Pereview: from lease PDF to portfolio

01Context

Investment decisions start in documents nobody wants to retype.

An investment team lives on a handful of numbers per building: occupancy, rent, income against budget, and when each lease ends. Those numbers come from rent rolls, ledgers and leases, which arrive as spreadsheets, exports and scanned PDFs.

Pereview brings the assets, the leases, the financing and the ownership structure into one platform. The hard part is getting the data in, and then trusting it once it is there.

The redesigned asset page: breadcrumb, title and badges, one row of tabs, an AI notice, five KPI tiles with trend lines and a labelled income chart — Fig. 02The asset page at desktop width

The original asset page: a dark blue header with a building photo, two rows of tabs and a grid of placeholder charts titled Heading of the graph — Fig. 02The asset page at desktop width

02Role

One view across four parts of the platform.

I worked across the platform rather than on one feature, which meant the asset pages, the stack plan, the import tools and the AI flow had to read as one product.

That shaped most of the decisions here. A status, a confidence score or a date looks and behaves the same wherever it appears.

Asset pages

Overview, KPIs and charts
Stack plan by floor and suite
Leases, documents and workflows

Data in

Spreadsheet import and mappings
Rent roll and ledger history
Saved templates

AI abstraction

Upload, classify and match
Review extracted fields with their source
Reconcile and schedule critical dates

03Starting point

Good bones, and a lot left for the user to work out.

The platform already had the right parts: a clear navy app bar, consistent tables, and a step-by-step import. Reading the screens closely, the work was mostly in what they left people to figure out.

Finding things

Eight items in the top bar, with Add and Admin beside the places to go
A second row of up to eleven tabs on each asset
Charts laid out before their questions were settled

Reading things

A stack plan legend of four colours with no labels
A white-on-orange badge at 3.2 to 1
Errors in banners that cut off mid-sentence

Doing things

"Select header row by clicking on it" before an import could start
Save & Continue on every step, whatever it did
Extracted values linked to the lease by section, not by line

The original import selection step: a table of rows with one row highlighted in blue and the instruction to select the header row by clicking on it — Fig. 03Original import steps: picking the header row, mapping columns, validation and the AI match

The original mapping step: rows of source headers mapped with arrows to Pereview labels and an example value — Fig. 03Original import steps: picking the header row, mapping columns, validation and the AI match

04Principles

Four rules for data people have to sign off on.

Show the source

Anything the AI extracted links to the clause and page it came from. Nobody should have to trust a number they cannot check.

Say it in words

Status is an icon and a word, never a colour alone. Committed, 6 to review, Missing account map.

Let numbers line up

Tabular figures, right-aligned amounts and units beside every value, so columns can be scanned.

One current step

Long flows show where you are, what is left and what will happen when you continue.

05Asset overview

An overview that answers the first five questions.

The overview now opens on the numbers an asset manager asks about first: occupancy, leased, net operating income, in-place rent and weighted lease term. Each tile shows its trend and what it is compared against.

Below them, every chart states its question in the title, labels its axes and units, and explains itself on hover. Lease expirations are shown as a share of the building, with the heavy year called out in words.

Fig. 04Income against budget, with the month explained on hover

Comparisons, not just values

Each KPI carries its change and what it is measured against: last quarter, budget or the previous year.

Three KPI tiles: occupancy 92.4% up 1.8 points versus Q2, leased 95.1% up 2.6 points, and NOI $6.93M up 2.0% versus budget, each with a trend line

Charts with their question

Titles say what the chart answers. Axes, units and legends are always labelled; the tooltip gives the exact figure.

The net operating income chart with labelled dollar axis, a budget line, forecast bars and a tooltip for August 2025: $590,400 against a $571,000 budget

New data announces itself

When the AI abstracts a lease for this asset, a notice at the top says what changed and links straight to the differences.

A notice reading: New lease abstracted by AI. Halcyon Bank, N.A., Suite 1200, 289 of 300 data points extracted, 4 differ from the rent roll, with a Review differences button

06Stack plan

Colour that means a year.

The stack plan shows every floor and suite in the building. In the original, four colours marked the suites and the legend had no labels, so the colours had to be remembered.

The redesign colours each suite by the year its lease expires, from 2026 in coral to 2032 and later in navy, with vacancy hatched. Every suite carries its year in text as well. Selecting one opens its terms and options beside the stack.

