BrewOS — the route-to-market platform we’ll build with one brewer

The industry ledger — public record, not our numbers

$52.5B

GMV AB InBev routed through its in-house-built BEES platform in FY2025 — 72% of its revenue now arrives through digital B2B. Owning the stack works.

1,200+

Engineers — and six years — AB InBev invested to build BEES, per its own investor seminar. That was the price of owning the stack, before AI-augmented engineering changed the arithmetic.

ERPs Heineken’s CEO told investors the company still runs — “non-harmonised process, non-harmonised data.”

£54.5M

“Indirect access” fees SAP claimed after Diageo’s Salesforce-based ordering systems read its SAP data — the High Court upheld liability (the sum itself was never awarded in full).

Sources, all public: AB InBev FY2025 results (SEC filing) & its BEES investor seminar, Dec 2021 · Heineken capital-markets event, Oct 2025 · SAP UK Ltd v Diageo Great Britain Ltd [2017] EWHC 189 (TCC). No affiliation, endorsement or customer relationship is implied — every company named is cited from public record only.

The pattern is already proven: owning route-to-market at brewer scale moves tens of billions and cuts the ERP dependency that blocks the best features. What’s new isn’t the idea — it’s that the next brewer shouldn’t need a 1,200-developer platform organisation to follow it.

Voices from the industry

The pain, in the industry’s own words.

The people who run the world’s biggest beer businesses have said all of this on the record — at investor events, in interviews, in transcripts anyone can read.

THE RETAILER EXPERIENCE

“At the end of the day, retailers shop for a living. We were seeing that all the benefits that consumers saw in their day-to-day lives — in terms of convenience, a wide variety of products at a very good price, flexible delivery — were not being translated to the retailer.”

Chief Growth OfficerAB InBev · interview, 2023

THE LEGACY ESTATE

“…we can leapfrog one or two generations of transformation, going from that world of very inflexible, static, wall-to-wall ERP systems to a world of modular, flexible, cloud-based global platform.”

Chief Executive OfficerHeineken · capital markets event, Oct 2025

WHY IN-HOUSE WINS

“This modern architecture also drastically reduced the platform’s dependencies on our legacy ERP systems, which would have been a barrier to providing many of the most valued and utilized features…”

Global VP of ProductBEES, AB InBev · investor seminar, Dec 2021

Quoted verbatim from public transcripts and interviews; speakers attributed by role and company from public record. No affiliation or endorsement implied.

What we’re proposing

We don’t have a product to sell you.

We have a blueprint and the team to build it — once, with one brewer, as that brewer’s own platform. BrewOS is looking for its founding build partner: a company that would rather own its route-to-market layer than rent it. Here is the deal, plainly.

You get

The platform as your IP — every line of code, from the first commit, in your repositories and your cloud
Roadmap control — built around your OpCos, your returnables rules, your trade
Founding-partner pricing, agreed in writing before anything starts
A clean exit at every stage — code, infrastructure and the knowledge base leave with you

We get

The reference — the next brewer will want to see this one
The domain, for real — live orders, live depots, live empties disputes to build against
A partner who runs the pilot like it’s theirs — because it is

You won’t find customer logos on this page — there are none yet. What you will find is a written offer we expect to be held to, and a public source for every number we cite.

Who we are

Twenty engineers who’ve been inside your problem.

We’re ex-startup founders and early employees who went on to build order capture, loyalty and returnables systems inside global brewers. We’ve sat in the change-advisory meetings. We’ve watched a promo wait a quarter for a middleware release window. We know why that happens — and that it isn’t anyone’s fault. It’s the architecture.

AB InBev proved a brewer can build this in-house — with six years and 1,200 engineers. We’re betting the next one takes twenty, working like a startup instead of a programme.

Nobody on this team closes tickets someone else scoped. Each engineer carries a piece of the platform end to end — knows what they’re building, why it matters, and what it does to the person holding the phone in the outlet. People like that are hard to hire and cost a premium. They’re also the reason twenty is enough.

Our references are individual, not corporate: the systems we’ve shipped, the startups we’ve built and sold. Names and track records on the first call.

How we work

The speed is an operating model, not a typing assistant.

Everyone says AI makes them faster. Here is our actual mechanism — and the published evidence, including the studies that cut against it.

Twenty senior engineers, not a ticket line. Everyone here knows what they’re building, why it matters, and how it lands on the outlet owner. People like that cost a premium — and they’re what makes twenty enough. The AI doesn’t replace them; it removes the waiting between them.

One knowledge base, next to the code. Every decision we make — and the reasoning behind it — is written where every agent can read it. Need to know what the stock team changed and why? Don’t book a meeting for Thursday. Ask your agent. It answers precisely, because it reads the same record everyone else’s agent does.

