Brenman Talent

The tool that managed our growth was also taxing it.

Brenman Consulting ran its people operations on Zoho People, a capable HR product priced as a per-employee, per-month subscription. It did the job. But the pricing model had a quiet problem built into it: every new hire added another paid seat, and every seat pushed a recurring bill that only ever moved in one direction.

For a company that intends to grow, that is the wrong shape of cost. The software meant to support hiring was quietly making each hire a little more expensive. The more successful the company got, the more the back office cost to run.

We were paying per seat for a problem that does not change per seat.

A single company's HR needs are remarkably stable. Attendance, leave, policies, a directory, onboarding, approvals. None of it gets more complex when you add a person. It simply gets applied to one more row. Yet a per-seat subscription charges as if every new employee were a brand new problem to solve.

• The cost scaled with headcount, not with the value or the complexity of the work being done.
• Most of what the subscription bought was generic flexibility built for thousands of companies, almost none of it shaped to how Brenman actually runs.
• Our real policy lived in a leave-policy document, with its own leave types, accrual, carry-forward, the sandwich rule, and director approvals. The tool could not read it, so we kept bending our process to fit the software.

The reframe was simple: per-seat economics only make sense when the problem is open-ended. Ours was bounded. A bounded problem can be owned instead of rented.

Build it once, scope it to exactly what we need, run it for almost nothing.

The plan was two weeks and a clear rule: build precisely what Brenman uses and nothing it does not. That constraint is the whole reason two weeks was enough. Instead of configuring a generic product, I designed and built one that encodes Brenman's own policy directly. The leave engine reads the company's leave-policy PDF and turns it into real entitlements; attendance, approvals, and digests run on scheduled automation rather than on a per-seat headcount.

The economics flip the moment you own the surface. Where the subscription charged for every seat, this runs on infrastructure with a flat, near-zero marginal cost. Adding the hundredth employee costs the same as adding the eleventh, which is to say it costs nothing.

"The subscription charged me more for hiring. The platform charges the same whether we are ten people or a hundred."

Designing and building at the same time, with AI in the loop, kept the two-week timeline honest. Every screen earned its place against one test: does an employee or an admin actually do this, the way Brenman does it? If the answer was no, it did not get built.

For the look and the interaction patterns, I studied how modern HR platforms like Deel make genuinely heavy operational software feel calm and human. Deel hides enormous complexity behind a surface that never feels heavy, and that set the bar: an internal tool an employee opens without dread. I borrowed that restraint, then pared it down further to only the surfaces Brenman actually uses.

A glimpse, not the whole thing.

Preview of selected surfaces · Full walkthrough shared on request

Brenman Talent is live and internal. Below is a walk through the surfaces people use every day, then the hiring flow from the public careers page to the offer. The architecture, data model, and policy engine sit behind a request. Click any screen to open it full size.

Attendance that logs itself

One tap to clock in or out, present and absent days at a glance, and the full history below. Automated reminders nudge anyone who forgets, so the record stays honest without a manager having to chase it.

Leave, driven by the actual policy

Casual, earned, and sick balances up top, requests below, and one button to apply. The entitlements are not hand-configured: the engine reads Brenman's own leave-policy document and turns it into these numbers, accrual and carry-forward included.

Ask HR, the assistant that reads the policy

Instead of messaging a person, employees ask in plain language: who is on leave this week, what is the casual leave policy, how do I apply for earned leave. It answers from the company's real data and its actual policy documents, so the answer is the policy, not a guess.

The company, as a directory

Every employee as a card with role, department, status, and location. Search and filters make it the fastest way to find a person and see what they do.

How the company is actually structured

An interactive reporting tree built from the same people data, so the hierarchy is never a stale slide. Expand, collapse, and search to trace any reporting line.

A manager's view of their people

Direct reports in one place with live status, who is in, on a break, on leave, or done for the day, plus attendance at a glance. It appears only for managers and above.

Skills and certifications, tracked

Each person builds a profile of competencies and certifications, which gives the company a real map of who can do what instead of a guess.

A public careers page, hiring's front door

A clean public page that lists open roles in the company's own voice. It is the one surface meant for the outside world, and it is part of the same platform, not a separate website or a paid job board.

This is the one surface that is public. See the live careers page →

Apply in a single page

Each role opens to the full description and an application form: details, a resume, and a portfolio link. No account, no third-party redirect, just apply.

Every applicant in one pipeline

On the admin side, each role has a pipeline: new, reviewing, interview, offered, hired. Applications land here automatically, filterable and exportable, so nothing slips through a forwarded email.

Review the candidate without leaving the system

Open any applicant to read their resume inline, leave internal notes, and move them through the stages. A hired candidate converts straight into an employee record, closing the loop from applicant to teammate. That retired a separate job board and applicant tracker we no longer pay for.

Built as a PWA, installable anywhere

Brenman Talent is a Progressive Web App. There is no app store and no second codebase to maintain. On desktop or on a phone, an employee can add it to their home screen and open it like a native app, with the layout collapsing to a thumb-friendly shell with bottom navigation on mobile. Clocking in, applying for leave, or asking HR a question is one tap away, at a desk or on the move.

A lean stack chosen to keep both the build and the bill small.

Every piece was picked for two reasons: it let one person move fast, and it runs on free or low-cost tiers that do not scale with headcount.

Framework

Next.js 16, React 19, TypeScript

Interface

Tailwind CSS v4, shadcn/ui, Base UI, Framer Motion

Backend & data

Supabase Postgres, Auth, row-level security, RPC

Auth

Google OAuth domain-restricted to the company

AI assistant

Vercel AI SDK with Cerebras inference

Email

Resend invites, approvals, digests

Automation

Vercel Cron reminders, accrual, year-end rollover

Documents

Server-side PDF parsing for policy and holiday calendars

Reliability

Sentry, Upstash Redis rate limiting

App shell

Installable PWA web and mobile, no app store

Components

react-day-picker, cmdk sonner toasts, driver.js tours

Hosting

Vercel fixed, near-zero monthly run cost

The cost curve, flattened.

The old model charged for growth. The new one does not. The build was a one-time, two-week cost; running it is a fixed monthly bill that does not move when the team does.

Owning the tool beats renting the workflow.

Per-seat software is the right call when a problem is genuinely open-ended and you need a vendor to keep solving it for you. People operations for a single company is not that. It is bounded, well understood, and it changes slowly. When a problem is that stable, the per-seat premium is mostly rent on flexibility you will never use.

The other lesson was about leverage. Two weeks was only possible because I could design and build at once, with AI in the loop, and because I refused to build anything Brenman would not use. The cheapest software, it turns out, is the software that does precisely its job and nothing more.