Two weeks. $7,500. You keep the findings whether or not you keep us.
A fixed-scope engineering diagnostic of the systems that move your freight and your invoices. Ten working days. We instrument what is already running, watch it through a peak window, re-price ninety days of your own carrier invoices, check what your AI costs and returns, and rank every failure by what it costs you per month — evidence attached to each line.
If you are not sure where the problem is, start here. Nothing after it is required.
What a call can and cannot do
Usually in the length of the sales process, and in a scope written to be won rather than to be right.
Take the free call if the question is whether we know the domain — thirty minutes, an engineer, bring the thing that is broken. What a call cannot do is tell you where the money is going. That needs read access to production data and two weeks of somebody’s full attention.
Few firms in this space publish a price for discovery. We publish one fixed number because a scoped engagement should be budgetable before the sales call: $7,500.
The price is also the filter, and it cuts both ways. It earns us the right to ask for a read replica and your carrier contracts on day one, and it means we spend the ten days on companies that intend to act on the answer.
in SLA penalties, when the window is missed once.
Ten working days, five phases. Here is all of it.
Published in full, because the ask is the revealing part. If this is more access than you would give a vendor in week one, better to know before either of us signs. NDA and access safeguards are agreed in the scope before work starts.
What we ask you for, named, so you can check whether you have it.
- 01Ninety days of carrier invoices as they actually arrive — PDF, EDI 210, portal export. Not a cleaned extract. The mess is the data.
- 02Rate cards and accessorial schedules for your top carriers, with the contracts behind them: base rate, fuel surcharge formula, detention, residential, reweigh, reconsignment.
- 03Ninety days of EDI logs — 850, 856, 810, 940, 945, and the 997 acknowledgments, including the rejections nobody read.
- 04A read replica or restored snapshot of the operational database, plus read-only monitoring access if monitoring exists.
- 05Twelve months of incident history — tickets, pages, post-mortems, or the channel where nights get handled when there are none.
- 06A named peak window in the next two weeks: the shift, the day, or the promotion you expect to hurt.
- 07Two hours with whoever built the integration layer, one with your AP lead, one with whoever carries the pager.
We measure the systems at normal volume, so the peak run has something to be compared against.
- 01Timing on order ingest, carrier rate calls and EDI dispatch — lightweight tracing where it can be added without a deploy, log-derived timings where it cannot.
- 02Carrier API latency and error rate per carrier, p50 and p95. One carrier changing an endpoint took a client's rating flow down for three weeks (C01); we look for the same failure waiting in yours.
- 03The database as it is: slowest twenty queries by total time, bloat, partitioning, what VACUUM costs and what it competes with. At one fulfillment platform VACUUM ran for more than six hours on a 3.2-billion-row orders table, competing with production writes (C08).
- 04EDI and ASN health: 997 acknowledgment rate, timeouts, rejections that surfaced to nobody, and transmitted 856 carton records against what physically shipped — a gap of about 4% at a non-asset 3PL with major US retail accounts (C05).
We watch the window you named, live — your volume, your SKU mix, your carriers, your staffing on that shift, not a synthetic benchmark.
- 01Queue depth and where it builds; latency in the worst hour rather than the average; cutoff proximity; where retry storms start; which manual workaround the floor reaches for.
- 02Warehouse software rarely errors at peak. It gets slower, and operations absorbs the difference in overtime.
We rebuild the invoice arithmetic ourselves, line by line, against your own contracted rate card. Ninety days, both directions.
- 01What carriers overbilled you — variance by carrier, lane, accessorial code and dollar band, including under $50, where the money usually is. At one LTL broker a manual dispute took 45 minutes to build, so anything under $50 went unchallenged — at 12,000 invoices a month, that was thousands of variances (C06).
- 02What you delivered and never billed. The quieter half. Retroactive accessorial reconciliation and unbilled services ran $45,000 a month at one non-asset 3PL (C01).
- 03Whether each line traces back to the operational event behind it. Where it does not, the missing trace is the finding.
Every finding gets a monthly cost, a confidence level, its evidence, and effort in engineer-weeks. Ranked by annualized cost, not by technical severity and not by what we would enjoy building.
- 01Where a finding is a day of work for an engineer who already knows your codebase, we mark it yours.
- 02Handover is a two-hour walkthrough with engineering leadership, and a session with finance when the leak sits on the invoice side. The document is delivered, not presented and taken away.
What you get
Scope limits, stated now rather than discovered in week two.
It will not fix anything.
Ten days buys measurement. Nothing we touch changes state. Construction is a different contract and we will price it.
It will not renegotiate your carrier contracts.
We read them to price the variance and hand you the clause each dispute stands on. We are not a freight-audit brokerage and take no percentage of recovery.
It will not stand in for SOC 2 or ISO 27001.
We hold neither certification. The document flags security issues we trip over; it is not an audit and we will not let it be read as one.
It will not tell you your WMS vendor is wrong.
Where the fault sits inside a licensed product we say so, show the evidence, and give you reproduction steps to take to the vendor.
It will not price a rebuild to the dollar.
Effort comes back in engineer-weeks with a range that narrows after the first month of real work. A precise number in week two is a proposal, not an estimate.
It is not an assessment of your people.
We look at systems and evidence. When the constraint is plainly organizational — as at the 3PL where the CTO was personally closing night incidents at 45 to 90 minutes each (C09) — we say it once, and go back to the code.
What happens next
Three doors. All three are normal.
One week-one saving, and one build that recovered $59,000 more per month
Eight months into AI, not one model in production, spend with no owner. In the first week we turned off LLM API calls running from staging into integrations nobody used. Two vendors selling “AI features” turned out to be regex underneath. Monthly AI spend went $42,500 → $16,000, down 62% with no working initiative lost, and the first priority use case — automatic carrier invoice classification — was in production nine weeks later.
LTL, ~12,000 carrier invoices a month, eight carriers with eight rate structures, four people normalizing them by hand in Excel against 30% annual volume growth. Every variance under $50 written off unchallenged, because building one dispute took 45 minutes. The completed audit pipeline increased monthly recovery by $59,000, using invoices the client already held — and the AP team absorbed 40% volume growth without adding a person.
Questions people actually ask
Start where the money is leaking.
Two weeks, $7,500, fixed scope. We instrument what you have, rank the failures by what they cost you, and hand you the findings document — yours to keep and to act on with us, with someone else, or with your own team.