What modern data architecture consulting should deliver in the first four weeks

Data architecture consulting is the assessment and design work that decides where an organization’s data lives, how it moves, who can reach it and which controls apply, before anyone builds a pipeline. The build is expensive and slow to change. The design is cheap and fast to change. The whole point of the consulting is to spend the cheap weeks well so the expensive months go in the right direction.

This guide sets out what a modern engagement should hand over in its first four weeks, because four weeks is long enough to make a decision and short enough that nobody has forgotten why they asked. It is written from the delivery side. Dedicatted designs and builds data platforms on AWS, Azure and Google Cloud, so the timeline, the deliverables and the warning signs below are the ones we use and the ones we would want a customer to hold us to.

Why four weeks

The classic data architecture assessment ran for a quarter and ended in a hundred-page document that described the current state in great detail and the target state in a diagram. By the time it landed, two of the source systems had changed, the sponsor had moved on, and the engineering team had started building anyway. The document was accurate and useless.

The alternative is to treat the assessment as a decision with a deadline. In four weeks a competent team can map the sources, pipelines, stores and consumers, find the bottlenecks and the risks, draw the target architecture, decide what happens to each workload and put a cost against each phase. What it cannot do in four weeks is build anything, and that is the discipline. Anything that needs more than four weeks of analysis before a decision can be made is usually a sign that the question was wrong.

What modern data architecture means

Modern is an overused word, so here is the concrete version. A modern data architecture in 2026 has five properties, and a consulting engagement that does not design for all five is designing the platform you already have with newer logos.

• Storage separated from compute. Data lives in object storage such as Amazon S3 in an open table format, and query engines such as Redshift, Snowflake, Databricks or BigQuery run on top of it. That is what people mean by a lakehouse. Cost scales with what you run, not with what you keep, and no single vendor owns the data.
• Streaming and change data capture instead of nightly batches. Changes flow from source systems through Kafka or Kinesis, with tools such as Debezium reading the database log, so the platform is minutes behind the business rather than a day. This is also what makes a migration run in parallel with production instead of in a weekend outage.
• A catalog with lineage. Every dataset has an owner, a definition and a traceable origin, in a catalog such as AWS Glue or Unity Catalog. Without it, governance is a policy document nobody can enforce.
• Governance designed in, as code. Role and attribute based access, retention, data quality checks and audit trails defined in the same repositories as the platform, so a HIPAA, PCI, GDPR or SOC 2 review is a document review rather than a project.
• Infrastructure as code. The platform itself in Terraform, reviewable in Git, reproducible in another account or region. A platform that was clicked together in a console is a platform nobody can safely change.

The property that separates a modern design from a fashionable one is the first. If the architecture locks the data inside one engine’s proprietary storage, the rest is decoration.

The first four weeks, week by week

Here is the shape of an engagement that produces a decision. The first two days can be bought on their own, and often should be, because they tell both sides whether the next three and a half weeks are worth it.

The per-workload decision deserves its own explanation because it is where most of the money is saved or wasted. Lift and shift moves data and applications as they are. It is the fastest route and often the most expensive to run, right for a deadline-driven data centre exit or a legacy system that rarely changes. Re-platform makes targeted changes on the way, swapping self-managed databases and ETL for managed equivalents. It is slower than a lift and cheaper every month after. Re-architect is the full redesign, the option where a lakehouse pays for itself, right for analytics platforms, AI workloads and systems already at their limits. Most estates end up a mix. An engagement that recommends the same treatment for everything has not looked.

What should be in the deliverable

Whatever the provider calls the document at the end of week four, it should contain the following, and you should be able to act on each item without the provider in the room.

• The current-state map. Sources, pipelines, stores and consumers, with volumes, freshness and the monthly cost of each, and the risks ranked by what they would cost if they happened.
• The target architecture. Drawn, with the storage, the engines, the streaming path, the catalog and the governance zones named, and the reasons the alternatives were rejected.
• A decision per workload. Lift and shift, re-platform or re-architect, with the reason in one sentence each.
• A phased, costed roadmap. Each phase with its scope, its dependencies, its estimated cost and what the business gets when it lands. You should be able to fund phase one without committing to phase four.
• The migration approach. How each phase moves without downtime, which usually means change data capture, parallel runs, reconciliation by row counts and checksums, and a rehearsed cutover with a known way back.
• The governance and compliance design. Which regulated zones exist, who can reach what, how retention and lineage are enforced, and how the next audit becomes a document review.
• A cost model. What the platform costs to run at today’s volume and at the volume the roadmap plans for, so the finance conversation happens before the build rather than after the first bill.

The warning signs

Four things tell you, usually by the end of week two, that an engagement is going to produce a deck rather than a decision.

• The target architecture appeared before the current state was mapped. A provider that already knows the answer is selling a platform, not consulting on yours.
• Every workload gets the same treatment. Everything to the cloud as it is, or everything rebuilt. Either way, nobody made a decision.
• There is no cost against any phase. A roadmap without numbers is a wish list, and the engineering team will find out the real cost in production.
• Governance is a later phase. Access control, retention and lineage designed in afterwards are a second project. Designed in from the start they are a week of work.

What it costs

The two-day snapshot and the four-week jump start are fixed-scope offers, so a reputable provider will quote them in the first call. What varies is the build that follows, and the right way to price that is per phase, estimated from the roadmap, so you see each number before you commit to it. The pattern to avoid is the one where the assessment is cheap or free and the build is open-ended. The assessment is the product. Pay for it, own the document, and then decide who builds.

How Dedicatted does it

Dedicatted runs data architecture consulting as two accelerators and then the design and build they call for. The two-day current-state snapshot ends with a C1 view of your systems and a written report your team can act on. The four-week architecture modernization jump start turns the snapshot into a target architecture, a lift, re-platform or re-architect decision per workload, and a costed roadmap. Both are fixed scope and quoted in the first call. The build that follows is designed as code in Terraform, reviewable in Git, with change data capture pipelines where downtime is not an option, parallel runs and reconciliation before anything is switched over, and your team trained to run the platform.

We work across AWS, Azure and Google Cloud, with the deepest bench on AWS as a Premier Tier Services Partner, and we design for Databricks, Snowflake, Redshift or BigQuery where each fits, rather than one answer for every customer. Our practices are SOC 2 Type 2 and ISO 27001 audited, which matters when the architecture carries regulated zones. The accelerators, the six consulting services and the stack are on the data architecture services page. The move itself is described under data migration consulting, and the warehouse or lakehouse that usually sits at the centre under data warehouse consulting.

Timelines, deliverables and the accelerator scopes are Dedicatted’s own, current at the time of writing.