the platform
Emily conversational AI
Three people open this platform. One is owed money, one is chasing it, and one cannot afford to pay it.
CollectCo, powered by Cedar Financial, is a recovery platform that agencies, banks, healthcare providers, utilities and subscription businesses buy to collect what they are owed. I designed the marketing site, the client dashboard, the onboarding, the AI agent surface and the consumer portal, around one uncomfortable fact: collections is an industry people already distrust before they open it.
Cedar Financial is a global receivables and recovery firm. CollectCo is the platform they sell. A business that is owed money, whether that is a hospital, a utility, a telecom, a landlord, a school or a software company, uploads its delinquent accounts and CollectCo runs the recovery: letters, calls, payment plans, disputes, and escalation to legal if it comes to that.
The commercial model is unusually design sensitive. There are three ways to buy: a flat fee for accounts under 120 days delinquent, where the client keeps everything recovered; a contingency fee for older accounts, where the agency takes a percentage; and a first party plan for accounts under 60 days, where collection happens under the client's own brand rather than the agency's. Those are not three price points. They are three different products with three different mental models, and the interface has to make the right one obvious in about fifteen seconds.
Under 120 days delinquent. A fixed price per account, from $9.95. The client keeps 100% of what is recovered, with a refund guarantee if targets are missed.
Over 120 days. The agency takes a percentage that scales with claim value and age, from 15% on large recent claims to 50% on small old ones.
Under 60 days. Collection runs under the client's own brand, so the customer relationship survives the conversation.
A conversational AI account concierge carrying three decades of collections practice, handling volume that human teams cannot cover consistently.
Most B2B SaaS has one user with one job. Collections has three parties whose interests are in direct opposition, and all three touch the same data.
A finance team that is owed money and has already written some of it off mentally. They want to know recovery rate, aging, and what is happening today, without learning a new discipline.
Collectors working a queue under legal constraint. Every contact is regulated, every note is potential evidence, and speed matters because recovery decays with age.
Someone having the worst financial month of their year, who has probably been contacted about this before, and who arrives expecting to be treated badly.
Having an account in collections doesn't have to be scary. We believe people come first and do everything we can to make the process easy and stress free.
Every element on that screen is doing trust work before it does transaction work. The greeting uses the person's name. The tone rule is stated in the second sentence, before any figure appears. Dispute status sits above the balance, because the first thing a person needs to know is whether this is even theirs. Credit reporting status gets its own block, because that is the fear underneath the call.
Only after all of that does the interface offer a way to pay, and it offers a route rather than an ultimatum.
That sentence is the entire design thesis. Collections products historically optimise for the collector: dense queues, aggressive urgency, and a consumer surface that exists only to take a payment. Cedar's actual competitive advantage, visible in years of public consumer reviews, is that their agents are patient and explain things. My job was to make the software behave the way their best agents already do.
A recovery platform has to sell certainty to a buyer, enforce compliance for an operator, and earn trust from a person who has every reason not to give it. Do the first well and you get an aggressive dashboard that scares consumers off. Do the third well and you get a soft product that finance teams do not believe recovers money. Most products in this category pick one and lose the other two.
Reframed as the design question I worked from: how might we make one system where the client sees rigour, the collector sees the next action, and the consumer sees a way out, without any of those three surfaces contradicting the others?
Three inputs shaped the work. Stakeholder sessions with Cedar's operations leadership, who know exactly where recovery leaks. A read of hundreds of public consumer reviews of the existing service, which is unusually honest research material because people leave reviews about collections when they are either relieved or furious. And a competitive audit of the category, where the pattern was consistent enough to be useful.
Positive consumer reviews almost never mention the payment. They mention an agent who was patient, who explained the balance, and who did not pressure. That is a designable experience, not a personality trait.
Category pricing is opaque by convention. Publishing the full rate card, including the unflattering 50% tiers, becomes a trust signal rather than a liability.
The product's own pricing proves it: under 60 days, under 120 days, over a year. Age is the single most important variable, so it had to be a first class object in the UI, not a filter.
Nobody wants an AI debt collector on reputation alone. It cannot be sold with a claim. It has to be demonstrated, which is where the simulator came from.
The hard constraint underneath all of it is regulation. Contact frequency, disclosure language, dispute handling and credit reporting are all governed, and a design that makes a prohibited action easy is a liability rather than a usability win. Compliance was treated as a material property of the components, not a legal review at the end.
Eight decisions carried the product. Each one came from one of the three parties, and each had to be checked against the other two before it shipped.
