Tessera Labs · Release 01 · Kestrel

The delay is notin your systems.It's between them.

Construction, mortgage, title, sales, and finance each run in their own system. Every one of them understands its own part of the transaction. None of them can see the whole path to close. Tessera reads all of them and tells you which homes will miss their date, and why, while there is still time to act.

Request a findings report How it works

Reads the systems you already run
Never writes back to any of them
One hour of your IT team, then nothing

The problem

Where the money actually leaks

Builders know how to build homes. Closing one is a coordination problem. The cost does not sit inside any single system — it accumulates in the handoffs, where a slipped inspection becomes a title delay, becomes a rate-lock extension, becomes a margin hit nobody priced until the week of close.

$10.8B

Annual industry cost of longer build times

NAHB / HBI, June 2025

16.3%

New-home cancellation rate — the highest on record

Redfin, December 2025

9–10 mo

Average build time, up roughly 2.5 months in a decade

US Census / NAHB

$500–700

Carrying cost per day, per delayed home

Assoc. of Professional Builders

The handoff chain

One home passes through five independent systems on its way to close. Each is competent at its own leg. None of them holds the whole chain.

01
Sales
CRM
02
Construction
Scheduling
03
Mortgage
LOS
04
Title & escrow
Settlement
05
Finance
ERP

When the second link slips, the fifth finds out after the date has already moved. The information existed the whole time — it just had nowhere to be assembled. That assembly is the entire product.

What we believe

Two commitments we will not trade away

Plenty of software makes a builder faster. We are trying to make a home cost less to deliver, and to be adopted without asking anyone to change how they work. Those two things constrain every decision we make.

Affordability for all

Every day a finished home sits unclosed is priced into the home.

Carrying cost, rate-lock extensions, incentives handed over to rescue a slipping date — none of it makes a house better. It lands in the price a family pays, or in the margin that funds the next community. Coordination waste is one of the few costs in homebuilding that can fall without cutting quality, cutting labour, or cutting the builder. That is the lever we chose on purpose.

Take the waste out of the transaction, not out of the house.

Ease to connect

The only thing a builder has to do is say yes.

No migration. No rip-and-replace. No new system for anyone to learn, and no change to a single existing workflow. We read the tools your teams already trust, and we never write back to them. A new system is a connector definition plus an optional normaliser — measured in hours of our work and about an hour of your IT team’s, once.

If adoption is expensive, it does not matter how good the insight is.

How it works

Above the stack, never inside it

Tessera is a read-only intelligence layer. It sits above the systems you already run, assembles their signals into one timeline per home, and surfaces the risk that only becomes visible when all of them are read together.

Connect

Connectors are declarative definitions, not bespoke code. Each one polls a system read-only and emits typed signals into the pipeline.

Definitions built for BuildPro, Procore, Encompass, Qualia, JDE, Salesforce and Acumatica
Nothing installed, nothing changed in the source system
Twelve months of history ingested before you make any decision

Correlate

Signals land on a dependency graph per home. The graph is what makes cross-system risk computable: it knows which milestones gate which.

51 signal types across six categories, resolved onto 45 milestones
Critical path recomputed as each signal arrives, not on a nightly job
A reason chain, not a red dot — the score always explains itself

Act

A finding is worthless if it arrives after the date has moved, or lands with someone who cannot act. Escalations are routed and timed.

Named blocking milestone, days overdue, and who owns it
Routing per division, falling back to a global rule
Earliest possible close date in business days, not a vague warning

Read-only by construction

A connector definition can only declare a GET or HEAD request. Anything else fails validation and the connector never loads — so a write back into a builder system is not a policy we promise to keep, it is a request the platform cannot express. Your system of record stays yours, untouched.

Release 01

Kestrel

Hovers above your stack. Changes nothing. Sees everything.

Kestrel · Home detail

1245 Elm Ct

Community: Windmill Ranch
Target close: Jul 11
Contract: Jan 18

Confidence

At risk

How the score is built

Closing confidence decomposition by transaction category
Category	Milestones	Complete	Weight	Points
Construction	11/12	92%	35%	32.1/ 35
Mortgage	3/10	30%	25%	7.5/ 25
Title & escrow	5/8	63%	20%	12.5/ 20
Sales	3/4	75%	10%	7.5/ 10
Financial	3/5	60%	5%	3.0/ 5
Utility	4/6	67%	5%	3.3/ 5
Weighted subtotal				65.9
Staleness — Encompass quiet for 31h				-5.0
Closing confidence				61

What is blocking

9 days overdue

Clear to close

Construction is all but finished and mortgage is the constraint. No single system can see that: the scheduling tool reports a home that is ready, and the loan system reports a file that is merely behind. Read together, this home misses its date.

Earliest possible close · Jul 24

Illustrative home — not customer data. The address and values are invented; the scoring arithmetic, milestone counts and category weights are the ones the platform actually uses.

