Schedule intelligence · forecasting · claims analytics
Schedule forecasting for construction programmes.
Zepth Rx is schedule intelligence for the owner’s side of the table. It reads the programme your contractor submits, learns how your project actually behaves, and tells you when it will finish, what it will cost, and whose delay it is.
- Deterministic engines compute every number
- Primavera P6 .xer and Excel
- Every figure opens into its own derivation
The programme describes intent. It never tells you the truth.
A contractor’s programme is a statement of what they mean to do. It arrives every month, re-baselined, re-coded, re-sequenced. The completion date on the front page is a single number with no confidence attached to it, produced by a network nobody outside the contractor’s planning team has recalculated.
By the time a slip is undeniable it is usually months old, the float that absorbed it is gone, and the question has stopped being are we late? and become whose is it, and what does it cost? — a question now answered by a forensic exercise costing six figures and taking months.
- The date has no confidence attached.
- A single finish date says nothing about the range of outcomes, or the probability of any of them.
- The truth is buried in the revisions.
- The evidence of what actually happened is spread across twelve monthly submissions in which activity codes have been renumbered.
- The analysis cannot be defended.
- A number in a spreadsheet, with no record of its inputs or the revision it was computed against, is an opinion.
Four things, in order
Read the programme. Learn the project. Simulate. Explain.
- 01
Reads your programme
Import the contractor’s submission — Primavera P6 .xer or Excel. Every import becomes a dated, locked revision. Zepth Rx runs its own critical path over the network they submitted and compares it to what they claimed.
- 02
Learns your uncertainty
It measures how long your trades have actually taken against how long they were planned to take — on this project, from your own completed work. That measured spread, not an industry assumption, is what drives the forecast.
- 03
Simulates the future
It runs your schedule thousands of times with durations sampled from what it learned, re-running the full critical path each time, and reads the distribution of finish dates.
- 04
Explains and advises
It names the handful of activities that decide your date, shows what recovery would actually take, and — where a delay has occurred — attributes it window by window and party by party, with the reasoning stored.
The spine
Your programme, understood — and comparable forever.
A project holds one or more programmes. Each programme accumulates dated snapshots — one per submitted revision, each with its data date. Comparing consecutive snapshots is how the platform sees change.
The problem: activity codes renumber between revisions. Insert twenty activities into a P6 programme and the outline numbers shift beneath everything downstream. Compare two revisions on activity code and you get nonsense — activities that appear deleted, durations that appear changed, delay attributed to the wrong work.
So every activity carries a stable identity that survives renumbering, and every cross-revision comparison, every attribution and every roll-up runs on that identity. Never on the code.
That is the difference between a comparison that produces evidence and one that produces artefacts.
Locked on import
A submission is immutable. Editing happens on an explicit working copy that keeps the same identities, so the evidentiary chain is never quietly rewritten.
Two pointers, deliberately apart
- The contract baseline
- What every variance is measured against. It moves only on a formal, audited re-baseline, with the reason recorded.
- The current approved revision
- What coordination reads. It moves monthly.
Conflating the two is how variance figures quietly lose their meaning.
Six modules. One record.
Everything a forensic exercise would reconstruct — captured as the project runs.
Program
Recalculate the submission. Compare every revision.
ExploreForecasting
Monte Carlo, calibrated on your project’s own record.
ExploreEarned Value
EVM on a cost basis built from your bill.
ExploreResource Management
Where the plan demands more crew than exists.
ExploreClaim Management
Windows analysis, with every attribution’s reason stored.
ExploreField Data
The site record that turns a plan into a measurement.
ExploreAcross the platform
Executive Dashboard
Portfolio view above the project level — health with P50/P80 completion against contract dates, risk pulse, and field manpower across sites.
Zepth Rx Copilot
An in-app assistant with a command palette that runs analyses and answers questions about the project.
Basis of Calculation
The audit trail behind every asserted figure — inputs with their sources, the arithmetic as steps, and what it was computed against.
Completeness declarations
Every engine result records the dataset it actually covered, and is structurally incapable of claiming completeness while reporting a limit.
