Start a conversation
← Portfolio

Daedalus Ventures where the data is the value.

Daedalus is an AI and data venture studio in the City of London. I co-founded it in 2022 and I am its CTO: I lead the data engineering, data science and software architecture behind the ventures it builds, mostly B2B and financial services.

Role
Co-founder and CTO
Since
2022
Focus
FinTech, banking and lending, B2B
Based
City of London
Underwriting
Decisions waiting
Search businesses, loans⌘KThis week ▾
Applications this week186+12 on last week
Decided automatically64%the rest go to a person
Median time to decision3 h 40 mfrom bank data in
Loan book£18.4m412 live loans
30+ days in arrears1.8%+0.2 pts on September
Ready to decide14Referred9Waiting for data21Decided today38
BusinessAmountTermScoreBandStatus
Thornbury Joinery LtdJoinery and fit-out£85,00036 mo712BReady to decide
Calder Valley DentalDental practice£140,00060 mo688BAwaiting bank data
Penrose Bakery LtdFood production£32,00024 mo741AAuto-approved
Harlow & Finch LogisticsRoad haulage£210,00048 mo655CReferred
Oakmere Garden CentreRetail£60,00036 mo702BReady to decide
Brightwell Print CoPrinting£45,00024 mo618CReferred
Sefton Physio ClinicHealthcare£28,00018 mo760AAuto-approved
Ashby Lane NurseriesChildcare£52,00036 mo724AReady to decide
Kingsway Motors LtdMotor trade£175,00060 mo590DDeclined
Thornbury Joinery LtdJoinery and fit-out · trading 7 years · 14 staff · applied 09:14Band B
Asked for£85,000
Term36 months
Debt cover1.46×
Score712
Cash flow from their bank data12 months, £k
money inmoney outbalance
Why the model scored it socontribution to the score
Revenue growth over 12 months+0.42
Cash buffer, 23 days on average+0.31
Tax paid on time, 8 quarters+0.18
Directors’ credit history+0.12
Sector: construction−0.21
Two returned direct debits−0.14
Suggest: approve £85,000 over 36 monthsOver the £75k automatic limit, so it needs your sign-off.ReferApprove
An underwriting desk. The kind of screen a lending venture runs on: applications in, bank data read, a score with its reasons, and a person signing off what the rules say needs one. Brookmere and every business in it are invented.

Most proofs of concept are built to be thrown away.

The usual way, the business case is done on the side of a desk, the proof of concept and the MVP are separate jobs on tech that was never meant to scale, and revenue arrives late. In financial services the data is often the real asset, and it gets treated as an afterthought.

The business case runs beside the build, and the PoC is real code.

We write the business case while the proof of concept is built, so the numbers and the product test each other. The PoC is built to grow into the MVP and then v1, with the same team throughout. My part is the foundation it all stands on: the data, the models and the architecture. The client owns the code and the IP, and the documentation is written as we go, so they can take it in-house whenever they choose.

