Services

From where you are to where you're going

Every company starts from a different place. Pick the description closest to yours to see what getting to your north star would involve: the work, the timeline, the costs, and what tends to go wrong. Most companies sit across two stages, and that's normal.

Where are you today?

Stage 1 · Running on spreadsheets

Data Foundations Sprint

A small, low-cost cloud data platform that replaces the manual reporting cycle with automated pipelines and one agreed set of KPIs.

Usually 10–100 staff with no data team. Reporting sits with a finance or operations lead. The founder, MD or finance director signs off.

Typical duration4–6 weeks

Discuss this offering

Where you are

Reporting is manual exports, Excel and copy-paste. Month-end takes days and nobody agrees on the numbers.

Signs you're here

  • Key reports are built by hand every week or month
  • The same KPI gives different answers depending on who you ask
  • One person is the only one who knows how the report works

What usually gets in the way

  • The real data is in personal spreadsheets, each with its own layout and its own manual fixes
  • Some systems have no API, or the vendor charges extra for data access
  • Departments define the same KPI differently, and agreeing one definition is a political decision as much as a technical one
  • The one person who knows the reports is also the busiest person in the business
Your north star

One trusted set of numbers, refreshed automatically, that leadership uses to run the business.

What I'll do

  • Source audit: where the data really comes from and who owns it
  • Right-sized cloud warehouse on the platform that suits your team
  • Automated loads from your core systems and spreadsheets
  • Agreed definitions for your 5–10 most important KPIs
  • A first dashboard built on the governed model, not on the raw exports

You get

  • Working pipelines and warehouse
  • KPI definitions document
  • Executive dashboard
  • Runbook and handover

What you'll need to provide

  • A named owner with 2–4 hours a week
  • Admin or export access to each source system
  • KPI definitions decided within an agreed timeframe
  • Copies of the reports you produce today

Not included

  • Cleaning years of historical data by hand. Anything that can't be fixed automatically gets flagged and scoped separately
  • Replacing your CRM, ERP or finance system
  • Software licences (bought in your name, so you own them)

Risks, and how I manage them

  • The dashboard gets built and nobody uses it. Replace one report people already depend on, designed with the people who chair the meeting where it is used.
  • People keep editing spreadsheets by hand and break the loads. Locked input templates or simple forms, with validation that flags bad rows instead of failing silently.
  • Nobody in-house can maintain the platform after handover. Managed services and a deliberately small toolset, a written runbook, and optional light-touch support.
Typical duration
4–6 weeks once access is in place. Getting logins and exports sorted often adds 1–2 weeks at the start
How we measure success
Hours of manual reporting removed each week · Report refresh moved from manual to scheduled · One agreed figure per KPI
Running costs after go-live
Indicatively £50–£500 a month for cloud and BI licences at this size. It mostly depends on how many people need report access.
What changes for your team
Someone needs to own the KPI definitions and handle new report requests, about half a day a week. Without an owner, the spreadsheets drift back in.
When this isn't the right fit
Businesses that run almost entirely on one system with good built-in reporting. Its native reports are cheaper than a platform.

Done before: Centralised reporting that cut one regular data retrieval from 5 hours to 10 minutes.

Stage 2 · Too many disconnected tools

Unified Data Platform

Bring every system into one modelled, governed cloud platform with conformed dimensions and a semantic layer that teams can self-serve from.

Usually 50–500 staff, with one to three analysts and rarely a data engineer. A head of finance, operations or BI usually sponsors it; IT and data protection have a say.

Typical duration8–12 weeks

Discuss this offering

Where you are

You run on several SaaS apps (CRM, finance, operations), each with its own reports, and none of them talk to each other.

Signs you're here

  • Analysts spend their week stitching exports together
  • Customer, product or property data is duplicated and inconsistent across systems
  • Every new question becomes a one-off project

What usually gets in the way

  • No shared customer or product ID across systems, so records have to be matched, and some matches will always need a human decision
  • SaaS API rate limits, paid API tiers and connector pricing that rises with data volume
  • Each department owns its own definitions and doesn't want to give them up
  • Personal data moving into a new platform needs a lawful basis and often a DPIA under UK GDPR before anything is loaded
Your north star

One integrated view across your main systems, so routine questions are answered from a shared model instead of hand-stitched exports.

