Roles I'm a fit for
- Forward Deployed Engineer
- AI Solution Consultant / AI Specialist
- Presales Solution Architect / Solution Engineer
- Customer Solutions Architect
- Delivery or Engagement Manager
- Enterprise Architect — CRM & Data
Forward Deployed Engineer · AI in production · Bergamo, Milan & Rome
Hi, I'm
Flavio
Macciocchi.
AI Solution Architect and Technical Manager, forward deployed. I design enterprise AI systems and run the delivery that puts them in production — on the client's site, on the client's stack, with the before-and-after numbers to show for it.
Associate Technical Manager, ten years on enterprise projects. I design the architecture, lead the delivery team, and present the result to the people who paid for it. Right now that work is large language models and customer data, on the Salesforce platform and outside it.
- 10+Years on enterprise projects
- 100+Professionals trained
- 6×Salesforce certifications
- 30+AI and data use cases assessed
6× Salesforce Certified
Bergamo · Milan · Rome
AI live in production
Role fit
The problems I get called for
Industries I have delivered in
Retail, fashion, food & beverage, automotive, media and services. The recurring problem is the same everywhere: a customer record split across five systems, a service process nobody has ever timed, and an ERP that has to stay exactly where it is. In automotive that shows up as dealer networks and after-sales, where the customer belongs to the brand and to the dealer at the same time and the data model has to survive that.
- Retail
- Fashion
- Food & Beverage
- Automotive
- Media & Telco
- Services
How I work
Based between Bergamo and Milan, on-site or hybrid, and available for engagements in Rome. A normal week contains a steering committee with the C-level and a working session with the delivery team, and translating between those two rooms is most of the job.
The fastest way to judge whether I fit is the case study below.
Forward Deployed Engineer
The engineer who moves in with the client
The job I have been doing for ten years, before I had the name for it
A Forward Deployed Engineer is a software engineer who works from inside the client's organisation: installing, adapting and running complex technology in an environment nobody designed for it. Not a demo and not a lab. Their systems, their data, their security team, their deadline.
Every project on this page was delivered that way — on the client's site, against the stack they already had, with my name on the result. The three roles below are not a wish list. They are what a normal week already looks like.
- On-site or hybrid · Bergamo, Milan, Rome
- 4–6 concurrent projects, €500K–€2M each
- Teams of 5 to 15 people
- 92% on-time delivery · budget variance under 5%
- Vendor-neutral: OpenAI, Claude, Gemini, Agentforce
Three jobs, one person
Software engineer
I write the code that has to run
OpenAI agents, integrations against ERP, e-commerce, iPaaS and REST/SOAP APIs, PL/SQL in the data layer, B2C Commerce storefronts, portals built end to end, and NLP research on BabelNet at Sapienza. A client lets me near the architecture because I have been the one called when it broke.
Food retail: agents that read a case, pull the data and draft the reply in 10 to 20 seconds.
Consultant
I sit in the room where the problem actually is
The brief a client writes down is rarely the problem they have. Finding the real one takes time on their floor, watching a process nobody has ever timed, and enough trust to be told what never reaches a requirements document. Then saying the same thing upstairs, in language a C-level can sign.
Infocert: 30+ use cases surfaced, scored and sequenced into a roadmap. CSAT 4.5 out of 5 across the portfolio.
Product manager
I decide what actually gets built
Between what the client wants and what the product does there is a gap, and someone has to choose what gets built across it, in what order, and what gets refused. The pass or fail criteria are agreed before the POC starts, so the decision at the end is arithmetic rather than opinion.
Baseline first: 18 minutes per case, measured before a single model was chosen.
What that looks like on site
-
Connect the systems
Wire the new platform into what is already there: ERP, e-commerce, legacy databases, iPaaS, REST and SOAP APIs, and identity resolution so one customer is one profile. Integration is where the project is actually won or lost, and it is never the part in the slide deck.
-
Clear the blockers
Security review, data governance, a DPO with questions, an IT team that has watched vendors come and go. Dedicated Data Spaces, explicit rules on what an AI agent may read, and a named person signing whatever reaches the customer. Most AI projects die here, so this is the part I design first.
-
Hand over the keys
As a Salesforce Authorized Trainer at Italy's only certified training provider, I build enablement into the delivery instead of bolting it on at handover: 100+ professionals trained and certified, on a format that transfers to whatever capability comes next. An engineer the account cannot run without has not finished the job.
-
Carry the field back
What I learn on site goes back to the people who build the product and the people who sell it: which gap costs us deals, which request is one client and which is the market. In practice that is Design Authority on architecture, plus pre-sales and effort estimation with a 68% win rate.
The short version: I am the person you send when the technology is sound and the deployment is the hard part. Vendor-neutral by habit — OpenAI, Claude, Gemini or Agentforce, whichever survives the client's constraints.
Artificial Intelligence
AI systems in production
Process automation
Manual, repetitive flows replaced by AI agents that pull the data, draft the reply and route the case. The team moves to validation and exception handling.
Intelligent customer experience
Unified customer profiles built on Data Cloud, with AI-driven journeys across email, WhatsApp, RCS and push. Contact rules live on the profile instead of being rebuilt channel by channel.
