promise.

The same work also supports better upstream invites for good-fit customers, so a pre-approval estimate and a shorter application helps both finishing and starting well.

What I would still learn before shipping

I would diagnose the top reasons people quit mid-form: handoff or login friction, long or confusing forms, bank-link problems, business-identity stuck points, and “save for later” with no return.

I would confirm which partner fields can be shared for prefill, with consent, starting with one rich partner type such as invoice or wallet workflows.

If almost everyone already finishes applications, or if most quitters were never eligible, I would stop leading here.

Ideas I seriously considered, then ranked below this lead

Application recovery (email and one-time code to resume). I took this seriously. It can still bring back people who leave with a partial application. I did not make it the lead because, in the simulation, most starters never finish. When finish rates are that low, preventing quit while someone is still in the form is the better first move. Recovery stays as a follow-on for people who still leave after in-form improvements.

Preventable declines after submit. I took this seriously too. Some declines come from bad or missing data rather than true credit risk, and those are worth fixing with validation and clearer data capture. I did not make it the lead because that pool is smaller and later than mid-apply quit. There is also no single fix for every decline reason. I would diagnose decline reasons after the finish-rate work, not before.

“Show financing at the right moment.” That was the assistant’s early suggestion. I set it aside for this case because I am assuming visibility is already solid enough, and because a generic “right moment” story is widely used by peers. If visibility turns out to be weak, I reopen that lane.

When I would stop or change course

If nearly everyone who starts already finishes, the opportunity is too small to lead with.

If most people who quit were never going to be eligible, polishing the form will not help enough.

If financing is barely seen on partner platforms, I would fix visibility before finishing-the-application work.

What is not this pick

Changing who gets approved (underwriting and data science own credit models). Launching brand-new loan types such as factoring or buy-now-pay-later as the first few-week project. Treating the old “right moment” wireframes under Screens as the answer, those are an archive.

· · pre-approval estimate notes

You are here: How it fits together
Where the estimate comes from, and who decides what. The partner shares information it already holds. Kanmon works out a rough amount from it and hands back a simple answer for each customer: eligible or not, and for roughly how much. The partner shows the pre-approval flow only to the customers who qualify. When someone applies, underwriting makes the real decision, exactly as it does today. Kanmon produces and presents the estimate. It does not decide who gets approved. The diagrams below go from the whole system down to individual products.
In one line, before the diagrams

Information the partner already has, turned into a per-customer answer: eligible or not, and for roughly how much. Kept up to date, and handed back so the partner can show the flow to the customers who qualify. Everything after that is the normal application, just much shorter.

4 · End-to-end sequence

Partner calls soft; soft returns state + token; apply still goes to UW for the hard path.

sequenceDiagram
  actor SMB as Merchant
  participant Partner as Partner app
  participant Soft as pre-approval estimate service
  participant Apply as Kanmon apply
  participant UW as Underwriting

  SMB->>Partner: Works in product
  Partner->>Soft: SoftCheck with partner customer id and context
  Soft-->>Partner: Soft state plus prefill token
  alt Soft strong or weak
    Partner-->>SMB: Soft invite or quiet check CTA
    SMB->>Apply: Open apply with prefill token
    Apply-->>SMB: Prefill form
    SMB->>Apply: Finish and submit
    Apply->>UW: Hard decision path
    UW-->>Apply: Offer or decline
  else No need or not now
    Partner-->>SMB: Dismiss without shame
  else Soft exclude or data unavailable
    Partner-->>SMB: No soft claim or standard apply only
  end

1 · Systems and trust boundary

pre-approval estimate ranks invite and hydrates apply. It does not mint a hard approval amount. UW and credit models own the hard decision after submit.

flowchart TB
  subgraph PartnerZone["Partner trust zone"]
    PApp["Partner app"]
    PData["Partner features and events"]
  end

  subgraph SoftZone["Kanmon soft layer - design assumption"]
    SoftSvc["pre-approval estimate service"]
    Prefill["Prefill token store"]
  end

  subgraph HardZone["Kanmon hard credit - UW owns"]
    Apply["Kanmon apply"]
    SoftPull["Soft bureau pull on apply"]
    UW["UW and credit models"]
    Fund["Fund and service"]
  end

