Delivery Physics

Built for 200 customers.
Treated like thousands.

Most enterprise software really serves maybe 50 to 200 customers — a small, knowable base, not a mass market. Vendors still treat the product as if it were built for thousands. The gap is absorbed by configuration, extensions, and services — which is where the months and overruns go.

The 50-to-200 range is a practitioner view, not a published statistic.

We still need product, development, and implementation skill. We do not need those skills locked in separate departments, handing work over a wall until the context is gone.

Whether the work is an enterprise software implementation, a supply chain software implementation, or an AI implementation, each of those customers matters enough to tune the product to them. AI makes that per-customer tuning economical: it reads the customer’s data, documents, and processes and drafts tailored configuration, test scripts, and integration mappings in days. On the product side, every implementation becomes structured evidence of what that customer actually needed.

One small team should own the go-live. We do not promise a 90% cut in end-to-end implementation time. The wedge is a near-90% cut in specific phases, starting with testing.

Placeholder: mailto is hello@example.com. TODO: replace with a scheduling link.

Fig. 1 — The gap

A knowable base treated like a mass market On the left, a product treated as if it were built for thousands of customers. On the right, the actual base of maybe 50 to 200 customers. Between them, a band of configuration, extensions, and services where months and overruns accumulate. Product Treated as if built for thousands of customers THE GAP Config, extensions, and services Months and overruns The base Maybe 50–200 customers, each worth tuning MASS-MARKET STORY KNOWABLE BASE A knowable base treated like a mass market A product treated as if built for thousands at the top, and a knowable base of maybe 50 to 200 customers at the bottom, separated by a band of configuration, extensions, and services. Product Treated as if built for thousands of customers THE GAP Config, extensions, services Where months and overruns go The base Maybe 50–200 customers, each worth tuning

The problem

A knowable base,
treated like a mass market

Enterprise supply chain and ERP software is usually sold as one product that must serve thousands of customers. In practice, the installed base is often maybe 50 to 200. That range is a practitioner view, not a published statistic — but it is a knowable set, not a mass market. Each of those customers matters enough to tune the product to them.

Vendors still treat the product as generic. The mismatch is absorbed by configuration, extensions, and services. That is the gap — and it is where programs lose calendar time, wait in reviews, and run over budget.

Delivery Physics exists to close as much of that gap as is honest to close. AI now makes per-customer tuning economical, without pretending that customer decisions, data ownership, or change will disappear.

Fig. 2 — Where the time goes

Work flowing through friction A wide flow from product design narrows through a friction chamber of configuration and services before reaching a single go-live outlet. PRODUCT FLOW CUSTOMER REALITY The gap GENERIC DESIGN ONE GO-LIVE Work flowing through friction A wide product flow and a customer-reality flow pass through a shared gap before a single go-live. PRODUCT FLOW CUSTOMER REALITY The gap GENERIC DESIGN ONE GO-LIVE

Two clocks, one program: the product is already built as if for thousands; each customer still has to be fitted.

Fig. 3 — Queues between departments

Handoffs versus one go-live team Three departmental boxes with waiting queues between them, contrasted with one small team organized around go-live. TODAY Product WAIT Development WAIT Implementation INSTEAD Go-live owned Product Dev Delivery Handoffs versus one go-live team Three departments stacked with wait queues between them, then one small team around go-live. TODAY Product WAIT Development WAIT Implementation INSTEAD Go-live owned Product Dev Delivery

Organization

Why the departments are the problem

The disciplines are not the problem. You need people who can design a product, write software, and get a customer live. The walls are the problem. When product, development, and implementation sit in separate departments, each handoff drops context: why a process was designed that way, what the customer’s data actually looks like, which exception matters on go-live day.

The work also waits. In theory-of-constraints terms, the binding constraint is often not the task itself but the queue between departments — a ticket, a review, a waiting period for the other team. Elapsed time piles up in the handoffs, while each group looks busy. The flow of value is a safe customer go-live, not a completed departmental task.

Organize around that flow instead: a small team with all the necessary skills, owning the outcome — in the spirit of Amazon’s two-pizza teams and you-build-it-you-run-it. For enterprise software, that means one small team that owns go-live, and then sees what the customer actually needed.

Both sides of the gap

How AI closes the gap

Not by bolting a chatbot onto the same relay. By making it economical to tune the product to one customer — drafting the artifacts that usually travel, slowly, between departments — and by turning each implementation into evidence the product can use.

Delivery side

Read the customer. Draft the fit.

AI reads the customer’s data, documents, and processes, then drafts tailored configuration, test scripts, and integration mappings in days. People still review, decide, and accept go-live risk. The draft is no longer the thing that takes a quarter to type.

Product side

Every go-live becomes evidence.

What the customer actually needed — the exceptions, the mappings, the tests that failed — is captured as structured data, not a slide deck that dies at the handoff. With a knowable base, every implementation is evidence the product can use to tune itself, instead of a one-off service project that starts from zero again.