Nothing merges on an agent’s word. Human review on every change, tests against production-shaped data, small batches deployed continuously. The studies showing AI erodes code quality describe teams without that discipline — it’s the failure mode senior engineers exist to prevent.

And one thing we won’t promise: fast features inside your current stack. A four-vendor estate is slow for structural reasons no tool fixes — every change waits on someone else’s release train. So we build beside it: clean services, ERP and payments mocked behind contract-first interfaces at first, then integrated through interfaces your SAP team approves, in a named phase of the pilot. Nobody is fast in that environment. That’s the reason to build a new one.

knowledge base — ask, instead of a meeting

Simulated — but this is how the team runs: decisions, dates and reasoning, queryable by any engineer’s or PM’s agent. Delivered to you as part of the platform.

The evidence — including what cuts against us

+26%Tasks completed by developers working with AI, in randomized trials across 4,867 engineers at Microsoft, Accenture and a Fortune 100 firm.Cui et al. · Management Science / SSRN 4945566

+56%Faster on well-specified greenfield work, in a controlled experiment — the environment we deliberately create.Peng et al., 2023 · arXiv:2302.06590

−19%Experienced developers were slower with AI on million-line legacy codebases. The strongest argument we know for building clean instead of extending the estate.METR, July 2025 · arXiv:2507.09089

#1DORA’s top unlock for AI-assisted delivery: AI-accessible internal data. The knowledge base is that capability — built from day one, not retrofitted.DORA · State of AI-assisted Software Development, 2025

+45%Average budget overrun on large IT programmes, which also deliver 56% less value than promised. This is the baseline we ask to be measured against.McKinsey–Oxford study of 5,400+ IT projects

You don’t have to believe any of it. The two-week build is the measurement.

The difference, concretely

The same feature, two ways.

Say revenue management wants a returnables-aware combo: buy ten cases, hand in your empties, earn 2× points. Here is what shipping it takes.

in a four-vendor stack

Week 1–2The request goes to the CRM vendor, the commerce vendor, the loyalty vendor and the integrator. Four tickets, four backlogs.
Week 3–6Scoping calls. Each vendor quotes its change request. Deposit netting is “not supported natively” — a workaround gets designed around the loyalty platform’s object model.
Week 7–10Build, across three vendor teams. Cross-system tests wait for a window when all four calendars line up.
Week 11–14UAT, regression, a release window. Empties netting gets cut — moved to phase 2.

Four months if nothing slips, three change-request invoices plus integrator days — and nobody did anything wrong. Coordinating four vendors is what it costs.

the two-week build

Day 1Your commercial lead and our engineers in one room. Combos, empties and points are one domain model — nothing to translate.
Day 2–8Built in the loyalty and returnables services. Agents hold the system’s full context; decisions land in the knowledge base as they’re made.
Day 9Tests against production-shaped data — real order volumes, messy catalog entries, outlets with disputed empties balances.
Day 10Demo. The feature runs end to end: order in, points accrued, empties netted, posting queued.

In our environment, not your stack: ERP and payments are mocked behind contract-first interfaces. Nobody can promise speed inside a four-vendor estate — removing that estate is the product.

The blueprint

What we’ll build, in the order we’ll build it.

Eight services, one domain model, one event bus. Generic commerce knows carts; this blueprint knows beer — crates, kegs, empties, depots, reps, telesales, outlet credit, delivery routes and SAP postings. The four phases below are also the pilot’s four quarters.

Phase 1 · pilot Q1

Capture demand

RETAILER APPS

iOS, Android and PWA from one codebase, branded per OpCo, offline-first for outlets on a two-bar connection. New markets launch as configuration — catalog, currency, language, tax.

REPLACES: agency-built apps, per market

ORDER CAPTURE & COMMERCE

B2B storefront and telesales console for the fragmented trade: customer-specific pricing, credit limits, suggested orders. First outlets ordering for real by the end of the quarter.

REPLACES: SAP Commerce Cloud, Salesforce B2B Commerce

Phase 2 · pilot Q2

Promise accurately

STOCK & DEPOTS

Availability by depot, allocation and substitution rules, promise dates the app can keep — because stock hears about every order the moment it happens, not in tonight’s batch.

SITS ABOVE your WMS — replaces the overnight sync and the ATP guesswork, not the warehouse

Phase 3 · pilot Q3

Execute & reconcile

DELIVERY & ROUTING

Drops, routes, proof of delivery, cash and credit settlement at the door — and empties pickup planned into the same run, because the ledger already knows the balance.

REPLACES: last-mile point solutions

RETURNABLES LEDGER

Empties, kegs, crates and deposits as first-class objects: balance per outlet, netting on the next delivery, reverse-logistics routing — and postings queued to SAP as they settle.

