Story 2 Wolters Kluwer · AI product design

AI-Assisted Drafting
for CCH iFirm

Removing the blank page from client documents

3/3Customers validated the concept
ShippedCanadian market
1CCH iFirm
Systems thinking Product strategy Human AI system MVP definition Stakeholder alignment
Built inside CCH iFirm at Wolters Kluwer, their cloud practice management product for accounting and tax firms, and released in the Canadian market. Some of this work cannot be shown as it was built. The interfaces here are redrawn and the data is altered; the decisions, constraints and outcomes are not.
01
The context

Our users were already using AI. They were just using it outside the product.

The brief was about editing. The real problem was starting.

The research insight

Professionals already used ChatGPT to draft emails and written content during their working day.

Product problem

Our product still gave them a blank page to start from.

Design opportunity

Bring that behaviour into the product. Improving the editor would make the blank page slightly better. AI-assisted drafting could remove it.

Research method
  • 5Foundational research sessions conducted in 2024 on AI and usage among Wolters Kluwer customers
02
The problem

From template to AI assistant

Accounting and tax professionals regularly create documents that clients need to review and sign. The original requirement was to improve the document editing experience.

Route

What it meant

Friction

Template

Rarely fit the actual client situation.

Heavy editing anyway

Write from scratch

Slow, and introduced typing errors in client-facing documents.

Risk in a signed document

Leave the product

Draft in an external tool, then paste the content back.

Client information leaves the system

Business goal

Make document creation faster and keep the workflow inside the product.

Friction

Every route was slow.

The problem

The product handed professionals an empty page at the exact moment they needed to start writing.

A new engagement letter open in CCH iFirm with nothing in it: a blank document, an empty editor, and the words nothing here yet above the choice between inserting a template and drafting with AI.
CCH iFirm · the blank page the requirement was written about
03
The opportunity

Make document creation faster and keep the workflow inside the product.

04
The strategic decision

Turn a document-editing requirement into an AI-assisted drafting experience

  1. Original brief  ·  Improve editing
  2. Reframed problem  ·  Remove the blank page
  3. Proposed direction  ·  AI-assisted drafting
The proposal was mine

AI was not in the original requirement. I proposed AI-assisted drafting based on research we already had.

Stakeholder reviews

I defended the direction using the existing user research rather than a new discovery cycle.

Result

Stakeholders expanded the scope of the initiative and phased the work into the roadmap.

Main principle

Don't optimise the blank page. Remove the blank page.

05
The system

AI had to fit a professional workflow, not replace it

AI

Creates a first draft

From the professional's instruction.

The professional

Reviews · Edits · Approves

  • 01Nothing is stored before approval
  • 02Nothing is sent before approval
  • 03The professional stays accountable for the content
Client

Receives the final document

Only after human approval.

Review is not a step after generation. Review is part of the product.

Main principle

Research shows users don't reliably verify AI outputs by default.

Appropriate reliance on Generative AI: research synthesis · Microsoft

06
The experience

The AI creates a starting point. The professional stays responsible for the final document.

Generate → Review → Decide

01

Start

A document is needed for a client.

02

Instruction

The professional describes what the document should say.

03

AI draft

A first version replaces the empty page.

04

Review and edit

The professional reads, corrects and adjusts.

05

Approve

Explicit human decision.

06

Save or send

The document goes to the client for signature.

CCH iFirm · the instruction step. Intent, not prompt engineering
Lo-fi wireflow in six frames: the blank document, the instruction in the user's own words, generating with a cancel, the draft landing in the document rather than a preview, review and edit, and approval as a named act.
The flow used to argue the scope change. The annotations are the argument, not the pixels
The instruction panel

Intent in plain language. No prompt fields and no model picker: choosing the model is the product's job, not the professional's. The panel states what it will not do before it does anything.

The instruction panel with its redlines: 24 pixel padding, a 104 pixel input, a panel between 640 and 720 wide, all spacing on an eight pixel grid, and a line stating that nothing is sent to the client.
The instruction panel · spacing on an 8px grid
Accessibility, written into the spec

Three of the notes the specification carries, so they are built rather than retrofitted.

  • 01The panel is a labelled region, not a dialog. It does not trap focus, and the document stays readable and editable behind it.
  • 02Generation progress is announced politely, so a screen reader hears that work is happening without the draft interrupting whatever is being read.
  • 03Colour is never the only signal. The rail that marks machine-written text is paired with a chip that says so in words.
The full interaction specification: entry point, instruction, review and edit, approve and send, each state drawn and annotated, with developer notes covering that generation is never terminal, that nothing persists before approval, that markers are paragraph-scoped, that failure is not a dead end, that the panel is a labelled region announced politely and colour is never the only signal, and what was left out of the MVP.
Every state, with the notes development needed to build it
07
MVP definition

Not everything useful belongs in the first release

ShippedInstruction-based AI drafting
ShippedFull human review before saving or sending
DeferredReusable templates for generated documents
Why templates were deferred

Templates introduced a second lifecycle: ownership, versioning, editing rights and what happens when the underlying rules change.

Decision

Validate the core drafting behaviour first. A smaller MVP gave us a safer way to validate the core behaviour before introducing lifecycle complexity.

08
The validation

Test small. Release for real.

What we needed to know
  • 01Is it intuitive?
  • 02Is it useful?
  • 03Does the review model feel appropriate?
Result

3 of 3 trusted customers validated the concept positively.

A directional signal, strong enough to justify building the MVP. Not statistical evidence.

09
The build

From strategic proposal to shipped AI feature

Product strategy

Reframed the brief from editing to starting.

AI product design

Proposed AI-assisted drafting and the human review model.

UX

Designed the end-to-end navigation and interaction flow, from wireframes to validation concepts.

MVP

Defined what shipped and what was deferred.

Stakeholders

Defended the direction; scope was expanded and phased into the roadmap.

  1. Concept
  2. Validation
  3. Roadmap
  4. MVP
  5. Canadian release
Build

I helped turn an AI opportunity into a product decision and a shippable MVP.

10
How I would measure it

Are people reading what the AI wrote?

No product analytics were available to me on this work, so this is the plan rather than the result. Written as it would go to the team.

Target behaviourStarting a document from a draft instead of an empty page, and editing it before approving
Event to instrumentDrafts generated, drafts edited before save, drafts discarded without saving
The questionWhat proportion of generated drafts are edited before approval?
At thirty daysA high edit rate is the good outcome here: it means people are reading. A high discard rate points at the instruction step, not the model
10
The outcome

AI moved from an external workaround into the product

User

A faster way to get started

No empty page, and no leaving the product to draft elsewhere.

Product

AI-assisted document creation inside CCH iFirm

With human review built into the flow, not bolted on.

Business and delivery

A validated MVP released in Canada

3 of 3 customers validated the concept. Users needed to accelerate their workflow with AI.

After release

I thought this was a tool for contracts. People used it for anything that needed a signature.

My working assumption throughout design was that the feature served formal, structured documents: engagement letters, contracts, the heavyweight paperwork of a practice.

In use, the pattern was broader. People reached for it for anything that required a client signature, including short, routine correspondence. The opportunity was wider than the brief, and wider than my own framing of it.