AI Assessment for Argea

Section 1

What we heard

∨

Company

Argea: premium Italian wine producer and global distributor. Buys finished wine, bottles it, creates brands, and distributes to importers worldwide. Main clients are large-scale retail importers (Costco, Aldi) and the HoReCa channel.

Contact

Giulio Rossano, Chief of Staff. Supports the CEO on strategy (M&A, investor relations, annual planning) and operations (new markets, org design, process review).

Team

550 people. More than half in operations: bottling, manufacturing, logistics, and purchasing. Remaining: winemaking, wine purchasing, finance, customer service, and sales.

Communication

Microsoft Teams for day-to-day collaboration and video calls. WhatsApp for urgent and executive communication. Email as a third channel. Giulio attends five to six meetings per day.

Meeting notes

Teams Copilot already in use for meeting capture.

Tech stack

Three separate ERPs across different entities. Salesforce CRM, acknowledged as underused ("the offering is done still manually"). MRP system. S&OP demand-planning tool. Dedicated wine movement software (from purchasing through blending to bottling). Bottling efficiency software (machine performance, spare parts, bottleneck identification). Excel and macros used extensively to bridge system gaps.

AI today

Growing and partly official. The CEO, IT team, and Giulio use AI tools. The company has formal access for a small group; the majority of staff use personal accounts.

Main pain

Three separate ERPs with no integration layer. Monthly manual reconciliation of financial data across entities consumes roughly two days of team time. Full ERP consolidation is planned but acknowledged as a long-term effort requiring significant investment.

Pain 2

Order entry: orders arrive by WhatsApp, phone, and email and must be manually keyed into the ERP by customer service. Feasibility checks (pricing, production slots, personalization requirements) are done manually before confirmation.

Pain 3

Production planning: a manual Excel-and-macros process bridges the wine movement software with the main ERP each planning cycle. Demand planning described as "running blind" on monthly forecasts, relying on the annual budget as a proxy.

Pain 4

Intercompany invoices: manual input and reconciliation across three entities, described as hours of work per day for dedicated staff.

Repetitive approvals

Sales offer approvals and expense management approvals mentioned as recurring judgment tasks, likely weekly or monthly cadence.

Key-person dependency

Named areas: winemaking ("a very crafty activity"), customer service department heads, bottling engineers, and IT engineers responsible for software management.

Reporting

ERP data extraction is gated through the IT team ("whenever we need to pull data from the ERP, we need to ask a guy"). Market data is cross-checked manually across multiple external sources.

New client acquisition

Tenders, trade shows, networking events, word of mouth, and referrals. Cold outreach described as rarely done in B2B wine distribution.

Marketing

Managed by a dedicated marketing team. Content production takes hours per piece; the team is increasingly using AI to accelerate it.

A 15-minute conversation cannot capture how a 550-person wine company actually runs across three separate entities and a full supply chain from wine purchasing to global distribution. What follows is what came through clearly enough to act on.

Section 2

Executive summary

∨

The most visible pain at Argea is also the most structural: three separate ERP systems that do not connect. Every month the finance function spends roughly two days reconciling what each entity recorded. Every day customer service manually keys orders that arrived by WhatsApp, email, or phone. And every production planning cycle runs through Excel macros because the wine movement software and the main ERP have no automated handoff. This is not primarily an AI adoption challenge. It is an integration gap that AI can help bridge today, in pieces, without waiting for the full ERP consolidation that Giulio described as a long-term effort requiring significant investment.

The AI picture is also split. Teams Copilot handles meeting transcription for those using it. But the majority of staff who are using AI for other tasks, including drafting, analysis, and email, are doing so through personal accounts. Data that belongs to Argea, its clients, and its supply chain is moving through systems with no contractual data protection. That gap can be closed in a single decision.

Where Argea sits

Argea is at Level 1: AI tools are in use, but fragmented across personal accounts with no company-wide policy or visibility. Teams Copilot is the one exception. The gap at this stage is not about finding better tools; it is about extending the official program to everyone and making the ERP and planning pain points the first automation targets. The most common mistake at Level 1 is continuing to explore new tools while the existing data-protection exposure accumulates.

5Autonomous AI & agents

4Connected flows across functions

3Process automation

2Team-level workflows

1Individual use · partial official

0No AI adoption

The problems divide into two groups: two specialty workflows where the highest value comes from building a connected layer rather than deploying a self-serve tool, and four quick wins the team can start this week without needing anything from outside.