  PApp -->|"customer id plus context"| SoftSvc
  PData --> SoftSvc
  SoftSvc -->|"soft state plus prefill token"| PApp
  SoftSvc --> Prefill
  PApp -->|"open apply with token"| Apply
  Prefill --> Apply
  Apply --> SoftPull
  SoftPull --> UW
  UW --> Fund

  SoftSvc -.->|"does not set hard amount"| UW

2 · Who enters pre-approval estimate

Program eligibility decides who gets a soft check. Cash need decides when to nudge, not whether soft runs.

flowchart TD
  Start["Partner surface in financing program scope"] --> Prog{"Program eligible?"}
  Prog -->|no| Quiet["No soft UI or quiet unavailable"]
  Prog -->|yes| Run["Run pre-approval estimate for everyone in set"]
  Run --> NeedNote["Need cash is for timing only - not a soft gate"]
  NeedNote --> Soft{"Soft state?"}
  Soft -->|strong| Invite["Invite plus prefill path"]
  Soft -->|weak| Careful["Careful check CTA or quieter invite"]
  Soft -->|unavailable| Std["Invite without soft claim - more fields later"]
  Soft -->|exclude| Suppress["Suppress campaign - no eligibility promise"]
  Invite --> NoNeed{"Merchant needs financing now?"}
  Careful --> NoNeed
  NoNeed -->|yes| ApplyPath["Open apply with prefill"]
  NoNeed -->|no| Dismiss["Dismiss snooze or no-need - no shame"]

3 · Where soft plugs into the funnel

Bird 2 helps See and Start. Bird 1 shortens Apply via prefill. Offer and Fund stay on hard UW.

flowchart LR
  See["See"] --> Start["Start"]
  Start --> Apply["Apply"]
  Apply --> Submit["Submit"]
  Submit --> Offer["Offer"]
  Offer --> Fund["Fund"]

  SoftInv["Soft invite Bird 2"] -.-> See
  SoftInv -.-> Start
  Prefill["Prefill shorter form Bird 1"] -.-> Apply
  HardUW["Hard UW unchanged"] -.-> Offer
  HardUW -.-> Fund

5 · Data flow overview

Partner sends features and events, not a raw ledger. Soft returns coarse state, never hard score or approve amount.

flowchart LR
  subgraph In["Partner to Kanmon"]
    ID["Identity seed"]
    Perf["Performance features"]
    Obj["Object cash context"]
    Risk["Partner risk flags"]
    Cons["Consent"]
  end

  Soft["pre-approval estimate service"]

  subgraph Out["Kanmon to partner"]
    State["Soft state codes"]
    Tok["Prefill token"]
    Evt["App lifecycle events"]
    Supp["Campaign suppress"]
  end

  ID --> Soft
  Perf --> Soft
  Obj --> Soft
  Risk --> Soft
  Cons --> Soft
  Soft --> State
  Soft --> Tok
  Soft --> Evt
  Soft --> Supp

6 · Product overview

Same soft contract; the trigger object and partner moment change by archetype.

flowchart TB
  Soft["Pre-approval estimate plus shorter application"]

  Soft --> A["A Invoice EDI"]
  Soft --> B["B Wallet"]
  Soft --> C["C ERP PO"]
  Soft --> D["D Payments"]
  Soft --> E["E Staffing"]

  A --> ObjA["Object: unpaid invoice"]
  B --> HubB["Hub: Financing tab plus volume spike"]
  C --> ObjC["Object: PO or stock risk"]
  D --> VolD["Volume plus boarding flags"]
  E --> PayE["Payroll vs unpaid client invoice"]

A · Invoice / EDI

Object CTA on the unpaid invoice; dismiss leaves invoice work intact. Partner never shows approved advance $.

sequenceDiagram
  actor SMB as Supplier ops
  participant Cleo as Invoice EDI partner
  participant Soft as pre-approval estimate
  participant Apply as Kanmon apply
  participant UW as Underwriting