Platform · Data
Pipelines
Search businesses, loans⌘KRun now
Rows in today1.3mfrom 5 sources
Tests passing211 / 2121 warning, not blocking
Median run12 m 18 slast 30 runs
Models with an owner41 / 41and a description
Data cost this month£412compute and storage
Lineage1 warning upstream of the model5 sources · 41 models · 212 tests · last run 06:00
SourcesStaged and testedReady to use
Open banking feedtransactions, every 15 min
Credit bureaureports on request
Companies registerfilings, daily
Core ledgerloans and payments, streamed
Payments providercollections, hourly
stg_transactions4.8m rows · categorised
stg_bureau_reports61k rows
stg_companies23k rows
stg_ledger_entries1.2m rows
applicant_featureswhat the model reads
affordabilityone row an application
loan_bookone row a loan a day
arrears_dailyfor collections
Checks212 tests, every run
stg_transactions: uncategorised rows0.4% of rows have no category; the limit is 0.25%03:00
loan_book balances match the ledgerto the penny, 412 loans06:00
Every application has one affordability rowunique and not null06:00
No future-dated transactions4.8m rows checked06:00
Freshnessage, and what each one promises
Open banking feed11 min
Core ledger2 min
Payments provider41 min
Credit bureauon call
Companies register19 h
Runsevery 3 hours, and on demand
RunStartedTestsResult
r-418212 m 41 sToday 06:00212 / 212Passed
r-418111 m 58 sToday 03:00211 / 2121 warning
r-418012 m 10 sYesterday 23:00212 / 212Passed
The data foundation. Every source staged, tested and traced to what reads it, with freshness against what each one promises. In these ventures the data is the asset, so it is built first.
Platform · Models
sme-pd: probability of default
Search businesses, loans⌘KCompare runsPromote challenger
Champion · v3.2 · live since March0.61Gini, last 30 days
Challenger · v3.3 · shadow scoring0.64+0.03 on the same applications
Approval rate at the cut-off58%unchanged by v3.3
Population stability0.06stable; review above 0.10
Gini by month— champion · - - challenger, scored in shadow
Drift by featurestability index, this month against training
Sector mix0.14
Months trading0.07
Average balance0.05
Revenue growth0.04
Returned payments0.03
Director history0.02
Calibrationpredicted against observed, by decile
Scores this monthbars now · line at training
Before v3.3 goes live3 of 5 done
Shadow scored for 60 days4,920 applicationsdone
Fair across sectors and regionsno group moved more than 2 ptsdone
Reasons readable for every declinetop four reasons, in plain wordsdone
Sector mix drift explainedmore construction applicants since Julyopen
Credit committee sign-offbooked for 22 Octto do
A model, watched. A credit model scored against a challenger in shadow, its calibration and drift on one page, and the checks that stand between it and going live.
Brookmere · build plan
Updated every Friday
Fixed scope, fixed price for the first phase · week 9 of 20
Brookmere: from idea to v1
On planMVP in build
Week1234567891011121314151617181920
Business casewritten while the PoC is built
Business casecosts, revenue, the regulator, with real numbers
Proof of conceptreal code, not a mock-up
PoC: bank data in, a decision outsuccess criteria agreed in week 1
MVPthe PoC, grown rather than rebuilt
MVP: first real customersunderwriting desk, applicant app, loan book
v1scale and controls
v1model monitoring, reporting, collections
Data foundationfrom the first week
Sources, tests, lineagethe data is the asset, so it is built first
Handoveryours to keep
Documentation, written as it is builtand help hiring your own team
One teamthe same people, start to finish
CTPMDEDSFSFSUXCTO, product lead, data engineer, data scientist, two full-stack engineers, design and QA
This week
  • Applicant app: bank connection live on test accounts
  • Affordability model reads the staged data, not a copy
  • Underwriting desk: sign-off over the automatic limit
  • First test loans with the design partners
Decided
  • The PoC’s scoring service is kept for the MVP, not rewritten
  • Every decline carries its reasons in plain words
  • Collections in v1, not the MVP
  • Which credit bureau to keep after the pilot
What Brookmere owns
  • All the code and IP, assigned in the contract
  • The documentation, as it is written
  • The choice to take it in-house at any point
  • Its own team, hired with our help in v1
The build plan. The business case is written while the proof of concept is built, and the PoC grows into the MVP and v1 with one team. The client owns the code and the documentation throughout.
BrookmereStep 4 of 4
Approved in principle
Good news, Sarah. Here is your offer.
For Thornbury Joinery Ltd, from the 12 months of bank data you connected this morning.
£85,000over 36 months
Monthly repayment£2,819
Interest, fixed11.9% a year
Total to repay£101,480
Paid outwithin 2 working days
Why we could say yes
Your money in has grown steadily over the year.
You keep a healthy cushion of cash.
Your tax has been paid on time.
Accept and sign
Talk it through with us first
Subject to final checks and a personal guarantee from a director.
The applicant’s side. What the same decision looks like to the business that asked for it: the offer, and why, in plain words.

Screens drawn with dummy data. No real customer, client or account appears.

What it does

Fractional CTO

Technology leadership for a venture that is not ready for a full-time CTO, paired with data, product, fundraising or the build itself.

Data engineering

Sources brought in, tested and traced, so what the product and the models read can be trusted.

Data science and AI

Models built into the product, and AI-assisted development checked by people who know the domain and the data.

Architecture that grows

One codebase that carries a venture from proof of concept to MVP to v1, rather than a demo thrown away each time.

Yours to keep

The code and IP are assigned to the client, documentation is part of the build, and there is help hiring the team that takes it over.

Start small

Work can begin with a fixed-term, fixed-price piece, so both sides see how it goes before committing to more.

Data engineeringData scienceAI and machine learningSoftware architectureFull-stack engineeringDevOpsProduct design
Daedalus

Building a venture where the data is the value?

Daedalus works with founders, startups and enterprises, from a first proof of concept to a product in market. Tell us what you are building.