Top priority · Urgent & important

Order & intake workflow automation

The single highest-value area AdapttoAI can help build at Argea. Orders arrive by WhatsApp, email, and phone and customer service types each one into the ERP manually. The fix is a connected intake layer that captures, classifies, and writes the order directly into the system, and surfaces the pricing and stock data needed for feasibility before anyone picks up a phone. Built right, this also becomes the foundation for the B2B self-service portal already in development.

See Area 1 in Section 5 →

Start this week · Four quick wins

What the team can implement without us

ERP reconciliation with Claude: export data from three ERPs to CSV and let Claude flag the discrepancies.
AI governance baseline: extend the official AI program from a small group to the whole company, with a one-page policy and a signed DPA.
Demand planning starter: use 12 months of historical order data to build the first rolling 90-day forecast.
Marketing content templates: save the prompts the marketing team already uses into a shared Claude Project so every piece starts from a consistent baseline.

See quick wins in Section 4 →

Section 4

Quick wins

∨

QW1 · Cross-ERP reconciliation

Reconcile three ERPs without an IT ticket

"There's a lot of manual reconciliation of the numbers from the different entities."

Claude Team

Why it fits

Three ERPs recording the same costs and revenues in different formats means someone has to do the translation manually every month.
Claude can parse, compare, and flag discrepancies across multiple CSV exports in seconds, producing a structured view of where the numbers disagree and why.
No custom connector or IT involvement needed: the data lives in every ERP as an export.
The finance team keeps control of the final numbers; Claude removes the mechanical comparison step.

How to start this week

Export the previous month's intercompany cost data from each of the three ERPs in CSV or Excel format.
Paste the exports into a Claude Team conversation with a structured prompt asking Claude to flag lines that do not match across all three systems, grouped by entity and account code.
Save the prompt in a Claude Project for monthly reuse.
Run the same process for intercompany invoices in week two.

Complexity

Low

Monthly cost

$20/seat (Claude Team)

Setup time

2 to 4 hours

Time recovered

~2 days/month (team)

Week 1 milestone: first automated reconciliation of intercompany cost data, with discrepancies flagged by line item across all three ERPs, without manual cross-referencing.

QW2 · AI governance

Extend the official AI program to everyone

"The company set it up officially for a small number of employees. When the rest use it officially." [note: voice artifact; the majority use personal accounts]

Claude Team · ChatGPT Business

Why it fits

Data entered into personal AI accounts (free or consumer tiers) is not covered by a data processing agreement. Client specs, ERP exports, financial data, and supply chain details are all potentially in play.
Teams Copilot covers meetings; it does not extend to the personal Claude and ChatGPT accounts the rest of the team uses for drafting, research, and analysis.
Enterprise AI tiers cost $20 to 25 per seat per month and include contractual data-handling commitments, audit logs, and usage visibility.
Half a day of configuration closes a data exposure that is currently open and growing.

How to start this week

Choose a platform: Claude Team ($20/seat/month, 5-seat minimum, 200K context window, strong for document analysis) or ChatGPT Business ($20/seat/month, Code Interpreter included).
Write a one-page acceptable-use policy: what data can and cannot be shared with AI tools, and which tools are on the approved list.
Migrate the 10 to 20 highest-use AI staff in the first week; roll out to the remaining team over 30 days.
Update the company's privacy notice to reflect AI-assisted data processing.

Complexity

Low

Monthly cost

$20 to 25/seat

Setup time

Half a day

Result

Data protection active

Week 1 milestone: 10 to 20 active AI users working through a company account with a signed DPA and a one-page policy; no company data going into personal AI accounts.

QW3 · Demand planning

Build the first rolling 90-day forecast from ERP exports

"We are basically running blind on demand planning. We basically base our forecast on the yearly budget."

Claude Team · NotebookLM

Why it fits

The annual budget is a fixed snapshot; actual order patterns change by month, by client, and by SKU. An AI analysis of the last 12 to 24 months of confirmed orders surfaces the patterns the budget cannot show.
Argea already has a S&OP planning tool in the stack, but Giulio described the team as "running blind" on monthly forecasts despite having it. This starter exercise uses raw order data to surface patterns the current setup is not delivering, and can inform whether the gap is in the tool, the data feeding it, or how it is being used.
This is not a demand planning system. It is a first forecast that makes seasonality and client-velocity patterns visible so that production and procurement can plan ahead rather than react.
The data already exists in the ERP. The tool is Claude or NotebookLM. No new integration is needed to start.