  SMB->>Cleo: Open unpaid invoice
  Cleo->>Soft: SoftCheck with invoice context
  Soft-->>Cleo: Soft state plus prefill token
  alt Soft strong or weak
    Cleo-->>SMB: Finance CTA on this invoice
    SMB->>Apply: Start with prefill
    Apply-->>SMB: Entity and invoice prefilled
    SMB->>Apply: Submit
    Apply->>UW: Hard path
    UW-->>Apply: Offer or decline
  else No need for this invoice
    Cleo-->>SMB: Dismiss - invoice work continues
  end
More product sequences · wallet, ERP, payments, staffing

B · Wallet

Financing tab or volume-spike moment; AP vs term picked after soft invite. Cooldown if they leave.

sequenceDiagram
  actor Seller as Seller
  participant PP as Wallet partner
  participant Soft as pre-approval estimate
  participant Apply as Kanmon apply
  participant UW as Underwriting

  Seller->>PP: Open Financing or hit volume spike
  PP->>Soft: SoftCheck with wallet tenure and volume
  Soft-->>PP: Soft state plus prefill token
  alt Soft strong or weak
    PP-->>Seller: Check if you qualify CTA
    Seller->>Apply: Start with KYC seed prefill
    Note over Apply: Product pick AP vs term after soft invite
    Seller->>Apply: Submit
    Apply->>UW: Hard path
    UW-->>Apply: Offer or decline
  else Not now
    PP-->>Seller: Leave Financing - cooldown on campaign
  end

C · ERP / PO

PO-tied soft invite into Capital hub. Soft UI must not imply stacking products.

flowchart LR
  PO["PO approved or stock risk"] --> Soft["pre-approval estimate"]
  Soft --> Hub["Capital hub CTA on that PO"]
  Soft --> Skip["Not now"]
  Hub --> Prefill["Prefill entity plus PO"]
  Prefill --> Submit["Submit"]
  Submit --> UW["Hard UW"]
  UW --> Fund["Fund AP"]

D · Payments processor

Volume + boarding features; chargeback or reserve → soft exclude. Less object-tied than invoice/PO.

flowchart TD
  Board["Boarding complete"] --> Soft["pre-approval estimate"]
  Vol["Rolling volume features"] --> Soft
  CB["Chargeback or reserve flag"] --> Soft
  Soft -->|exclude| Quiet["No soft claim"]
  Soft -->|strong or weak| CTA["Hub CTA - less object-tied"]
  Soft -->|unavailable| Std["Standard apply only"]
  CTA --> Prefill["Prefill entity plus volume context"]
  Prefill --> UW["Submit then hard UW"]

E · Staffing

Soft still runs in program scope during quiet weeks; dismiss is normal when there is no cash pressure.

sequenceDiagram
  actor Agency as Staffing agency
  participant Vert as Staffing partner
  participant Soft as pre-approval estimate
  participant Apply as Kanmon apply
  participant UW as Underwriting

  Note over Vert,Soft: Soft still runs if in program scope - even in quiet weeks
  Vert->>Soft: SoftCheck with unpaid client invoice and payroll calendar
  Soft-->>Vert: Soft state plus prefill token
  alt Soft strong or weak and cash pressure
    Vert-->>Agency: Bridge payroll CTA
    Agency->>Apply: Prefill employer entity
    Agency->>Apply: Submit
    Apply->>UW: Hard path
  else Quiet week - no need
    Vert-->>Agency: Dismiss is normal
  end

Optional deep notes: architecture prose · standalone diagrams HTML

You are here: The demo
Click through the application twice, once each way. Switch between Today and With a pre-approval estimate at the top of the demo. The counters in the corner show screens, fields typed, documents and time. Numbers are simulated. The estimate is never a binding approval.

Open the demo full screen for presenting. Arrow keys move between screens, and pressing T switches between the two versions.

You are here: Screens (archive of an earlier idea)
ARCHIVE · NOT THE LIVE ANSWER
These drawings belong to an earlier idea: show financing at the right moment. That is not my live pick. The live pick is a pre-approval estimate and a shorter application, so finishing the application is easier in flow. Live pre-approval estimate and prefill screens are under . For real partner flows today, also use Partner examples.