Permissions
Seven modules across five functions, with approver as a separate track on claims and schedule.
Language models where judgement matters. Proven maths where numbers matter.
The boundary is absolute, and it is the reason the outputs are usable in a contractual setting.
The AI never produces a number.
Not a date, not a probability, not an impact, not a party. It is used for exactly what a language model is good at: turning free text into a category, and turning a computed result into readable prose.
Classifying an activity’s trade
From its name and work-breakdown position — after deterministic rules have had first refusal, and instructed to answer “unknown” when it is unsure.
Mapping a free-text delay note
A foreman’s note to a fixed condition, under an absolute rule that it must never indicate who is responsible.
Writing the narrative
The prose that accompanies a computed result — commentary on a figure, never the source of one.
And the narrative is checked.
Every figure the model states in prose is reconciled programmatically against the engine’s own value. Anything that cannot be reconciled is flagged on screen, and the calculated figures govern.
Everything else is deterministic.
The critical path, the float, the simulation, the earned value, the attribution and the money are computed by engines whose arithmetic you can open and re-derive by hand.
Every number opens.
Ask any figure in Zepth Rx where it came from, and it answers. This is not a feature on one screen — every earned-value KPI, every money figure in a claim and every milestone forecast carries one.
- Result
- The headline figure, its label and its unit.
- Inputs
- Each value with its source — the exact field, setting or prior computation it came from.
- Steps
- The arithmetic, so a reviewer can re-derive the headline by hand.
- Computed against
- The reproducibility stamp: which snapshot, which data date, which simulation run.
- Caveat
- The honest framing, including what the figure does not establish.
Results declare what they covered
Every engine result records the dataset it actually computed over. If anything limited the run, the result says so — and it is structurally incapable of claiming completeness while reporting a limit. A cap that changes a number is treated as a defect; a cap that changes only a display is labelled as a view.
A plausible number computed from a fraction of the data is worse than no number, because nothing about it looks wrong.
Who it is for
- Owners and developers
- An independent read on the programme you are being given, and an early view of your own exposure.
- Project management consultants
- The analysis at portfolio scale, with the working shown.
- Cost consultants and quantity surveyors
- Earned value on a cost basis built from the bill, with the bill and the programme reconciled rather than assumed to agree.
- Claims and delay consultants
- A contemporaneous windows analysis maintained continuously, instead of reconstructed two years later.
- Contractors
- The same instruments the other side will use, before the conversation starts.
What it takes to run
Stated plainly, because a technical buyer will ask.
- A programme
- Primavera P6 .xer, or Excel with a guided column mapping. Microsoft Project users export to .xlsx.
- More than one revision
- A single snapshot gives CPM, float, schedule health and a Gantt. Comparison, attribution and calibration need a chain.
- A bill of quantities
- For earned value and the money side of a claim. Earned Value stays locked until a cost basis exists, and says so rather than showing zeros.
- Completed work
- For the forecast to calibrate on. Early in a project it falls back to a project-wide spread and then a domain default — and tells you which of the three it used.
- A desktop
- Zepth Rx is a desktop application by design.
On the roadmap — not shipped
Real-time voice interaction · an assistant that acts rather than answers · a daily AI briefing · automatic drafting of the claim narrative · site vision from photographs.
Explore the Platform Suite
See How Our Platforms Work Together
Zepth AI
AI-powered construction automation with specialized agents for risk detection, compliance, and decision acceleration.
Zepth Core
Documents, project controls (cost and schedule), quality and safety, the site, and risk — on one record.
Zepth Vector
Procurement end to end: tendering, purchase orders, deliveries, three-way matching, invoices and vendors.
Zepth Edge
The operating asset and its money: registers, maintenance, statutory inspections, CapEx, budgets and reporting.
Zepth Node
The workforce and its record: verified attendance, leave, payroll, performance and consent-first monitoring.
Zepth API
Open integration platform to connect construction intelligence with your existing tools and custom applications.
Not another scheduling tool. An AI that thinks about your project.
Book a demo and we will run it on your programme.