The redesigned stack plan: a summary row, floors 18 down with suites tinted by expiry year and labelled with the year, and a panel for the selected Halcyon Bank suite — Fig. 05The stack plan, from an unlabeled legend to colour by expiry year

The original stack plan: floors 8 to 1 with tenant blocks in green, yellow, pink and grey, and a legend of four unlabeled colour squares — Fig. 05The stack plan, from an unlabeled legend to colour by expiry year

Fig. 06Hover for the lease, click for its terms

07Import

Four steps the AI takes, one the team takes.

Leases come in as PDFs, Word files and scans. The import center shows each document moving through the same pipeline: read, classify, match to an asset, extract, then hand to a person for review.

Confidence is explained once, on the same screen: 90% and up is accepted but still editable, 70% and up needs a quick look, and missing values must be filled in before anything is submitted. The history below says what happened to every file in words.

The redesigned import center: a drop zone with the five-step pipeline, three documents processing with progress, and a guide to confidence — Fig. 07Data upload, from a drop zone to a visible pipeline

The original data upload: a dashed drop zone with a Select a file button above a history table of Committed and Error badges — Fig. 07Data upload, from a drop zone to a visible pipeline

Fig. 08Each document shows which step it is on

08Review

Every value keeps its receipt.

Reviewing an abstract is where trust is won or lost. The lease now sits beside the fields. Click a value and the document opens at the clause it came from, highlighted.

Values the AI worked out rather than read, like a rent start date set two months after commencement, are flagged as calculated and show the arithmetic. Missing values are typed in with the page they came from, and the counts at the top fall as the team works.

The redesigned review: fields on the left with page links and a highlighted rent commencement, and the lease open beside them at section 3.2 — Fig. 09Lease data review, with the lease beside the fields

The original lease data step: a seven-step arrow bar and cards of extracted fields with an AI percentage and View lease and Edit buttons per section — Fig. 09Lease data review, with the lease beside the fields

Fig. 10Calculated values show their working; missing ones are filled from the page

300+

Data points read from each lease

Steps from upload to reports

Standard clause types scanned

±30 days

Tolerance when matching lease dates

09Reconcile

Settle the differences, then set the reminders.

Before a lease is saved, the abstract is compared with what the rent roll already says. Matching fields stay quiet. Each difference shows both values and the clause behind the new one, and the team picks which to keep.

Submit stays disabled until every difference is settled, and a panel says what submitting will change. Critical dates, from rent start to renewal notice, become alerts with an owner and a lead time.

The redesigned finalize step: the abstract compared with the rent roll, four differences resolved with Lease or Rent roll choices, and a ready to submit panel — Fig. 11Finalize, with the rent roll comparison moved ahead of submission

The original finalize step: four count cards and a list of critical-date notifications with toggles, day dropdowns and team dropdowns — Fig. 11Finalize, with the rent roll comparison moved ahead of submission

Fig. 12Four differences, settled one at a time

10System

A system narrow enough to stay consistent.

The working file had 49 colour styles, 24 text styles, 68 effect styles and 476 variables, and text set in six typefaces. The system keeps Pereview's navy, amber and blue, and narrows the rest to 30 colour tokens and two interface typefaces: Manrope for headings and key numbers, Inter for everything else. A serif is kept for one job, quoting the lease.

Amber stays the brand accent but never carries text on white, where it measures 1.9 to 1. A violet is reserved for one meaning: this value came from the AI.

5.8 : 1

Ink blue on white, for every action

1.9 : 1

Amber on white: accent only, never text

Colour tokens, from 49 styles

Interface typefaces, from six

Fig. 13Foundations: contrast, type, data colour and status

Fig. 14Components built for review: one field in every state

11Mobile

For the minutes between meetings.

The phone app does a few things well rather than everything: see what needs attention, check an asset, confirm a clause, settle a difference and keep track of dates. Floors become bars, tables become lists and the main action sits in reach of a thumb.

Fig. 15Eight phone screens across the same journey

Review

Fields with their clause

Each field shows the sentence it came from, with Accept and Edit beside it.

Reviewing on a phone: rent commencement confirmed with the clause quoted, and the lease execution date entered

Clauses

The lease, quoted

Key dates on top, the verbatim clause below with the terms that matter highlighted.

Dates

What is coming up

Notices, expirations and options across the fund, grouped by month with days remaining.

Critical dates on a phone: November and December items with days remaining

12Outcome

One platform, one way of showing what is true.

The redesign carries one idea through the platform: show people the source, the status and the next step. An asset page that answers questions, a stack plan that reads without a key, and an AI flow where every value can be traced back to the lease and confirmed by a person.

Screens, 5 desktop and 8 phone

Colour tokens

Interface typefaces

5.8 : 1

Contrast for every action