What I'll do

  • Managed or custom ingestion from SaaS APIs and databases
  • Medallion layers (raw, cleaned, business-ready) on a lakehouse or warehouse
  • Kimball star schemas with conformed customer, product and date dimensions
  • A semantic model so BI tools share the same measure definitions
  • Data quality tests and freshness alerts on every pipeline

You get

  • Integrated data platform
  • Dimensional model and semantic layer
  • Data catalogue and definitions
  • Self-serve reporting starter pack

What you'll need to provide

  • An owner for each source system who can grant access and answer questions
  • A business owner for each domain to sign off definitions
  • Input from your data protection lead early on
  • A decision-maker who can settle definition disputes

Not included

  • Fixing data quality inside the source systems. I find and report the issues; the process owners fix them at source
  • Buying or implementing master data management software
  • Connector and platform licences

Risks, and how I manage them

  • Connector costs grow faster than expected. Volumes estimated before choosing tools, with managed connectors where they're cheap and custom ones where they aren't.
  • "Just one more source" scope creep. Sources ranked by business value and delivered in phases, each with its own sign-off.
  • The central model becomes a bottleneck. A clear change process and documented ownership for each domain.
Typical duration
8–12 weeks for 3–5 sources once access is in place. Each extra source adds roughly 1–2 weeks, depending on its API
How we measure success
Number of sources unified · Report load and refresh times · Share of questions answered without an engineer
Running costs after go-live
Indicatively £500–£3,000 a month. Connector fees are often the biggest line, so they get modelled before anything is chosen.
What changes for your team
Analysts move from stitching exports to analysis, which needs training on the shared model. Someone has to own the platform: an internal hire or a fractional engineer.
When this isn't the right fit
Companies where one ERP or platform already holds most of the data. Start with its native analytics and add sources only when there's a clear need.

Done before: Consolidated three operational systems into one source of truth, saving 8 hours a week of manual processing and cutting report load time from 2 hours to 30 minutes.

Stage 3 · Tied to a legacy system

Reconciled Legacy Modernisation

Migrate safely: read the legacy system strictly read-only, rebuild it as a governed cloud warehouse, and prove after every load that the numbers match the source.

Established, often regulated businesses (insurance, finance, healthcare, manufacturing, public sector). The CIO, CFO or chief data officer sponsors it, with procurement, security and sometimes the regulator involved.

Typical duration3–6 months for the first domain

Discuss this offering

Where you are

Your system of record is on-prem or on a mainframe. It works, but it can't be switched off, nobody fully understands it, and reporting from it is slow and contested.

Signs you're here

  • Thousands of cryptic tables and few people who can explain them
  • Reports disagree, and finance, actuarial or operations argue over whose number is right
  • Earlier migration attempts stalled or made the numbers worse

What usually gets in the way

  • Security approval for even read-only access to a production system routinely takes 3–6 weeks
  • Network connectivity between the on-prem system and the cloud (VPN, private link, gateway) needs the infrastructure team's time
  • Business rules live in code and in the heads of long-serving staff, some of whom are close to retiring
  • Extraction has to fit around batch windows so it doesn't slow down production
  • Vendor licences or data residency rules can restrict where the data may go
Your north star

A governed cloud warehouse beside the legacy system that is checked against it every day, with any remaining differences known, explained and signed off.

What I'll do

  • Read-only extraction with change data capture, so the source is never put at risk
  • Domain modelling of the core business (claims, policies, orders, ledgers)
  • Two-sided reconciliation of business measures after every load
  • Load audit, lineage and point-in-time recovery
  • Phased cut-over plan, so the legacy system can retire one report at a time

You get

  • Cloud warehouse running in parallel with the legacy system
  • Reconciliation suite and daily evidence
  • Decoded business-rule documentation
  • Cut-over roadmap

What you'll need to provide

  • Regular time from business experts and the legacy IT team
  • A read-only service account, plus network and security approvals
  • Infrastructure team support for connectivity
  • A senior sponsor who decides which number is "right" when the departments disagree

Not included

  • Replacing or re-platforming the operational system itself
  • Rewriting transactional applications or user screens
  • Cloud and licence costs, including the legacy system's own running costs during the parallel run

