AFDSOLUTIONS

Technology

Data — Azure, GCP, AWS

Decision-ready data, on the cloud you already run.

Most of the data a CRM needs already sits on a hyperscaler — half-modelled, owned by a different team, and trusted by nobody in particular. We work where it lives rather than moving it for the sake of it, and build the ingestion, modelling and governance that makes it fit to decide on instead of merely available.

Talk to usSee all services
  • Azure, Google Cloud, AWS
  • Warehouse and lakehouse patterns
  • Governed, documented, tested
  • Designed for the run cost

The platform is rarely the constraint

Synapse, BigQuery and Redshift will all hold your data perfectly well. What decides whether anyone can act on it is duller than the platform choice: whether a load can be safely re-run, whether the model answers the questions people actually ask, whether a failed pipeline is noticed the same day, and whether two teams agree what a customer is.

We are deliberately not evangelists for one cloud. We work on the one you already pay for, and we design pipelines with their running cost visible while the design can still change — the same discipline we apply to Salesforce credits.

Scope

What we do across the hyperscalers

  1. Ingestion and landing

    Getting data in reliably — batch, change-data-capture or streaming — with the failure and replay cases designed for rather than discovered.

    • Source selection and load pattern
    • Idempotent, re-runnable loads
    • Schema drift handling
    • Monitoring, alerting and backfill

  2. Modelling and transformation

    A layered model — raw, conformed, serving — with transformations that are tested and readable by someone who did not write them.

    • Warehouse and lakehouse layering
    • Incremental transformation logic
    • Tests on the assumptions, not just the syntax
    • Version-controlled, reviewable pipelines

  3. Identity and entity resolution

    Matching customers, accounts and products across systems, with match quality that is measured rather than assumed.

    • Match rules and survivorship
    • Cross-system key management
    • False match and miss measurement
    • Iterative tuning with evidence

  4. Governance and access

    Catalogue, lineage and permissions set up so people can find what exists and only see what they should.

    • Catalogue and lineage
    • PII classification and handling
    • Least-privilege access design
    • Retention and residency constraints

  5. Feeding CRM and agents

    Moving the governed result into Salesforce — and into agent grounding — without turning the warehouse into a second CRM.

    • What belongs in CRM and what does not
    • Sync patterns, volumes and latency
    • Grounding and retrieval sources
    • Sharing and exposure boundaries

  6. Cost and performance

    Compute and storage are consumption decisions. We model them alongside the design instead of after the first invoice.

    • Partitioning, clustering and pruning
    • Storage tiering and lifecycle
    • Cost forecast by pipeline
    • Tracking actuals against forecast

Connections

Related work

Data Flow Automation

The service engagement that builds and hands over these pipelines.

Learn more

CRM — Salesforce Platform

Where the governed data lands, and the processes that consume it.

Learn more

Grounding for agents

Using resolved, governed data as trustworthy agent context.

Learn more

Definitions and semantics

Metrics that reconcile because the meaning was settled first.

Learn more

What good looks like

What a well-built data platform gives you

Pipelines you can re-run

Loads that are idempotent and monitored, so a bad day is recoverable rather than forensic.

Entities you can rely on

Resolution with measured quality, not an assumed match rate carried forward from a demo.

A model that answers

Structured around the questions the business asks, so analysts stop rebuilding it themselves.

Governance that survives audit

Catalogue, lineage and access decisions written down rather than reconstructed on request.

Cost you can forecast

Compute and storage modelled per pipeline while the design is still changeable.

A team that owns it

Pipelines under source control with runbooks, so the next change does not need us.

Related from AFD

Is your data actually fit to decide on?

Tell us where it lives and what you need it to answer. We will tell you what is missing between those two things.

Start the conversation