How to start this week

Export 12 months of confirmed orders from the main ERP, broken down by client, SKU, and month.
Upload the export to Claude Team with a prompt asking it to identify the top 20 SKUs by volume, surface seasonal patterns, and flag any clients with declining order frequency.
Save the output as a planning baseline document in a shared drive.
Run the same export and prompt each month to update the baseline and track drift from the annual budget.

Complexity

Low to medium

Monthly cost

$20/seat (Claude Team)

Setup time

3 to 4 hours

Outcome

Monthly rolling forecast

Week 1 milestone: first AI-generated SKU and client pattern analysis from historical ERP data, with at least one seasonal pattern or declining account identified that was not visible in the annual budget.

QW4 · Marketing content

Formalize what the marketing team already does with AI

"It takes hours. But they're increasingly using AI to accelerate that part."

Claude Team

Why it fits

The marketing team is already using AI to produce content but without shared templates or prompts, each person rebuilds the context from scratch each time.
A library of saved prompts in a Claude Team Project cuts production time again and keeps tone consistent across markets and languages.
Global distribution means English, Italian, and potentially other language versions of the same materials. Claude handles all three natively.
This also gives the team a structured starting point for trade show follow-ups, importer briefings, and product launch communications.

How to start this week

Identify the three most repetitive content types: product description, post-trade-show email follow-up, and vintage launch social post.
Write a prompt template for each, specifying Argea's brand voice, target audience, and any mandatory brand elements.
Save all three in a shared Claude Team Project so every marketing team member starts from the same baseline.
Before saving each final prompt, paste it into promptoptimizer.tools (free, no signup) to sharpen the instructions before the template goes live.

Complexity

Low

Monthly cost

$20/seat (Claude Team)

Setup time

2 to 3 hours

Time recovered

1 to 3 hrs/piece

Week 1 milestone: three brand-voice prompt templates saved in a shared Claude Project; first content piece produced from a template in under 20 minutes.

Section 5

Areas worth a closer look

∨

Four areas came out of the call where the right move depends on details we could not get to in 15 minutes: how orders actually flow through each channel, where the approval gates sit today, what the wine movement software exposes via API, and how much of the winemaking knowledge exists in any written form. These are not quick wins. They are the kind of work where what AdapttoAI would build, and how much it would be worth, depends on a closer look at the actual workflow.

Top priority · Urgent & important

Order & intake workflow automation

What we noticed

"When we receive an order from a client, come from WhatsApp, from phone calls, from email. So there's a manual order entry to the ERP. And that takes a lot of time." And separately: "There is something that can be improved maybe in that area" (referring to a client-facing platform to place orders without going through a human).

What this could look like

An intake layer that reads incoming orders from WhatsApp, email, and phone, classifies them, extracts the key fields (client, SKU, quantity, delivery date, personalization requirements), and writes the record directly into the ERP without a customer service agent re-typing it.
A feasibility check built into the same flow: stock levels, production slots, and pricing data from the ERP surface in the intake confirmation before the order is committed, removing the back-and-forth that adds days to response time.
The B2B self-service portal in development becomes one channel in a unified intake layer rather than a parallel system. Orders placed by clients directly feed the same pipeline as orders arriving by WhatsApp or email.

Why this is more complex than it seems

WhatsApp order messages are unstructured. "I want 500 cases of the Barolo by Friday" requires extraction logic that handles variations in how different importers phrase the same request, including different languages.
The ERP integration point matters enormously. If the target ERP has a usable API or webhook, the write-back is straightforward. If orders must go in through a UI, the architecture is different and more brittle.
Customer service currently also performs a feasibility check before confirming. Automating the intake without also surfacing the feasibility data means a human still has to interrupt the process to verify stock and slots.

What we would want to understand better

What percentage of orders come by each channel (WhatsApp versus email versus phone), and does that split vary by client type (retail importer versus HoReCa)?
Which of the three ERPs receives order entries, and does it have an API or webhook capability?
What fields does customer service check before confirming an order, and where does that data currently live in the system?

Connects to QW1 (ERP reconciliation). The reconciliation quick win makes the underlying data more reliable across all three systems. The intake automation makes the data-entry step automatic on the front end. One reduces the cleanup cost each month; the other eliminates the re-entry step from day one of each order cycle.

Important

Production planning integration

What we noticed

"There is a manual process that basically matching the needs, the client orders, together with other softwares including this wine moving software... and then the interface between those two software is done manually, the Excel and macros." And: "there is some cumbersomeness in that passage. It is not flowing seamlessly and automatically from order confirmation to production."