The homepage headline is People First Financial Solutions, over a night skyline rather than a stock handshake. In a category that sells fear, the differentiator is calm. The subhead names the actual promise, an empathetic approach delivering measurable outcomes, which is the only sentence that speaks to the finance buyer and the nervous consumer at the same time.
First party versus third party is the single hardest concept to explain to a new buyer, and it determines which plan they need. Rather than a comparison table, the page walks the account forward in time: early intervention under your own brand, then advanced recovery under ours. Time is the axis, because age is the variable that actually decides the answer.
Signing up to a collections platform means handing over a portfolio of delinquent accounts, which is commercially sensitive and emotionally awkward. The flow is deliberately paced: pick a plan, then sign up and submit claims, then track. Each screen asks for one category of thing, with a visible position in the sequence.
A finance lead opens this to answer one question: is the money coming back? So the stack is ordered exactly the way they think. Total outstanding AR, then recovered amount with a month over month delta, then how much has escalated to third party, then recovery rate as a single percentage. Aging analysis sits directly below, because that is the follow up question every time.
A claim record is a wall of transactions, contact attempts and status changes. Before any of that, the collector gets a plain language summary of what has actually happened on this account: hesitation early, good response to flexible repayment, a three month installment agreement, risk of default decreasing, reminders scheduled.
The ledger stays immediately underneath, unabridged. In a regulated product the summary is a reading aid, not a source of truth, and the interface has to make that hierarchy obvious or it becomes a compliance problem.
An unnamed AI in a collections product reads as an autodialer, which is exactly the thing consumers hate most about the industry. Naming her Emily and describing her as an account concierge that embodies 30+ years of credit and collection expertise does two jobs at once: it sets a service expectation for the consumer, and it tells the buyer that the AI is trained on domain practice rather than generic language.
The interface carries that through. Emily speaks in full sentences, offers options rather than demands, and every conversation surface shows a live state so nobody is unsure whether they are talking to a machine.
Every line here is doing compliance work as well as tone work: disclosing the creditor, offering the dispute right, stating the credit reporting consequence, and removing pressure to decide immediately. Written well, those are the same sentence.
The strongest thing in this product is the simulator. Instead of claiming Emily handles complex conversations, the platform lets a prospect pick a scenario and listen in, or join the call live. Open ended, unscripted conversation is offered first, which is the confident choice: the hardest case is the demo.
Scenarios are grouped by the consumer's actual position rather than by feature, starting with willing and able, because collections practice sorts people that way and it makes the taxonomy honest.
This is the surface the industry neglects, and it is the one that decides recovery rate. It opens by name, states the tone rule explicitly, then shows dispute status before it shows the balance, because the first question a person has is whether this is even real.
Cedar's blue does the structural work. It is the only saturated colour in the product, so it can carry primary action, active state and brand simultaneously without competing with the data. Everything financial is set in near black on white, because in a product about money the numbers should be the loudest thing on screen and nothing else should be.
Three rules held the system together across a marketing site, a client dashboard, an AI surface and a consumer portal. Money is never abbreviated. Every state has a word, not just a colour, because status carries legal meaning here. And no surface uses urgency styling, since anything that looks like pressure is both a brand violation and, in some jurisdictions, a regulatory one.
The project is live and still moving, so the honest position is that the outcome numbers are not mine to publish yet. What I can point at is the shape of the decisions: pricing made public, recovery performance moved to the first screen, an AI given a name and a scope, and a consumer surface built for a person rather than a payment. If those hold, the recovery rate follows, and if they do not, I would rather find out from the data than defend them here.
The consumer surface is designed from public reviews and stakeholder knowledge. That is good evidence, but it is second hand. Talking to people who have been through the process would sharpen the language more than any amount of internal review.
The moment Emily passes a call to a human is the highest stakes interaction in the product, and it got attention after the happy path rather than alongside it.
Disclosure requirements differ by jurisdiction across 100+ countries. Treating that as a content problem rather than a component property created rework.
Disputes are where trust is won or lost, and they are currently measured as a support cost rather than as a designed outcome.
Everything in this case study is shipped and running. Open the marketing site, the product site, or the pricing and plan pages and click through it yourself.
Pakistan's largest health platform, rebuilt around symptoms instead of specialties. +30% conversion.
Read case study →
Fourteen HR modules in one subscription platform, designed end to end in six weeks.
Read case study →
Three products in one system: customer app, worker app and admin console, all reading the same job.
Read case study →
NFT event ticketing that kills counterfeits and scalping, sold without ever leading with blockchain.
Read case study →
School payments and a learning app inside a platform with 16 roles and 4 languages.
Read case study →