Back to Explore
Interactive Exploration

When Decisions End in (Ex)Cell

Cost estimates, vendor comparisons, change orders — they all end in a spreadsheet. The question is what happens when the spreadsheet starts thinking with you.

7 Interactive Demos
Document-Connected Workflows
Provenance & Verification

1. The problem: numbers without judgment

It's 11 PM. Dana — procurement lead at a $180M mechanical sub — stares at cell D14: $142/CY for concrete. The formula is right. But where did that number come from?

Across thousands of projects, teams reverse-engineer the reasoning behind their own spreadsheets every night. Not because the data is wrong — because the judgment doesn't travel with it. Click through to trace a cell back to its source.

ItemQtyUnit CostTotal
2
Concrete (4000 PSI)
240 CY
$142
$34,080
3
Rebar #5
18 tons
$1,180/t
$21,240
4
Formwork rental
1 lot
$8,600
$8,600
5
Labor — Foundation
320 hrs
$68/hr
$21,760
Clean numbers. Zero provenance. Can you defend these in a meeting?

2. The workbook stays. The infrastructure changes.

Instead of replacing Excel, add an intelligence layer underneath it. Five capabilities — ingestion, querying, template filling, report generation, and verification. Click each capability below to see what it does.

PELLES

Click a capability to explore

3. Upload and unlock

Before anything else, the system has to understand what your workbook means — not just raw cells, but headers, tables, key-value pairs, and document structure. Watch how a workbook gets analyzed on upload.

Schedule_of_Values_v2.xlsx — SOV
1Riverside Plaza — Schedule of ValuesContract: $2.4MAug 2024
2GC: Riverside Builders
3
4LineDescriptionScheduledCompletedBalance
53.1Foundations$340,800$312,000$28,800
65.1Structural steel$286,000$143,000$143,000
715.1Mechanical$194,200$58,260$135,940
816.1Electrical$167,500$41,875$125,625
9
10Total$2,412,000
11Retainage (10%)$241,200
12
13Per AIA G703 format. Retainage held per contract.
Every GC formats their SOV differently. Can the AI tell data from structure?

Primavera P6

.xerSchedules, activities, resources

MS Project

.mppTasks, calendars, assignments

Accubid

.xlsxEstimates, bids, change orders

Smartsheet

.xlsxExported for AI analysis

4. Every number carries its source

Dana's $142 looked clean — but was it based on Rev 2 or Rev 3 delivery terms? Verification cross-references your spreadsheet against source documents and flags mismatches before they become disputes.

Cited Answers

Ask “where did this come from?” and get a response pointing to the exact spec section, contract clause, or drawing.

Stale Data Flags

When a unit price doesn't match the latest contract revision, or delivery terms changed between versions — the system flags it.

Construction-Native

Built for JIS, cost estimates, schedules, and industry-standard document structures — not generic document search.

5. Full reports from a single prompt

“Analyze cost variance for Phase 2 against approved budget.” The query engine scans your project documents, extracts the data, and assembles a formatted spreadsheet. Step through the workflow below.

Report Workflow
Press Play to generate a cost variance analysis
PromptScanningExtractingAssemblingComplete

6. Your format, your methodology

Dana's vendor comparison template encodes twelve years of judgment. Upload any .xlsx — the system fills it from your project documents while preserving formatting, formulas, and layouts. See a template get filled in real time.

Dana’s Submittal log
Template

Submittal Log — Building C Foundation

Project: ________ PM: ________

ItemSpec §SubcontractorDue DateStatus
3
4
5
6
7
Your format. Your columns. Your institutional knowledge encoded in layout.

7. Ask questions where the work happens

No copy-paste, no tab switching. Highlight a cell, type a question, get a cited answer — right inside the workbook. Try the plugin demo below.

Pay_App_07_Draft.xlsx
ItemScheduledPreviouslyThis PeriodSource
2
3
4
5
6
7

AI writes results directly into your workbook — not a chat bubble

8. Dana's Thursday night, revisited

Same workbook. Same cell D14. Same question. But now with an intelligence layer behind the spreadsheet. 46 minutes becomes 4. Compare the before and after.

Press Play to see Dana’s Thursday night

9. What we're still figuring out

This isn't a silver bullet. Three hard problems remain.

Template ambiguity

A section labeled “Misc.” with twelve unlabeled rows forces the system to guess. Cleaner templates get better fills — your worst-maintained ones need the most hand-holding.

Revision hell

Six drawing revisions, three referenced across different email threads. We flag conflicts, but “which revision did we actually bid from?” sometimes still needs a human call.

The last 10%

90% done in seconds. 100% — matching formatting quirks, merged cells, formula dependencies — still takes iteration. Faster than from-scratch, but not hands-off yet.

Where this works best today: Teams with well-structured templates and consistent naming see immediate value. Teams with fifteen versions of “Final_Estimate_v3_FINAL_USE_THIS.xlsx” have cleanup work first. The system multiplies your existing organization — it doesn't replace it.

10. Before and after

Excel Today

  • -46 minutes to trace one number back to its source
  • -A cell comment that says 'from Martin bid' — no revision, no date, no section
  • -Six tabs of vendor data with no connection to the actual contract terms
  • -Re-keying Accubid exports into your own template, row by row

Excel + Intelligence Layer

  • +Highlight a cell, ask where the number came from — get a cited answer in seconds
  • +Describe a report in plain English, get a structured spreadsheet back
  • +Cross-reference against specs before the number becomes a dispute
  • +Upload P6, Accubid, MS Project natively — no manual re-keying

“The workbook stays. The infrastructure underneath is what changes.”