SaaS prototype
Showing the operator's console, including where the prompts live
A planning platform demo that switches from the customer's view to the vendor's — tenant health, system status, and a sandbox where the AI's prompts are edited. That last screen is the one that sells.
- Customer view and operator view in a single demonstration
- One switchCustomer view and operator view in a single demonstration
- Where the assistant's instructions are edited and tested, shown on screen
- Prompt sandboxWhere the assistant's instructions are edited and tested, shown on screen
- Forecasts stacked rather than swapped, which is how planners actually work
- Scenarios layeredForecasts stacked rather than swapped, which is how planners actually work
The client
A financial planning and analysis software vendor selling a multi-tenant product with an AI assistant, to buyers who increasingly ask not what the assistant does but who controls what it says.
The engagement
A prototype with a client-and-administrator switch, so one demonstration serves both the end-user story and the operator story.
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.
The problem
AI features in financial software now fail procurement rather than demonstration. The capability convinces on screen and then stalls on a governance question nobody can answer live: who writes the assistant's instructions, how are changes tested, what happens to one tenant when another's usage spikes. Answering those in a follow-up document loses the momentum the demo created.
What I did
I put the governance on screen. The administrative view is not an afterthought tab — it carries tenant monitoring, system status and a prompt sandbox where the assistant's instructions can be edited and tried, which turns 'how is this controlled' from a written answer into a thing the buyer watched. On the customer side, scenarios are layered rather than swapped, because planners compare futures side by side rather than one at a time, and a demo that gets that wrong signals you have not sat with the users. Metric cards carry their own trend so a number never appears without its direction. The switch between the two audiences is a single control, so a fifteen-minute meeting can cover both stories without a reload or an apology.
What was built
A customer-facing planning surface with layered scenario forecasting, metric cards carrying their own trend, a revenue chart, a scenario panel and an assistant, paired with an administrative view covering tenant monitoring, system status and a prompt sandbox — with a single switch between the two.
On the table at the end
- Client-facing planning and scenario surface with assistant
- Administrative view: tenant monitoring, system status, prompt sandbox
- One-switch transition between the two audiences
What it changed
Answered the governance question in the demo rather than in a follow-up document: the buyer sees the tenant monitoring, the system status and the sandbox where the assistant's instructions are edited and tested — which is what makes an AI feature procurable in finance.
How it ran
- 01
Build the operator's side properly
Tenant monitoring and system status as first-class screens, not a settings page.
- 02
Show the prompt sandbox
Where the assistant's instructions are edited and tested — the answer to the governance question, demonstrated rather than described.
- 03
Layer the scenarios
Forecasts compared side by side, because that is how planning actually works.
- 04
Never show a number alone
Every metric card carries its own trend, so a figure arrives with its direction.
- 05
One control between audiences
A single switch from customer to operator, so both stories fit inside one meeting.
Other work
All case studies →- Insurance
One product, two audiences, one prototype
An annuity platform has an administrator who lives in it all day and a customer who opens it twice a year. I built both, because the demo that shows only one always gets the same question.
- Telecommunications
Every clause gets a human verdict
Contract review is the most demoed AI use case in the enterprise and the least trusted. I built the version where the model proposes a redline and a lawyer accepts, edits or rejects each one.
- Asset management
Three layers for two audiences on a quantitative platform
A portfolio platform has to convince a business buyer and a quant, and they cannot be shown the same thing. I packaged the pitch, the library and the notebooks as three separate entry points.
Something similar on your plate?
Thirty minutes, no deck. I will tell you whether it is worth doing at all.