Portfolio confidence

Every active home scored and ranked, rolled up through communities and divisions, so a leader sees the pipeline and a superintendent sees their street.

Parallel category tracks

The six legs of the transaction shown side by side, because they run concurrently and block each other. Progress on one is not progress overall.

A reason, never a red dot

Every score decomposes into the categories that produced it, the milestone that is actually blocking, and how far past its expected date it is.

Escalations that name an owner

Plain language: what is blocked, why, who should move, and how long the window has been open. Routed per division, deduplicated so nobody is spammed.

Works with every system you already run

Sixteen systems across the six legs of the transaction — scheduling, purchasing, CRM, loan origination, title and ERP. Every one of them reads into the same signal vocabulary, so whatever stack a division has grown into, Kestrel speaks to it.

BPBuildPro

ENEncompass

QQualia

JDJD Edwards

PCProcore

SFSalesforce

ACAcumatica

EMEmpower

BTBuildertrend

LALasso

RWResWare

HSHubSpot

SPSoftPro

BLBlend

CCCoConstruct

SAPSAP S/4HANA

BPBuildPro

ENEncompass

QQualia

JDJD Edwards

PCProcore

SFSalesforce

ACAcumatica

EMEmpower

BTBuildertrend

LALasso

RWResWare

HSHubSpot

SPSoftPro

BLBlend

CCCoConstruct

SAPSAP S/4HANA

Proven against live systems

Salesforce is connected to a real org over JWT and verified by a live test. Procore is verified in sandbox over OAuth2. We label the rest honestly: a definition that has been tested against our own harness, not against your tenant. We would rather tell you that now than during a pilot.

What is built

Running today, not a roadmap

The platform exists. Ingestion, streaming, the dependency graph, scoring, escalation and the portal are built and running end to end. These are counted from the codebase, not estimated.

Signal types: 51; Across six categories of the transaction
Milestones per home: 45; 48 dependencies, 35 of them hard gates
Lifecycle templates: 3; Standard, cash buyer, and condo
Systems catalogued: 16; Seven with definitions built

Critical path, properly computed

Hard dependencies only, topologically sorted with cycle detection, then a longest-path pass that skips edges already satisfied. The output is an earliest possible close date in business days.

Stale data is a signal, not a failure

When a source system goes quiet the score is penalised rather than silently inflated, and milestones never touched by a signal still count against the denominator. A builder who is behind on data entry sees a lower score, which is the honest answer.

Replayable by design

Signals stream through Kafka keyed by deal, with idempotent production and a dead-letter path. The graph can be rebuilt from the log, so a scoring change can be re-run over history instead of only applying going forward.

What we will not claim

The shipped score is a weighted completion index with staleness and overdue penalties — not a calibrated probability, and it says so in its own output. Category weights start from operator judgement and are meant to be recalibrated per builder against that builder’s history. The machine-learning path exists behind a flag and is off by default. We will publish accuracy when we have earned the right to, on real data, with a control group.

Who is building it

Operators, not spectators

This is not a team that read about homebuilding. One of us ran the field organisation this product is built for, and left to fix it. The other has spent a career building the kind of detection pipeline it needs.

Teresa Lynch

Founder & Chief Executive

Former Vice President, Lennar

Teresa spent her career inside the largest homebuilder in the United States, running the field operations where the close date is either protected or lost. She has sat in the meeting where five departments each report that their part is fine and the home still misses its date by three weeks. She left to build the layer that would have told her the truth on week one instead of week nine.

She is not guessing at the workflow. She ran it.

Bhavdeep Singh Sachdeva

Chief Technology Officer

Five years shipping detection systems at AWS

Bhavdeep spent five years at Amazon Web Services building the data infrastructure behind GuardDuty and Security Hub — services that ingest billions of events across thousands of accounts and surface the handful that actually matter. He designed organizational-scale pipelines, Delta Lake architectures on EMR, and cross-region replication systems that had to be right every time. The problem Tessera solves is structurally identical: pull fragmented signals from many systems, correlate them in real time, and tell you which deals need attention before it is too late. He built this platform end to end.

The same engineering that detects threats across AWS now detects risk across your pipeline.

Two further operators from the same world are with us and closely involved. We will introduce them when it is right for them, not when it is convenient for us.

Start here

See it on your own homes

Not a demo on invented data. We read your history, score it, and hand you a written report on what we would have caught and when. You keep it either way.

The findings report

Read-only access to one system, or public records if you would rather share nothing
Twelve months of your own history, scored as if we had been running
Which homes we would have flagged, and how many of them actually slipped
One cohort comparison — the community or plan that runs longer, and by how much
Method and limitations stated plainly, including what we could not determine

Your time: About one hour, total
Your IT team: One read-only credential
Your commitment: None

The report is yours to keep whether you continue with us or not, and it is free. If it tells you only things you already knew, we would genuinely like to hear that — it is the most useful thing you could tell us.