Risks, and how I manage them

  • The new numbers don't match and trust collapses early. Reconciliation from the first load, with every variance explained and signed off, never hidden.
  • Running two systems in parallel costs more for a while. The parallel-run period is costed up front, with a retirement date for each legacy report.
  • The programme stalls at 80% done. Phased by business domain, so each phase delivers usable value on its own.
  • A key expert leaves mid-project. Business rules captured early as documented, tested code.
Typical duration
3–6 months for the first business domain, after 3–6 weeks of access and security approvals. A full estate usually takes 12 months or more, phased
How we measure success
Measures reconciled to source after every load · Share of rows moved by change capture rather than full reload · Reports retired from the legacy system
Running costs after go-live
Indicatively £2,000–£15,000 a month, depending on data volume and the compute model, on top of legacy costs until reports are retired. Incremental loads keep it towards the lower end.
What changes for your team
The legacy team's knowledge becomes the most valuable input, so they need to be involved, not bypassed. New skills (cloud, SQL/Spark, infrastructure as code) are built by pairing, not left in a handover document.
When this isn't the right fit
Businesses replacing the legacy system with a packaged product within about 12 months. The priority then is data migration into that product, not a parallel warehouse.

Done before: For an insurer on a decades-old IBM i system: ~100M rows across ~160 tables refreshed daily in ~90 minutes, with ~70 business measures reconciled after every load.

Stage 4 · In the cloud, but costly or fragile

Platform Health & Cost Optimisation

An audit-then-fix engagement: find where money and trust are leaking, then move the platform to incremental loads, automated testing and infrastructure as code.

Companies that moved to the cloud 1–5 years ago, often as a lift-and-shift or through contractors who have since left. The CTO or head of data is usually under pressure from finance about the bill.

Typical duration6–8 weeks

Discuss this offering

Where you are

You've already moved to the cloud, but the bill keeps rising, pipelines break quietly, and people have stopped trusting the dashboards.

Signs you're here

  • Cloud or capacity costs are rising faster than usage
  • Failures are found by business users, not by monitoring
  • Changes are made by hand in the portal, with no code or CI/CD

What usually gets in the way

  • Without cost tagging, nobody can say which team or pipeline is driving the spend
  • Reserved capacity or committed spend already signed limits short-term savings
  • Pipelines without tests are risky to change, because nobody knows which reports depend on them
  • Production is changed by hand in the portal, so what's running doesn't match any documentation
Your north star

A platform that costs what it should, fails loudly, and can be rebuilt from code.

What I'll do

  • Cost and architecture audit with a prioritised savings plan
  • Full reloads replaced with incremental, scheduled processing
  • Data quality and reconciliation checks that catch silent failures
  • Infrastructure as code (Terraform) and CI/CD with secret scanning
  • Access control, sensitivity labels and governance tidied up

You get

  • Audit report and savings roadmap
  • Optimised pipelines
  • Terraform and CI/CD pipelines
  • Monitoring and alerting

What you'll need to provide

  • Read access to billing and cost management
  • Admin access to the platform
  • Agreed change windows
  • Someone with authority to approve changes to production

Not included

  • Renegotiating vendor contracts or commitments. I provide the evidence to take into the negotiation
  • Rewriting every pipeline. Work is prioritised by cost and risk
  • Moving to a different cloud. That is a re-platforming programme and is scoped separately

Risks, and how I manage them

  • An optimisation quietly breaks a downstream report. Outputs reconciled before and after every change, released in small batches with a rollback plan.
  • Savings get eaten by new usage within months. Budgets, alerts and a named owner for each cost area.
  • Bringing existing resources under Terraform causes drift or downtime. Resources imported gradually, starting with the least critical.
Typical duration
A 2-week audit, then 4–6 weeks of fixes, prioritised by saving and risk
How we measure success
Monthly platform cost · Pipeline run time · Incidents found by monitoring rather than by users
Running costs after go-live
The aim is to lower it. Quick wins (idle compute, needless full reloads, oversized capacity) often save 20–50%. The ~80% saving I achieved came from one specific always-on setup, so it isn't a promise. The audit gives you a realistic figure before you commit to the fix phase.
What changes for your team
Engineers move to changes via pull request and CI/CD. That feels slower for the first few weeks and is much faster after that. Portal "quick fixes" stop.
When this isn't the right fit
Platforms that are fundamentally the wrong product for the workload. Tuning won't fix that; it needs a planned re-platform.

Done before: Cut platform capacity cost by ~80% and pipeline processing time by ~80% on two separate platforms.

Stage 5 · Ready for AI and real time

AI-Ready Data Products

Turn your platform into owned, documented data products, add streaming where it pays for itself, and build the governed pipelines that AI and LLM use cases need.

Organisations with an established platform and a data team of five or more. Sponsored by the chief data officer, CTO or a product leader, with legal, compliance and security involved from the start.

Typical duration6–8 weeks per use case

Discuss this offering

Where you are

Your platform is solid. Now the business wants AI, real-time insight and data shared safely across teams and partners.

