SELF-INITIATED PRODUCT CONCEPT

RenewalPipeline

From fragmented policy datato a renewal workflow brokers can act on.

I designed and built a B2B workflow for small commercial insurance brokerages that imports the spreadsheet they already maintain, extracts policy data from PDFs, surfaces premium at risk and prepares renewal communications for human review.

The product gives AI narrow, inspectable tasks while the application owns the workflow and the broker retains final authority.

Explore the interactive demo

Role: Product design and full-stack development
Scope: Workflow strategy, interaction design, AI integration and working build
Product: B2B vertical SaaS
Status: Self-initiated product concept

RenewalPipeline dashboard showing $590,638 premium at risk, overdue and upcoming renewal windows, a prioritised To do list, renewal rate and pipeline status, using sample data. — One operational view of premium at risk, overdue policies and the next action. The public demo uses sample data so the workflow can be inspected without exposing a real brokerage book.

The renewal book was visible. The risk was not.

This concept starts from an operating pattern familiar to small commercial insurance teams: policy information spread across management-system exports, working spreadsheets, PDF schedules and individual inboxes.

The data exists, but the work still depends on somebody reconciling formats, checking dates, remembering the last contact and deciding what needs attention next.

That makes the commercial risk difficult to see. An overdue policy, an incomplete record and a valuable account approaching expiry can all sit in the same spreadsheet without creating the same sense of urgency.

The business consequence

When renewal work remains recall-driven, managers get a late view of premium exposure, follow-up varies between brokers and client communication becomes difficult to audit.

The product hypothesis was that a shared operational view could make three things visible sooner:

Which premium is currently at risk
Which renewals need action now
What has already happened on each policy

These are intended outcomes, not measured customer results.

The adoption constraint

Replacing the brokerage’s existing system would create a second problem: migration effort, retraining and resistance from people already under renewal pressure.

The intervention therefore begins with the files teams already use.

RenewalPipeline imports the existing book, structures what it can, flags what still needs review and adds a controlled workflow around it.

Import policies screen with Upload, Map columns, Review and Done steps, accepting a spreadsheet or PDF policy schedule for review before anything is saved. — The product accepts the brokerage’s existing spreadsheet or a PDF policy schedule. Structured columns are mapped into the book of business, while extracted fields are flagged for review before anything is saved.

Start with the existing input

Decision

Support the spreadsheet or management-system export already used by the brokerage instead of requiring a clean migration before the product becomes useful.

Why it mattered

The product needed to create value before asking the team to change its operating habits. A forced replacement workflow would increase setup time and make adoption depend on perfect data.

Intended outcome

A shorter route from the current book of business to a usable renewal view.

Prioritise work, not records

Decision

Make premium at risk, overdue windows and the next action the centre of the dashboard.

Why it mattered

A searchable policy database still leaves the broker to interpret urgency. The interface needed to turn stored records into an ordered workload.

Intended outcome

Faster identification of renewals requiring attention without hiding the underlying policy context.

Separate system truth from AI interpretation

Decision

Keep dates, premium values, policy status, renewal windows and workflow state deterministic. Use AI for narrow interpretation tasks such as extracting unstructured fields, explaining context and preparing editable communication drafts.

Why it mattered

A model should not become the source of truth for policy state or financial totals. Those values must remain testable, predictable and traceable.

Intended outcome

AI assistance without making core workflow behaviour probabilistic.

Show the evidence beside the recommendation

Decision

Keep the policy, expiry date, premium, previous contact, risk context and proposed draft visible in the same review surface.

Why it mattered

A risk label alone asks the broker to trust a conclusion. The surrounding record gives them enough information to assess it.

Intended outcome

A faster decision with less blind trust and less navigation between systems.

Prepare overnight but never send by default

Decision

Generate renewal drafts for upcoming policies as background work, then place them into a broker review queue.

Why it mattered

Automation creates value by removing repetitive preparation. Automatically contacting a client would cross a more consequential authority boundary.

Intended outcome

Less preparation work while keeping the broker responsible for the final message.

Make approval part of the policy history

Decision

When a broker approves a communication, record that event against the policy.

Why it mattered

The action should not disappear after the draft leaves the queue. The wider team needs to know what happened and when.

Intended outcome

A more inspectable renewal process and a clearer communication trail.

Renewal review queue showing Northbridge IT Solutions policy context, a high-risk explanation, an editable draft, and Regenerate and Approve controls. — AI prepares a risk explanation and an editable renewal draft. The broker can inspect the policy context, change the message, regenerate it or approve it. In the public demo, approval is logged but delivery is not connected.

AI sits inside the workflow, not above it.

RenewalPipeline does not ask a model to manage the book of business.

The application owns the records, dates, totals, statuses and approval state. AI is used where the source material or output is unstructured and where a person can inspect the result.

Source layer

The product accepts structured spreadsheet inputs and unstructured PDF policy schedules.

Spreadsheet columns are mapped into known fields.
PDF data is extracted into a structured review state before it can become part of the book.

Deterministic product layer

The application owns:

Policy records
Renewal and expiry dates
Premium values
Due and overdue windows
At-risk totals
Workflow status
Tasks
Communication history
Approval state

These values must not depend on generative output.

AI assistance layer

The Anthropic API supports narrow, reviewable tasks:

Extracting policy fields from unstructured PDFs
Producing contextual risk explanations
Preparing editable renewal communications
Regenerating a draft when the first version is unsuitable

Human authority layer

The broker:

Reviews extracted information before it is saved
Inspects the policy context
Edits or regenerates communication drafts
Approves the final communication
Retains responsibility for client contact

Outcome hypotheses, not reported results

Because RenewalPipeline is a self-initiated concept, it does not have real customer or commercial outcome data.

The appropriate next step would be to baseline the current renewal process and measure whether the product changes observable behaviour.

Operational measures

Time from import to a usable renewal view
Time required to identify the next policy needing action
Percentage of upcoming renewals contacted before expiry
Number of overdue policies with no recorded action
Task completion rate
Frequency of policy-record corrections after import

Commercial measures

Premium at risk by renewal window
Renewal rate
Lost premium and recorded loss reasons
Number of renewal opportunities left unworked

AI-quality measures

PDF extraction correction rate
Draft approval rate
Draft regeneration rate
Amount of editing before approval
Frequency of missing or invalid fields
Fallback or manual-processing rate

Adoption measures

Successful import completion
Time to first useful dashboard
Repeat dashboard use
Repeat review-queue use
Percentage of active renewals processed through the workflow

What this demonstrates

RenewalPipeline demonstrates how AI can be introduced into an existing operational workflow without asking users to surrender authority.

The value is not autonomous email generation. It is a faster, inspectable route from fragmented source data to a decision a broker can review, approve and audit.

WORKING PRODUCT

Inspect the workflow yourself.

Explore the dashboard, import workflow, policy records and AI-assisted renewal review queue using sample data. Nothing sends, so the experience can be inspected safely from source data through human approval.

Explore the interactive demo

Next project

HubSpot's AI Breeze, reimagined.

View case study

Have a complex workflow to solve?

I help B2B SaaS teams turn fragmented operational work into products users can understand, control and adopt.

Explore the interactive demo Book a Call

01Context and product tensionThe operating pattern

02Decision logSix product choices

03Build and AI architectureWhere the model sits

04Measurement and outcomesWhat I would measure

Inspect the workflow yourself.

Have a complex workflow to solve?