Architecture review and rescue
An independent read on a build that is slipping or heading somewhere expensive. The output is a ranked assessment of what is actually wrong, with effort and risk against each item.
Salesforce Technical Architecture
Salesforce architecture, and AI applied where it actually helps.
We work on Salesforce programmes that have grown complicated: integrations that need redesigning, releases that have slowed down, data models that no longer fit. Sixteen years of enterprise systems experience behind the practice, ten of them on Salesforce, across banking, insurance, telecoms, public sector, and maritime.
Selected engagements
Architecture and delivery for teams at
- CaixaBank
- DNB
- Hurtigruten
- BT
- Wefox
- Crèdit Andorrà
- Arbetsförmedlingen
- Wilhelmsen
- Unifon
- IBM
- Salesforce
Where programmes get stuck
The hard part is rarely the feature.
It is usually what has accumulated underneath. Integrations built for a pilot and never revisited. Automation added over automation, each change reasonable on its own. A data model that suited the first market and now has to serve four. None of it was a mistake at the time.
It surfaces later as a delivery problem: releases slip, defects reopen, the team is busy and shipping less than it should. The cause is normally structural, and that is the useful part, because structural problems respond to being addressed deliberately.
Find the constraint, remove it, and leave the team able to keep going without us.
What we do
Where we are usually brought in.
Integration design
Point to point sprawl, batch jobs whose ownership has moved on, and sync patterns that will not survive the next volume step. Rebuilt around explicit contracts, idempotency, replay, and failure you can see.
Duplicate records and customer identity
The same customer held several times over, matching rules never tuned past the defaults, and reporting that counts one person twice. Measured, remediated, and prevented at the point of entry.
Data Cloud and customer data
Ingestion, identity resolution, and activation designed for production rather than for a demo. Consumption cost and GDPR treated as design constraints, because retrofitting either is expensive.
DevOps and release automation
Environment strategy, branching, and CI/CD that a multi vendor team can actually follow. Gearset, Copado, and AutoRabbit, with AI used to shorten review and catch regressions earlier.
AI in delivery, and Agentforce where it fits
Two different things. Agentforce for customer facing automation when the use case suits it. General purpose models inside the delivery pipeline, which is where most of the measurable gain sits today.
AI
Useful AI is mostly unglamorous.
Specification review, test generation, metadata diffing, release notes, pull request analysis. Work that is repetitive, well bounded, and cheap to verify. That is where models are dependable now, and it compounds quietly.
Track record
By the numbers
16
Years building enterprise systems
10
Years on the Salesforce platform
9
Countries delivered in
Get in touch
Tell us what has stopped moving.
A short description of where the programme is stuck is enough to start. If we are not the right fit for it, we will say so.