· reasoning ·

Archive only. Kept to show what was explored and thrown away. These screens are not the live answer.
You are here: Numbers (simulated volumes and flow chart)
Does the idea pencil out? Win means more people get funded because a pre-approval estimate and a shorter application help more finish applying, and because forms are better pre-filled. Do not win by approving riskier loans. Numbers below are practice examples, not Kanmon’s real book. Scoreboard first, then the flow chart showing where people leave.
Push for more funded loans
  • Main move: a pre-approval estimate and a shorter application, raise the share who finish after starting, so more complete apps and more funded loans.
  • Pre-fill: less time in the form; later, better return paths for people who still quit.
  • Bonus: better invites only for customers who look like a fit.
Don’t break credit quality
  • Stay inside the approval band. Don’t loosen who can get a loan.
  • Invite softly. Never say “you’re approved for $X” without Credit.
  • Don’t spam partners or over-promise if the light check is too loose.
Example month: about 51 funded, about 288 start then quit. If finish-rate rises about 10 points (+1,000 bps · ~40% → ~50% · +25% relative), about $230–250k more funded that month. Sizing notes · number sheet · · assume financing is already visible enough
Practice numbers · not disclosed

Simple scoreboard

Funded / month
~51
loans paid out
Start to finish
40%
main leak · move ~40% → ~50%
Quit mid-apply
288
already interested
Medium upside
~$240k
extra funded / mo · +10 pts (+1,000 bps · +25%)
Steps from “saw it” to “funded”
8,000
Saw it
960
Clicked
480
Started
192
Submitted
82
Offer
51
Funded

Why people quit (guess mix): login/handoff about 22%, long form about 28%, bank link about 20%, business ID about 18%, “later” and never return about 12%. I’d fix in-form blockers first.

Clicks alone don’t count. Approval quality still matters.

All volumes simulated, not Kanmon disclosed
Funnel lift, pre-approval estimate · baseline vs with-program side by side · Small / Medium / Large assumptions
Open full page ↗

Three things to see at a glance: new people enter from soft invite, existing starters finish (thinner abandon), and funded count / financed $ rise vs the simulated baseline.

New inflow
+80
Extra clicks (soft invite · Medium)
Saved completers
+48
Finish ~40% → ~50% · +10 pts (+1,000 bps · +25%)
Total funded lift
+18
51 → ~69 funded / mo
$ lift
+$330k
Extra financed / mo (simulated)

Callouts track the size selected in the chart (Small / Medium / Large). Math: lift assumptions.

Flow chart (practice data) · by partner and product · drop-offs under each step · “no click” faded for this case · focus on people who start then quit
Open full page ↗
You are here: Show (approach story and process log)
Live idea

A pre-approval estimate and a shorter application

Makes finishing the application easier while people are still in the form, and supports better invites for good-fit customers. Soft ≠ a hard loan amount. In the simulation, about four in ten starters submit. Recovery and preventable-decline fixes were seriously considered and ranked below. “Right moment” ads are set aside for this case.

·

Approach story

How I approached this case

The brief asks how I would own Kanmon’s product roadmap if success means more originations funded through partner platforms. I structured the problem first, then picked one idea and went deep. Here is the path I took.

  1. Clarify the win. An origination is a business on a partner’s platform taking financing and getting funded, not traffic to Kanmon’s own site.
  2. Map the big steps before picking a feature. More live partners, more visibility, more people finishing the form, more offers accepted, and capital to fund them; new loan types come later.
  3. Test “show financing at the right moment,” then set it aside. It was the AI’s early idea, but it is a common pitch and assumes people already finish applying, which the numbers say they do not.
  4. Fold in how the company runs. One PM, a wide funnel, low volume, little A/B, and underwriting owning credit, so the feature grows volume without changing who gets approved.
  5. Size the problem with a simulation. In it, about 51 loans fund a month while about 288 people start and quit before submitting, and a flow chart makes the drop-offs visible.
  6. Choose the lead feature: a pre-approval estimate and a shorter application. People who start already want it, so I cut friction in the flow instead of chasing cold demand; the estimate stays soft, never a promised amount.
  7. Show how it works. Data partners already hold shortens the form and powers a light “you look like a fit” check; credit and legal gate anything stronger.
  8. Name the second payoff. The same estimate makes an honest, low-effort invite for good-fit customers, so it helps new starters too, not only those already in the form.
  9. Rank recovery second. Email and a resume link can bring some people back, but when most never finish, fixing the form beats chasing people after they leave.
  10. Rank preventable declines later. Some declines are bad or missing data rather than real credit risk; worth fixing, but a smaller and later pool than mid-apply quit.