REPLACES: Excel & custom SAP Z-developments

Phase 4 · pilot Q4

Grow revenue

LOYALTY & PROMOTIONS

Points, tiers, challenges, combos — per market, per outlet segment. A new mechanic is a deploy, not a cross-vendor programme. This is the quarter you pick one and watch it ship inside a week.

REPLACES: Talon.One, Salesforce Loyalty

SALES FORCE & BACK OFFICE

Rep app with visit plans, tasks and perfect-store checks; catalog, pricing and promo administration your commercial team drives without a ticket.

REPLACES: Salesforce, Microsoft Dynamics 365

DATA, API & MCP

Every event lands in a queryable store as it happens. A public API and an MCP server mean your analysts — and your AI agents — work against the whole system, not last night’s export.

REPLACES: SAP PI/PO, Boomi, MuleSoft & ETL sprawl

Beside SAP — not instead of it

Every service publishes to and subscribes from one event stream. An order placed at an outlet is known to stock, loyalty, delivery, analytics and finance in milliseconds — no nightly batch, no middleware translation chain, no “the numbers reconcile on Tuesday.”

SAP stays your financial and audit core. It receives orders, stock movements and postings the way it already consumes IDocs, BAPIs and OData — through interfaces your team approves. The platform takes ownership capability by capability, a strangler pattern with old and new running side by side, and nothing dual-writes the same truth. No step is a big bang.

System of record, per flow

SAP stays authoritative for finance, customer master and pricing conditions; SD, MM and your WMS stay in charge until their flow cuts over. Stock truth is never in doubt.

Per-OpCo by configuration

Catalog, pricing, tax, language, promo rules and returnable logic are configuration per OpCo — one platform, not forty parallel implementations.

Audit by construction

An event stream is an immutable record. Access control, change management and deploy approvals are designed to your ITGC framework from day one.

Your cloud, your code

Runs in your tenancy. The repository is yours from the first commit. Walking away from us is a staffing decision, not a migration.

The economics

Budget lines that phase out, flow by flow.

Real numbers this time, with sources — list prices and public rate cards, flagged as such, because your negotiated figures are yours to model in.

Per-seat CRM and field-sales licences — Salesforce Consumer Goods Cloud, the SKU a brewer’s field organisation licenses, lists at $275 per user per month. A 5,000-rep org is ~$16.5M a year at list.

sunset over rollout

Commerce platform take — Salesforce prices B2B Commerce as a share of GMV; procurement advisors report 1–2%. On $1B of digital orders, that is $10–20M a year, growing with your own success.

retired as flows move

Integration middleware — MuleSoft contracts run a median $69k/yr in recorded transactions (mid-market; enterprise deployments run far higher), and SAP Integration Suite lists at £4.6k per tenant per month.

retired as flows move

The change-request programme — at published framework rates (£1,880–£2,100 a day for a level-6 Accenture or Deloitte consultant), a ten-person blended change team runs roughly £2.8M a year.

avoided per feature

Indirect-access exposure — the licensing risk SAP pursued in SAP v Diageo: your own portals reading your own ERP data through someone else’s contract.

risk → capped

For context: Deloitte’s 2023 survey of 1,179 technology leaders found 48% of tech budgets go to keeping existing systems running, and only about 20% to creating new value. The lines above are where that 48% lives.

Run-rate, side by side — modelled in the open

the brewos team — transparent math

20 senior engineers · €10k / month€2.40M /yr

Performance bonus · 20% of salary€0.48M /yr

AI agents & tooling · ~€1.5k / engineer / month€0.36M /yr

Platform infrastructure · your cloud, pilot → scale€0.4–0.8M /yr

All-in≈ €4M / year

Fixed and predictable — and it covers everything: features, maintenance, new markets, support. No per-seat licences. No change-request invoices.

a typical stack today — unoptimized

CRM, SFA & commerce seats · Salesforce / Dynamics, thousands of users€5–10M /yr

Integration middleware & ETL platforms€0.6–1.5M /yr

Loyalty & promo suites€0.5–1M /yr

Integrator change requests & upgrade programs€2–5M /yr

Per-market agency builds & rebuilds€1–3M /yr

Run-rate€9–20M+ / year

Illustrative — modelled from public list prices and industry figures, not your invoices. And it buys the delivery speed described in the changelog above.

Honest caveats: vendor figures are list prices — real enterprise agreements are discounted 20–40% — and consultancy rates are framework ceilings. That’s why week one of any engagement rebuilds this model with your procurement team, against your actual contracts. You keep the model whatever you decide.

Sources: salesforce.com & microsoft.com published pricing · Vendr marketplace transaction data · Accenture & Deloitte UK G-Cloud 14 rate cards and the public SAP Integration Suite G-Cloud pricing document · Deloitte Global Technology Leadership Study 2023 · SAP UK Ltd v Diageo [2017] EWHC 189 (TCC).