Data governance by design
Dedicated Data Spaces, separation between business lines, and explicit rules on what an AI agent is allowed to read. Governance designed up front costs a fraction of governance added after go-live.
How I run an AI project
-
Put a number on the current process
Measure the baseline before choosing any model: minutes per case, volumes, error rate, cost. Without a baseline there is no ROI to report, only impressions.
-
Design the architecture and the guardrails
Which model, which integration pattern, where the data sits, who can read it, what happens when the model is wrong. Settled on paper, with IT and security at the table.
-
Prove it on a real process
A POC on production-like data, with human-in-the-loop validation and quality measured against the baseline, and the pass or fail criteria agreed before it starts.
-
Industrialise and hand over
Scale-up, training and clear ownership. As a Salesforce Authorized Trainer I plan the enablement into the delivery rather than bolting it on at handover.
Career
My experience
Lutech Group · TenEnigen
AI & Innovation Lead · Program Manager
2024 — Present · Milan
End-to-end lead on enterprise AI solutions, vendor-neutral by design: OpenAI GPT, Agentforce, Claude or Gemini, whichever fits. Process discovery, agent design, integration with the core systems, human validation and go-live.
Lutech Group · TenEnigen
Associate Technical Manager
2022 — Present · Milan
Technical and delivery lead on enterprise Salesforce and AI programmes. I lead cross-functional delivery teams and hold Design Authority on architecture and integration, with technical support on new business.
Lutech Group
Manager — Digital Experience
2022 — Present
Consulting engagements across Marketing Automation, Data Cloud and AI. Single point of contact for delivery and escalation with C-level stakeholders, from the roadmap to the numbers reviewed in the steering committee.
TenEnigen · Cloud Computing Consulting
Senior Salesforce Consultant
2018 — 2022 · Milan
Marketing Cloud, Data Cloud and CRM solutions for enterprise clients in retail, fashion and food & beverage. Technical lead for configuration, integration with ERP and legacy systems, and client training.
Lutech · TenEnigen
Salesforce Authorized Trainer
2019 — Present
Over 100 professionals trained on Marketing Cloud, Sales Cloud, Data Cloud and Agentforce for the first and only Salesforce Authorized Training Provider in Italy. Teaching a platform to people who will then sit a certification exam is an efficient way to find the gaps in your own knowledge.
Expertise
Skills & Salesforce certifications
AI & language models
- OpenAI / GPT
- Agent design
- Prompt Engineering
- Human-in-the-Loop
- Salesforce Agentforce
- AI use-case discovery
Architecture & integration
- Solution Architect
- Solution design
- Design Authority
- REST / SOAP API
- iPaaS & middleware
- ERP & legacy integration
- Data governance & segregation
Cloud & platforms
- Google Cloud
- Heroku
- PaaS & SaaS architectures
- Salesforce Platform
- Multi-cloud delivery
Data & CDP
- Salesforce Data Cloud
- Data Spaces
- Identity Resolution
- CDP Architecture
- Segmentation
- Data Analytics
Delivery & leadership
- Team leadership
- Delivery & portfolio management
- Effort estimation
- Pre-sales & proposals
- Forward deployed delivery
- C-level stakeholder management
- Training & change management
Marketing stack & channels
- Salesforce Marketing Cloud
- MC Next
- Account Engagement
- Journey Builder
- WhatsApp Business
- RCS
- SMS & Push
Development
- Development
- PL/SQL
- Oracle Database
- SQL
- AMPscript
- JavaScript
- HTML5 / CSS
- Drupal CMS
- Web portals & apps
- Linux
Salesforce certifications
- Marketing Cloud Administrator
Salesforce - Marketing Cloud Consultant
Salesforce - MC Email Specialist
Salesforce - Salesforce Administrator
Salesforce - Data Cloud Consultant
Salesforce · 6× Certified Trailblazer
Beyond the platform
Salesforce is where much of my delivery happened, not the limit of what I design. The AI work runs on OpenAI models called through APIs, the platform work spans Google Cloud and Heroku, the data work is CDP architecture that applies to any stack, and the integrations touch ERP, legacy systems and middleware. A good part of my delivery history has nothing to do with Salesforce at all.
Portfolio
Selected projects
Enterprise client · Food Retail
From 18 minutes to 3 per case, with a person approving every reply
The problem
Customer care handling every request by hand: read the message, find the data across systems, write the answer. Eighteen minutes per case, and volumes growing faster than the team could.
What I designed
OpenAI agents that read the request, extract the relevant data and draft the response in 10 to 20 seconds. Salesforce keeps the case lifecycle. Nothing reaches the customer without a person approving it.
The outcome
Three minutes per case, 85% less effort, six times the throughput, around 400 hours a month freed per team. No new headcount: the same people now handle supervision and exceptions.
Happy to walk through the parts that are less flattering: how we handled wrong outputs, what we measured to trust the quality, and what I would design differently now.