Where this stands now: the live answer is a pre-approval estimate and a shorter application, so finishing apply is easier in flow. The “right moment” screens are kept only as an archive of an earlier idea, not the pitch.

Detailed process beats are below if you want the full work log. Simulation numbers are illustrative, not Kanmon disclosed data.

Case outline. How originations work through partners, which steps matter, why I picked a pre-approval estimate and a shorter application, the simulated size of the quit-mid-form pool, and how I used AI, including what I threw away. One-line version: structure the steps, pick a pre-approval estimate and a shorter application so finishing apply is easier in flow, show the simulated sizing, then walk the process log.
Optional timed walkthrough notes private prep
Rough hour order
MinutesPhaseWhereCover
0–3OpenPromptRestate success: funded originations through partners
3–12GroundingBusiness basics and examplesHow lending works inside partners; company limits
12–20StepsBig pictureWhich steps matter; assume visibility is solid enough for this case
20–30My pickMy pickpre-approval estimate and shorter application; finish-in-flow job; alternatives weighed
30–42SizingNumbers and flow chartSimulated quit pool; upside; credit quality guardrails
42–52ProcessShowApproach story; what was thrown away
52–60Open questions, Where do people quit mid-form? What partner data can be shared?

If the conversation jumps to “show the feature,” spend about a minute on why, open My pick, then the flow chart. Drawings under Screens are an archive of an earlier idea, not the live answer.

Core points
  • Live idea: a pre-approval estimate and a shorter application
  • Job: easier to finish while still in the form; also better invites
  • Recovery and preventable declines were seriously considered
  • Simulated sizing; “right moment” was set aside
Plain summary

They already started, so interest is there, but in the simulation most never finish. I would ship a pre-approval estimate and a shorter application so the application is easier to complete while they are still in it. That same work also helps bring more of the right people into the funnel. Recovery and preventable-decline fixes matter, but they are not my lead.

Effort / process log NARRATIVE DEFAULT

How the work was executed

Chronological story beats. Tap a beat for full prompt, workbench autonomy, tools, and thrown-away notes.

All process beats (expand/collapse individually)
Loading efficiency…
Secondary: swimlanes · judgment scorecard · time-in-stage

Dense lanes kept for completeness, not the primary read.

You
Autonomy
Agent

Judgment scorecard

DecisionAI tendencyMy interventionWhy

Time-in-stage EST

Relative efficiency = autonomy hits per user prompt. Costly minutes = judgments (which lever, what to cut, what not to claim).

Evolution / snapshots
  • Full changelog · open any prior version
  • v013a · pre-docx-alignment
  • v013b · e2e-in-workbench
  • v014 · docx-alignment
  • v015 · e2e-ui-mocks
  • v016 · log-fold-exploration
  • v017 · log-ideas (Ideas under Ground)
  • v018 · ideas-in-solve · Ideas = Solve main scratch · source-check logged (b37–b38)
  • v019 · back-to-live-banner · sticky ← Back to live
  • v020 · framing pass · b42 · label only, no frozen folder
  • v021 · Anthony raw notes → Solve · Notes · T15–T17 / F14–F15 · b43
  • v022–v029 · leverage → Sankey beats b44–b52 · labels only, no frozen folders
  • v030 · clarity-approach-b56 · catch-up freeze through b56–b57 · openable snapshot. LIVE ahead (architecture / experience / funnel lift embeds, b58–b63)
  • LIVE · this file · pre-approval estimate and shorter application · Architecture / Experience / Funnel lift embeds · process log through b63

Changelog: kanmon-case-snapshots/changelog.html · Last freeze v030; LIVE includes post-v030 embeds (b58–b63). No capture script, copy live into a new vNNN-…/ folder before the next freeze.

Raw timeline filters

Loading…

Artifacts

Thrown away