Fair questions

What your board will ask.

These are the questions enterprise procurement actually asks a twenty-person firm with no product. We’d rather answer them here than have you wonder.

You’re twenty people with no product. What happens if you fold?

You lose a supplier, not a system. The code lives in your repositories and your cloud tenancy from the first commit; payments are staged against delivered milestones; there is no licence to expire and nothing hosted on our side. If we disappear, continuing is a hiring decision — the repos, the infrastructure definitions and the knowledge base are already yours.

Custom builds fail. McKinsey says large IT projects run 45% over budget.

And deliver 56% less value than promised — we cite that study ourselves. It describes megaprojects: multi-year scope, big-bang cutovers, value deferred to the end. The ladder is built to be the opposite — a two-week build you can walk away from, then a one-OpCo pilot with exit criteria in writing, and you keep every artifact at every exit. The worst case is designed to be cheap.

Twenty people means key-person risk.

The knowledge base is our answer before it’s a speed tool: every decision, its date and its reasoning is written where any successor — human or agent — can read it. Context lives in the system, not in heads. Named-team and replacement commitments go in the contract.

Who owns the IP — including the AI-generated code?

You do. Work-for-hire plus explicit assignment of everything created for you, an enumerated list of our pre-existing tools with a perpetual licence, and indemnification that covers AI-generated output. Your lawyers get standard clauses, not novel ones.

Which AI tools touch our code and data?

Named model providers under enterprise agreements — zero retention, no training on your data. The knowledge base and the repositories live in your tenancy. Agent actions are logged, nothing merges without human review, and model vendors appear on the subprocessor list like any other supplier. You get the toolchain in writing, not a black box.

You’re mocking SAP in the early phases. Isn’t that deferring the hardest part?

Integration is the hardest part — mocking is how we sequence it, not how we skip it. The mocks implement contract-first interfaces designed against your real SAP objects from day one, and the pilot has a named integration phase with your SAP team as a planned dependency, not a surprise. What the mocks buy is speed on everything that doesn’t need SAP to be answered: does ordering work, do reps use it, does the returnables ledger reconcile.

Who runs this at 3 a.m.?

During the pilot, we do — with SLAs and credits in the contract. After it: hypercare, then handover. We train the engineers who take it on, yours or anyone’s. We work to make ourselves optional; owning the code means you can hold us to that.

How do we exit mid-way?

At any stage boundary, with everything: code, infrastructure-as-code, data, documentation and the knowledge base itself — all already in your systems, because that’s where they live from day one. Your exit cost is designed to be approximately zero at every rung.

Why not give our own teams AI tools instead?

Do — and measure it. Bain found teams using coding assistants alone gain 10–15%; the 25–30% gains came from redesigning the whole delivery lifecycle around them. The operating model is the hard part, and it transfers with the code. Also honest: your existing estate is exactly where METR measured experienced engineers getting slower with AI. Greenfield is where the gains live.

We just spent years consolidating our ERP. Now this?

Keep it — that consolidation is the foundation we build on. This platform deliberately does not compete with your financial core; it replaces the layer above it, where features die in vendor queues today. A lean ERP core plus an owned engagement layer is exactly where your ERP programme is trying to get.

Who this is for

Built for brewers who’ve outgrown the sprawl.

A founding partnership needs a specific kind of company. This is the honest sketch of the fit — in both directions.

It works if you have

Multiple OpCos and fragmented B2B ordering across markets
An ERP consolidation already underway — and worth protecting
Separate CRM, SFA, commerce, loyalty and middleware vendors
Returnables and empties complexity your systems fight
Features stuck in vendor queues, quarter after quarter
The ambition to own the route-to-market layer

We’re the wrong choice if

You need certified packaged software with a global 24/7 support organisation from day one
Procurement can’t contract a twenty-person firm — better we both know on the first call
No OpCo can own a pilot operationally — someone on your side has to want this

Ways in

Start with a conversation.

Three steps, each bigger than the last — and each one earns the next. Stop at any of them and keep everything produced so far.

Free · 45 min

Discovery call

A normal first call: how your route-to-market runs today, where features get stuck, how we work, whether there’s a fit. We mostly ask questions. NDA welcome.

Paid · 2 weeks

Feature build

A pod of our engineers takes one real feature — a combo, a returnables rule, a telesales flow — end to end in a clean environment, ERP and payments mocked behind contract-first interfaces. Scoped and priced on the call. The working software and the code are yours.

Pilot · 12 months

One OpCo, real orders

The four blueprint phases become the four quarters. Exit criteria agreed in writing — digital order share, feature lead time, cost-to-serve, returnables-reconciliation accuracy, stack cost retired. Miss them and you keep everything anyway.