The edge pattern

How we work

We do not try to convert the whole delivery factory from the inside first. We build one workflow fresh, with AI, alongside the old process. We measure it. The numbers decide what happens next.

  1. 01

    Diagnose the delivery system

    Map the handoffs, the queues, and where calendar time actually goes. Be clear about what AI can draft and what must stay human: decisions, data ownership, and change.

  2. 02

    Pick one workflow

    Start with testing — the phase with the strongest published case for a large labor-hour cut. Alternatives include first-pass documentation or data-mapping drafts. Not UAT sign-off. Not the change program.

  3. 03

    Build it alongside the old process

    An Edge path: the new, AI-assisted workflow runs in parallel with the current one, on the same definition of done, with a hard date. The mothership does not have to be reorganized first.

  4. 04

    Measure, then decide

    Compare labor hours and cycle time to a written baseline. Report those separately from decision wait time and sign-off time, which may not move. Let the numbers sell the next step — or stop if they don’t.

Engagements

Two offers. Fixed scope, then proof.

No retainer required to start. The diagnostic is a fixed fee. The proof is one workstream, ninety days, starting where the evidence is strongest.

Delivery Diagnostic

3–4 weeks · fixed $10,000–$15,000

A map of delivery handoffs and where time actually goes. A phase-by-phase read of what agents can draft versus what must stay human on this program. A 90-day experiment design with metrics, and a short decision memo for the sponsor or operating partner.

90-day Proof

One workstream · starting with test scripts

Run the Edge path on a single workflow, against a baseline written down before any agent is introduced. Report labor hours and cycle time honestly, including what did not move. Scoped after the diagnostic — no proof fee is published here until that scope is set.

Benchmarks, not slogans

What the evidence actually says

These figures are from named sources, not Delivery Physics surveys. They are why we will not sell a 90% end-to-end time cut, and why we start with testing.

~9 mo

Median enterprise software project timeline

Panorama Consulting 2026 ERP Report, 170 organizations

30%

Of projects finished over budget

Panorama 2026 — 22.9% slightly over, 7.1% significantly over

22.3%

Of projects finished late

Panorama 2026 — 18.2% slightly late, 4.1% significantly late

20–40%

Estimated effort cut if GenAI is applied to today’s delivery model

BCG, April 25, 2025 — a consulting estimate, not a survey

60–70%

BCG’s highest task-level range, for testing labor

The reason we start the 90-day proof with test scripts

Not 90%

End-to-end calendar time is not a promise we will make

Customer decisions, data ownership, and change still set the clock

The wedge is a near-90% cut in specific phases, starting with testing — and only if a measured proof shows it. BCG’s 20–40% figure is an effort estimate for applying GenAI to the current factory, not a program-duration guarantee. We will not claim otherwise.

Audience

Who this is for

Operators in the program

CIOs, COOs, and vice presidents of supply chain or operations who are implementing — or rescuing — supply chain planning, WMS, TMS, or ERP software, and need a clearer picture of where time is actually going.

PE operating partners

Partners sitting behind a delayed or over-budget implementation who want a fixed-scope diagnostic and a proof on one workstream, not another open-ended services plan.

Not a fit

Teams looking for a tool that replaces judgment, a promise to cut a whole program to 10% of its calendar, or a substitute for the customer still having to decide and own their data.

Founder

Founded by a veteran enterprise supply chain software delivery executive

Questions

Honest answers

No. We do not take over the program, and we are not a substitute SI. The diagnostic maps the delivery system you already have. The 90-day proof runs one workstream in parallel with the current path so you can see what actually gets faster. Your integrator, your vendor, and your team stay accountable for the go-live.

The work is about the physics of enterprise delivery — handoffs, queues, and the gap between a product treated like a mass-market system and a knowable set of customers who each have to go live — on supply chain planning, WMS, TMS, and ERP implementations. It is not a single-vendor practice, and we will not list platforms we have not agreed to name.

Not for end-to-end calendar time on a typical enterprise program. Panorama’s 2026 survey shows a nine-month median, with 30% of projects over budget and 22.3% late. BCG estimates a 20–40% effort cut if GenAI is applied to today’s delivery model, with testing the only task-level range near 60–70%. We will not promise 90% of program duration. We will test a near-90% cut in specific phases, starting with testing, and publish what the proof actually shows — including what did not move, because customer decisions, data ownership, and change still set much of the clock.

No. We start on the edge: one workflow, built fresh with AI, running alongside the current process. If the numbers are good, you have evidence for a wider change. If they are not, you have not rebuilt the factory to find that out.

Start

Bring one program. We’ll tear down how delivery actually moves.

Thirty minutes. No slide theater. We talk about where the handoffs are, which workflow is worth proving first, and whether a diagnostic is even the right next step.

Placeholder: mailto is hello@example.com. TODO: replace with a scheduling link.