What this could look like

A direct integration between the wine movement software and the main ERP, triggered when an order is confirmed: the system checks wine stock, dry goods availability, and production capacity automatically, without anyone downloading from one system and uploading to another.
The macro logic captured and translated into a maintainable automation, so that when the product mix or ERP field structure changes, the logic does not need to be rebuilt from scratch by whoever currently owns the spreadsheet.
Demand planning outputs from QW3 flowing directly into the production schedule as a structured input, rather than being manually aligned each planning cycle.

Why this is more complex than it seems

The most important question is whether the wine movement software exposes any kind of API or data export format beyond a screen-scraped UI. The answer determines whether this is a two-week integration or a six-month project.
Excel macros built over years accumulate undocumented business logic. The person who built them may not be the person who runs them now. Migrating the logic requires reverse-engineering decisions that were made informally and never written down.
Three ERPs means the production planning output may need to feed multiple downstream systems, not one. Each one is a separate integration point with its own data format.

What we would want to understand better

Does the wine movement software have an API, or does it only expose data through scheduled exports or a UI?
Is there one person who owns the current Excel macros? How much of the business logic exists only in that file?
How often does the production planning handoff run (daily, weekly), and what triggers it?

Important

Approval workflow automation

What we noticed

"When you make a sales offer, either you do it directly or somebody who needs to approve it... Expense management. Also, I think there's somebody who reviews and approves the expenses of employees. And I imagine that there is something there in the invoice processing and order entry."

What this could look like

A structured approval flow where sales offers route automatically to the right approver with the relevant margin, pricing, and client data from Salesforce visible in the request, removing the need for a separate ERP lookup before approving.
Expense approvals with a rules-based routing structure: amounts below a threshold auto-approved or routed to a line manager; amounts above go to a second level. No email chain, no CC list.
Invoice approvals with automatic matching against the relevant purchase order in the ERP: if the invoice matches, it routes for standard sign-off; if it does not, it flags for review with the discrepancy highlighted.

Why this is more complex than it seems

Approval logic is almost always more nuanced than it appears from the outside. Thresholds, exceptions, delegations, and off-cycle approvals are usually undocumented and inconsistently applied until someone has to write them down formally.
Salesforce holds some of the data relevant to sales offer approvals, but if it is not fully utilized (as Giulio noted), the data quality may not yet be sufficient to drive automated routing decisions.
Connecting approval flows to the ERP for invoice matching requires the ERP to have a reliable PO structure. If POs are inconsistently created or filed, the matching logic will generate false positives that create more work, not less.

What we would want to understand better

What does a typical approval chain look like today: how many emails go back and forth before a sales offer or expense is approved?
Are approval thresholds and routing rules currently written down anywhere, or do they exist only in people's judgment?
Is Salesforce the system that holds the pricing and margin data for sales offers, and is that data reliable enough to use as a routing input?

Important

Knowledge system for craft expertise

What we noticed

"There are few areas including winemaking, because it's a very crafty activity... In customer service, probably, have few key people who are the heads of the department, and then these are key."

What this could look like

A structured knowledge capture for the winemaking team: blending decisions, vintage-specific adjustments, supplier quality assessments, and exception handling documented in a format that a new team member can search and consult without calling the person who holds the knowledge.
A custom AI assistant trained on Argea's internal winemaking notes, production records, and supplier history, so that the institutional knowledge is accessible as a question-answerable resource rather than locked in one person's memory or buried in email archives.
A similar capture for customer service leads: key account preferences, common exception requests, escalation patterns, and the informal rules that experienced agents apply automatically but never document.

Why this is more complex than it seems

The largest risk is not technical. The winemakers who hold the knowledge need to be willing to document decisions as they make them, or sit for structured knowledge capture sessions. If that is a cultural or contractual challenge, the project stalls before it starts.
Craft knowledge in winemaking is partly tacit: a blend decision reflects sensory judgment, not just a formula. Capturing the rationale behind a decision is different from capturing the decision itself, and the former is what makes the knowledge useful.
A knowledge base that is not actively maintained becomes outdated. Each new vintage, new supplier, or production line change requires someone to own the update process or the system drifts away from what is actually happening in the cellar.

What we would want to understand better

Is any of the winemaking decision logic currently documented in any form, or is it entirely in people's heads?
Would the key winemakers be willing to participate in a knowledge capture process, and is there a business case they would find compelling?
How often does the customer service key-person dependency actually slow things down, and are there specific account types or request types where it happens most?

AI Assessment for Argea

What we heard

Executive summary

Impact versus effort

Quick wins

Areas worth a closer look

Your next steps