Signs you're here

  • AI pilots stall because the data isn't documented, governed or fresh enough
  • Teams want near-real-time operational data, not overnight batches
  • There's no clear ownership of data products

What usually gets in the way

  • AI is wanted, but there's no specific use case with a measurable return
  • The most valuable content (documents, emails, call notes) isn't in the platform at all
  • AI tools must respect the same row-level permissions as reports, and most pilots ignore this
  • Regulation (UK GDPR, the EU AI Act where it applies, sector regulators) needs an audit trail
  • Data product ownership is an organisational change, and domain teams may not want the responsibility
Your north star

Documented data products with named owners, feeding the one or two AI and real-time use cases that have a measurable return.

What I'll do

  • Data product design: owners, contracts, SLAs and documentation
  • Streaming or micro-batch pipelines for time-sensitive use cases
  • Feature and retrieval (RAG) pipelines from governed, permission-aware data
  • Lineage, cataloguing and access policies built for AI use
  • An operating model for how teams publish and consume data

You get

  • Data product catalogue
  • Streaming and AI data pipelines
  • Governance and access framework
  • Operating model and playbooks

What you'll need to provide

  • A business owner for each use case
  • Named data owners for each data product
  • Early involvement from security, legal and compliance
  • Agreed criteria for judging whether the AI output is good enough

Not included

  • Training custom models or building end-user AI applications. I build the data foundations and pipelines and work alongside your ML or app team or partner
  • Model, API and streaming platform costs

Risks, and how I manage them

  • Pilots that never reach production. One or two use cases with a business owner, a measurable target and a decision date.
  • Sensitive data leaking through AI answers. Permission-aware retrieval, redaction of personal data, and logging of every query.
  • Streaming that costs more than it's worth. Micro-batch by default; true streaming only where the latency pays for itself.
  • Unpredictable AI usage costs. Usage monitoring, caching and a budget for each use case.
Typical duration
6–8 weeks for each use case or data product, with a go/no-go decision at the end of each
How we measure success
Data products with named owners and SLAs · Data freshness for priority use cases · AI use cases moved from pilot to production
Running costs after go-live
Highly variable. AI inference and streaming costs depend on volume and latency, so each use case gets its own cost estimate at design stage, before anything is built.
What changes for your team
Domain teams take on ownership of their data products, with SLAs. That's an operating-model change, not just a technical one, and it needs leadership backing.
When this isn't the right fit
Organisations whose core numbers still aren't trusted. AI amplifies bad data, so fix stages 1–4 first.

Not sure where you fit?

Data Platform Assessment

Not sure which stage you're at? I'll review your systems, pipelines, costs and reporting, then give you a maturity score, a north-star statement and a costed, phased roadmap you can act on with or without me. If the honest answer is "don't build a platform yet", that's what the report will say.

  • Stakeholder interviews
  • Architecture and cost review
  • Maturity scorecard
  • Phased roadmap with quick wins

2 weeks of work, usually over 3–4 calendar weeks to fit around interviews

Book an assessment

What every engagement involves

Whatever the stage, these shape cost and timeline more than the technology does.

Access is usually the first delay

Accounts, security reviews and network approvals often take longer than the first build. I send the access checklist before kick-off so the clock starts early.

Your people's time is the scarcest input

Definitions, sign-offs and business rules can only come from your team. Most delays I've seen were waits for decisions, not technical problems.

There are costs beyond my fee

Cloud, connector and BI licences are billed to you directly, so you own them. Every proposal includes an estimate of running costs, not just build costs.

Compliance comes first, not last

Personal and regulated data needs a lawful basis, retention rules and access controls designed in from the start: UK GDPR, sector rules and your own policies.

Adoption is part of the job

A platform nobody uses has failed, however good the engineering. Training, documentation and retiring the old reports are in scope.

Some data will be messy

Every source has gaps and contradictions. I report them with evidence and a proposed fix; the fix belongs to the process that creates the data.

How we can work together

Fixed-scope sprint

A defined outcome, timeline and deliverables. Best for assessments and stage 1–2 work.

Phased programme

Delivered domain by domain, with a go/no-go review between phases. Best for migrations and modernisation.

Fractional data engineer

A set number of days each month to run, improve and extend your platform once it's live.

Platform-agnostic: Microsoft Azure & Fabric, AWS, Google Cloud, Snowflake, Databricks and dbt. I recommend whichever one fits your team, your contracts and where your data already lives.

Work with me

Need data you can actually trust?