Tinexta · TIH
Data Cloud & MC Next setup
End-to-end configuration of a dedicated Data Space and deployment of Marketing Cloud Next with Data Kits, Data Streams and Identity Resolution on an enterprise org.
Infocert
30+ Data Cloud use cases
Feasibility study across more than 30 use cases, scored and sequenced into an implementation roadmap, with a dedicated Data Space architecture for segregation and governance.
Sky
Marketing Cloud integration
Implementation and integration of Marketing Cloud into the Sky ecosystem, on a subscriber base of several million. Multichannel campaigns and communication flows rebuilt so the contact history belongs to the customer rather than to each channel.
Enterprise clients
Contact strategy advisory
Multivariate test design and communication optimisation across channels, content, offers and timing for retail, fashion and food & beverage clients. Every change went live on the back of a test result.
Lutech · TenEnigen
Salesforce certified training
Certified training programmes for Italian enterprise clients: over 100 professionals across Marketing Cloud, Sales Cloud, Data Cloud and Agentforce, delivered by Italy's only Salesforce Authorized Training Provider.
Background
Where the engineering comes from
Before the architecture diagrams there were ten years of writing code that had to run on Monday morning. The list is in reverse order, so the further down you read, the closer you get to the part that still shapes how I design.
AI and LLM project leadership
Enterprise AI systems built and run in production with OpenAI GPT, Salesforce Agentforce and Data Cloud
2024 — Present
Salesforce Authorized Trainer
Lutech TenEnigen, the first Salesforce Authorized Training Provider in Italy
2019 — Present
6× Salesforce Certified Trailblazer
MC Administrator · MC Consultant · MC Email Specialist · Administrator · Data Cloud Consultant
2018 — 2025
Salesforce B2C Commerce — Developer
Storefront development on cartridges and templates: product catalogue, checkout flow, and the order and stock integrations behind them. The first time I had to answer for a page that took money.
2017 — 2018
Oracle Database — SQL and PL/SQL Developer
PL/SQL stored procedures, packages, cursors and triggers: business logic living in the data layer, nightly load jobs, and query tuning read off execution plans rather than guessed at. Where I learned that the data model decides what an application can ever do.
2016 — 2017
Freelance developer — apps and web portals
Web applications and portals built end to end for direct clients: front end, back end, hosting and content management on Drupal. Scoping, quoting and then supporting my own work taught me more about estimation than any methodology since.
2013 — 2016
BSc in Computer Science — Theoretical Informatics
Sapienza University of Rome
2009 — 2012
Frequently asked
Questions I get in every interview
What does an AI Solution Architect actually do?
Two jobs at once. Before the project: turn a business problem into an architecture: which model, which data, which integration pattern, what happens when the model is wrong. During the project: keep the delivery team pointed at that design, and change the design when reality disagrees with it. The title matters less than being accountable for both the drawing and the result.
How do you measure the ROI of an AI project?
Against a baseline measured before the model exists: minutes per case, volumes, error rate, cost. On the case management project the baseline was 18 minutes per case; after go-live it was 3. Every other number (85% less effort, six times the throughput, 400 hours a month freed) comes from those two. Without a baseline you are not measuring ROI, you are collecting impressions.
What is human-in-the-loop, and why does it matter?
The model drafts, a person approves. It costs a few seconds per case and it buys the one thing that makes an AI project deployable inside a company with a compliance function: a named human accountable for whatever reaches the customer. It also produces a steady stream of corrections, which is the cheapest quality metric you will ever get.
Your background is Salesforce. Does that experience transfer elsewhere?
The AI work is API calls to OpenAI models, not a platform feature. The data work is CDP architecture: identity resolution, segmentation and segregation apply to any stack. I have delivered on Google Cloud and Heroku, designed PaaS and SaaS architectures, and integrated ERP and legacy systems that had nothing to do with Salesforce. The platform is where much of the delivery happened, not the boundary of what I design.
Which clients have you worked with, and where are you based?
Enterprise clients in retail, fashion, food & beverage, automotive, media and services — Sky, Tinexta and Infocert among the ones I can name. I am based between Bergamo and Milan, work on-site or hybrid across Lombardy, and take on engagements in Rome, on projects that usually combine marketing technology, customer data and AI.
What is Salesforce Data Cloud used for?
It unifies customer records scattered across systems into a single profile through identity resolution, then makes that profile available to marketing, service and AI agents in near real time. In practice most of the work sits upstream: deciding which sources can be trusted, how business lines stay segregated in dedicated Data Spaces, and which fields an AI agent is allowed to read.
What is the difference between a Forward Deployed Engineer and a Solution Architect?
Distance from the client, and how much of it you build yourself. A Solution Architect can design the system, hand over the drawing and move to the next account. A Forward Deployed Engineer lives with the consequences: the same person designs it, writes the integration, argues it through the security review, and is still there the month after go-live when the real numbers arrive. I have worked both ways. The second is where I am useful, and it is the only version where you find out whether the design was right.
Let's talk about what you need built
Hiring for an AI, architecture or delivery role in Bergamo, Milan or Rome? Send me the job description and I will tell you where I fit and where I don't.
Bergamo · Milan · Rome · Usually replies within a day