Practice piece · Information architecture

One front door, organized around the questions people bring

One way I've brought separate data and AI guidance together behind one front door, with the two disciplines side by side instead of blended. Shown through an invented transit agency.

In shortWhen guidance lives on two sites, people have to guess which one holds the answer. This piece shows one way I've merged them: map everything, sort it by the questions people bring, and pair the two disciplines instead of blending them. It's an approach that has worked for me, not the only way to do it. The organization in it is invented.

Reading as

Approach: Current-state map, crosswalk, paired architecture
Example: Fernlow Regional Transit, an invented transit agency
Practised in: Enterprise intranet and knowledge-hub work

Try it · find the guidanceinvented menus

Find the rules for using AI with rider data. There's a data part and an AI part. Try the old sites first, then the front door.

Shaped like a two-attempt tree test, menus only. The menus are invented, so your clicks illustrate the idea; they aren't usability data.

data rulesAI rules0 clicks

Fernlow Regional Transit and its menus are invented. Nothing you click leaves this page.

The crosswalk

I map everything before merging anything

A starting point I've often met: data guidance and AI guidance on two sites, each structured its own way.

Why it matters · an illustrative example

Someone at Fernlow needs the rule for using AI with rider data. They check the data office's site, then the AI group's site, then ask a colleague. One front door means one place to start.

Two sites

People have to know which site holds the answer

At Fernlow, the data office's site is organized by team. The AI group's site is a tools list and an FAQ.

inferredIn my experience, a split like this also hides how closely data rules and AI rules depend on each other.

Current state

Every page and section, mapped

Before changing anything, I map every page and section on both sites as they stand.

The crosswalk

One shared set of categories, written as questions

Every piece of content on both sites is mapped to the same categories. I write them as the questions people bring: where do I start, what are the rules, is my use safe, how do I learn it, where's the data, and who can help.

A crosswalk is a table that maps every old page to its place in the new structure.

In crosswalk terms, the dashed lines are one-to-many splits and the shaded categories are many-to-one merges.

The merge plan

I decide what to merge once I know what exists

Mapping first is how I make every merge a decision, so nothing is lost or duplicated by accident when the two sites come together.

Current state · invented sections

Which site has the answer? every page and section mapped

half where each half of the find-it answer lived

Crosswalk · the same sections, mapped to the shared categories

one section split across categoriesa category gathering several sectionsgathers from both sites: a merge decision

Left: the invented sections from above. Right: the shared categories, written as questions. Hover, focus or tap a box to trace it.

The structure

Data and AI, side by side

The structure I reached for here puts one homepage above paired data and AI pages. It's one shape among several; a single merged taxonomy is another. Pairs keep two distinct disciplines distinct.

The homepage

Quick links for the questions people bring

One homepage, with a quick link for each question and a short list of what's new.

Pairs

Connected, and kept distinct

Rules, safe use and skills each get a data page and an AI page side by side. inferredThe pairing shows how the two connect without merging two distinct disciplines.

Learning paths can sit underneath, drawing on both.

In IA terms: two parallel hierarchies, cross-linked level by level, instead of one merged taxonomy.

Guardrails with skills

I put the guardrails next to the skills

A page that teaches a skill links to its safe-use page at the same level, so the guardrails arrive with the skill instead of after it.

Reference lists

Spreadsheets become governed lists

Reference content, such as the list of data domains (the major subject areas of an organization's data, such as riders, routes or fares), moves out of spreadsheets into a list with an owner for every row, so it can be governed and searched.

Design review

Comparing approaches with AI

When I compare design approaches with AI, I keep what the evidence shows separate from what I recommend.

Front door · invented exampleschematic

Home

where do I start?what are the rules?is my use safe?how do I learn it?where's the data?who can help?

what's new

DataAI

Rules you found it here

Data rulesAI rules

Safe use

Data safeguardsAI safeguards

Skills links to safe use

Data skillsAI skills

Learning paths data and AI

Reference list · spreadsheet to listschematic

governedsearchable

Design review

AI comparison of approaches→what the evidence shows

what I recommendmy call, kept separate

AI fed the evidence. The recommendation stayed with me.

Where it sends you

A front door that points onward

I want a front door to send people onward: to the right owner, the right page for their role and the right next step.

Link to the source

Owners keep their own content

The front door links to guidance its owners maintain instead of copying it, so each owner keeps their own content current.

The single-source principle: one canonical owner for each piece of guidance, linked rather than copied.

Role-based next steps

Start from your role

Each role gets one page that points to what applies to it, and everyone gets a “start here”.

From principle to action

Every principle ends in a next step

A page that states a principle ends with what it means for you, what to do, and who to ask, so nobody reaches a dead end at "be responsible".

Link, don't copy

Front doorlinks to →the owner's own guidance maintained by that owner

copied guidance kept on the front door

By role

Everyone→“Start here” orientation

Data stewards→Steward page

People building with AI→Builder page

From principle to action · schematic

The principle

What it means for you

What to do

Who to ask

What this approach gives

One starting point instead of several

One place to start

Every question has one first stop, and the two disciplines stay side by side there.

Owners keep their content

The front door links to what owners maintain, so nothing drifts out of date in two places.

A check you can repeat

The find-it task at the top is cheap to run with real menus before and after any merge.

Consolidation isn't the goal in itself. In my experience, a single front door works when it's organized around the questions people actually arrive with.

What it can claim

What this story can support, and what it can't yet

I graded these myself. Every claim is about this approach or this page, so you can check most of them from here.

Claims register · 5 claims, and how well each is supported

observed: Visible on this page.
document-supported: Established by a public source, linked.
inferred: My professional judgement, not tested here.
unresolved: Not determinable from the available record.

Claims register
Evidence	Claim	What supports it
observed	In these invented menus, finding both halves takes at least seven clicks across the two old sites and two from the front door.	The find-it test on this page.
document-supported	A tree test measures how findable things are using the navigation alone, without visual design or search.	MeasuringU, “Tree Testing”.
inferred	In my practice, mapping every page before merging prevents accidental loss and duplication.	My practice.
inferred	Paired data and AI pages connect the two disciplines without blending them.	My practice.
inferred	In my experience, a single front door works when it's organized around the questions people arrive with.	My practice.

Mine

The approach, the invented example, the writing and the interactive pieces.

Where I've practised it

I've used this approach in enterprise intranet and knowledge-hub work.

AI

I build pages like this with AI assistants and check everything they produce.