Five high-stakes software projects
Some were stalled. Some were overestimated. Some had already failed. Here is what changed.
A five-year, $25M estimate
Under $2M actualglobal product catalog
Replaced in eighteen months, in flight, without taking the old one down.
$50M already spent
Under $1M to productionB2B marketplace
Four years and two failed attempts before us. Production in ten months.
After two abandoned attempts
90% cheaper to build and maintain800+ KYC bots
Both earlier attempts had been abandoned as too expensive to maintain.
Written the same way every time — what was wrong, what we did, what came out — so you can compare them rather than just read them. Each one opens with the architecture drawn out, because the decision is usually more interesting than the code.
These numbers are real, and we can prove it privately
Every figure on this site comes from an actual engagement. Client names are withheld because our agreements require it, not because the detail is thin — we can walk you through any of these in a conversation, and put you in touch with people who were there.
In detail
A five-year, $25M estimate. Delivered in 18 months for under $2M.
Everyone else quoted five years and $25M, and none of them had an answer for the price and availability errors running live that week. We separated the two problems: errors stopped in three months, full replacement and sunset in eighteen, under $2M, ETL ten times faster.
Read the case study Fortune 5 retailer · Marketplace · 2019–2023After $50M and two failed attempts, live in 10 months for under $1M
$50M and four years of previous effort, including a vendor who spent a year without reaching one at all. A scalable proof of concept in seven months for under $500K, production three months after that, live to 100,000+ merchants.
Read the case study Big 4 accounting firm · Compliance automation · 2015–2017KYC from 800+ sources, after two attempts were abandoned
Both earlier attempts failed on maintenance cost, not on writing the automation — which is exactly the trap AI coding tools walk people into today. 100+ bots in weeks, 800+ in production within six months, 90% cheaper to build and keep alive.
Read the case study Lending · Machine learning · 2019They asked whether to hire more collection agents
The useful question was a different one. A model reading 1,000+ features per loan sorted the queue into three piles so agents only ever saw one. 92% accuracy, workload halved, collections up about 10%, nobody hired.
Read the case study Fintech startup · Product build · 2017–2023Three platforms. Multiple brands. One startup budget.
Daily portfolio rebalancing with fiduciary and compliance requirements, on web and native mobile, white-labeled per partner. One codebase made it affordable. MVP in six months; the business ran for years and was eventually sold.
Read the case studyAlso delivered
More of the work from the same twenty years. Happy to talk through any of them.
A rebranding that had slipped several times
A configuration scheme met the new business requirement with marginal design change, cutting effort by about 20%. Parallel streams with fixed integration checkpoints landed it in time for the client’s annual sales conference.
Reporting that arrived weeks late
Channel data landing in an AWS data lake, aggregated in parallel by EMR, triggered by Lambda and read in Tableau. Operational within months, and nobody assembles a report by hand any more.
A search vendor who could not make the date
The incumbent quoted more time than the deadline allowed and more money than the plan. We implemented the MVP in roughly half the proposed time, and the client saved close to $100,000 in license fees.
We’ll prove it in six weeks.
Choosing an outside engineering team is slow and opaque, so we go first. Name the problem, we build the context, and within six weeks you have something you can judge — usually a working proof of concept, sometimes an architecture or a validated plan. A fixed fee, credited back if you go ahead with the full project.