Assessment to roadmap

A proposal the client could click

Predictive models were running and nobody could see them. I shipped the assessment as a working prototype on synthetic data rather than a document, then survived the meeting that changed the cloud, the tool and the pilot customer.

Clickable prototype instead of a PDF proposal
8 screensClickable prototype instead of a PDF proposal
Pilot, then multi-tenant hub, then AI priced separately
3 releasesPilot, then multi-tenant hub, then AI priced separately
To first live attribution for the pilot customer
~4 weeksTo first live attribution for the pilot customer

The client

A marketing consultancy in the United States serving member-owned financial institutions. Small team, no in-house data platform, dependent on a specialist prediction vendor and understandably nervous about depending on another one.

The engagement

A paid assessment producing a three-release roadmap, with the third release — the AI one — deliberately priced as a separate later decision.

A working demo of this build exists.

It is not public yet — ask for access, and where the NDA allows I will send a link or walk you through it on a call.

Request demo access

The problem

The client had predictive attribution running on a specialist platform and no way for their own customers to see any of it. The obvious answer — pick a business intelligence tool and build dashboards — would have locked a small organisation into whichever vendor happened to be in the room that quarter, and would not have survived their customers' wildly different data formats.

What I did

I proposed an orchestration layer the customer owns, with visualisation as the first tangible piece of it, and one founding principle: the client is never asked to change their data format. Rather than describe that in a document I built it — a password-gated proposal site with a clickable prototype on synthetic data, several mock tenants, an agency-versus-customer view switch and an export path — so the conversation was about a thing rather than an intention. The release plan deliberately fenced AI capabilities into a third release priced separately, so nobody bought a language model before they had a working number. When a later meeting flipped the cloud provider, dropped the committed reporting tool and replaced the pilot customer, the artefact was updated in days because the layers had been separated on purpose.

What was built

A vendor-agnostic orchestration layer with the visualisation tier as its first tangible piece: schema-agnostic ingestion so the customer never changes their data format, a non-personal-data key boundary, multi-tenant delivery, and a clickable eight-screen prototype on synthetic data standing in for the proposal document.

On the table at the end

  • Password-gated proposal site with a clickable prototype
  • Five-layer reference architecture
  • Three-release roadmap with effort ranges and a separately priced AI release
  • Non-personal-data key map for the upstream data contract

What it changed

Replaced a vendor-locked reporting conversation with a layer the institution owns, and fenced AI spend into a release nobody had to fund before seeing a working number. When the client reversed three foundational choices in one meeting, the proposal was updated in days instead of rewritten.

How it ran

  1. 01

    Reference architecture first

    A layered, vendor-agnostic model with a clear statement of which layer we were selling and which ones the client already owned.

  2. 02

    Draw the personal-data boundary early

    A key map that keeps personal data on the client's side of the line, agreed before any data contract discussion with the upstream vendor.

  3. 03

    Prototype as the proposal

    Eight screens, synthetic tenants, tenant switcher and role switch — on a real URL, sent as a link rather than an attachment.

  4. 04

    Re-scope after the pivot

    Cloud, reporting tool and pilot customer all changed in one meeting; attribution became the spine and the secondary models folded into it as tabs.

  5. 05

    An estimate that separates the bet

    Pilot and multi-tenant hub costed as core scope; AI and activation quoted as a distinct, later decision with its own range.

Something similar on your plate?

Thirty minutes, no deck. I will tell you whether it is worth doing at all.