The situation
Before the engagement, this winery's back office ran the way most small wine distribution operations do: on retyping.
- Vendor invoices arrived as PDF attachments. Someone opened each one, read it, and re-entered every line into the accounting system by hand.
- Provincial compliance reports were assembled at month-end from exported spreadsheets — a slow, careful exercise where one wrong cell meant redoing the numbers.
- Sales, inventory and orders lived in tools that didn't talk to each other, so nobody had a current picture of the business. The numbers were always days or weeks old.
None of this was anyone's fault. It's what happens when a business grows on top of disconnected tools: each tool works, but a person has to carry data between them. Here, that carrying added up to roughly ten hours every week — and every hour of retyping was an opportunity for a copy-paste error to reach the books or a regulator.
What was built
The system went live as eight connected modules, all built on top of the software the winery already ran — accounting, spreadsheets, inbox. Nothing was migrated, and the team kept working in the tools they knew. These are the five that carry most of the load:
AP invoice reading → accounting drafts
AI reads each vendor PDF, extracts the line items, and matches vendors and accounts. A draft entry appears in the accounting system, ready for one-click approval. Nothing posts without a human sign-off — and the posting itself is deterministic code, not AI. More detail on this pattern in our invoice automation guide.
Compliance reporting
The reports the provincial regulator expects, assembled from live sales data in the exact format required, every period. No AI touches the figures — the same tested logic runs each time, which is why the error rate is near zero. See how this generalizes in automated business reporting.
Live operations dashboard
Revenue, inventory, channel performance and receivables on one screen, updated from the source systems and readable on a phone. The owner stopped asking for exports — the answer is already there. This is the pattern behind our operations dashboard work.
B2B ordering portal
Trade customers place orders directly instead of emailing them in. Orders flow into fulfillment and accounting without anyone retyping a line.
Shipping tracking
Outbound shipments are tracked automatically and their status surfaced alongside the order — no more checking carrier sites by hand or fielding "where is it?" emails without an answer.
The remaining modules do quieter work: keeping records in sync between systems, monitoring the pipeline, and alerting a human when something genuinely needs judgment.
Where AI is used — and where it isn't
A rule shaped every module: anything that must be exact runs as deterministic code. Postings, compliance figures, order totals — tested logic, same result every time. AI is used only at judgment points: reading a messy vendor PDF, matching an unfamiliar line item, triaging an email. And anything AI produces goes through an approval step before it counts.
Why this matters: "AI everywhere" systems fail in ways you can't predict. Splitting the work — deterministic code for the exact parts, AI for the judgment parts, a human at the approval gate — is what lets a small team trust the system with its books and its regulator.
The results
Nobody lost a job. The hours that used to go into retyping moved to the work that actually grows a winery — customers, trade accounts, the tasting room. What changed is that the repetitive layer of the operation now runs itself, and the team reviews instead of types.
How the engagement worked
Audit
We mapped the winery's workflows end to end: which tools held which data, where hours were going, where errors crept in. The output was a costed roadmap of what to automate and in what order — the same deliverable as our AI Ops Audit.
Install
We built the modules on top of the existing tools and ran everything against real invoices, real orders and real reports until the output matched what the team would have produced by hand. First call to production: three weeks.
Operate
The system reports to a monitoring dashboard and alerts us on failure — usually before anyone at the winery notices. Fixes and improvements are covered by a flat monthly plan.
If you're comparing wine distribution software
Most winery operations automation starts from the opposite direction: buy a wine distribution platform, migrate everything into it, retrain the team, and hope the workflows fit. Sometimes that's right. But if your tools already work and the problem is the manual carrying between them, a platform migration is a year of disruption to solve a connection problem.
The alternative demonstrated here: keep the accounting system, keep the spreadsheets, keep the inbox — and remove the manual steps between them. It's faster to deploy, there's no data migration, and the team doesn't have to learn a new system. You own the accounts, the credentials and the roadmap; there's no platform to be locked into.
If you run a winery, a distributor, or any operation that moves inventory on paper-heavy processes, the honest first step is to find out where your hours are actually going — before choosing any tool. That's a 30-minute conversation.