From blind vision to billion-dollar AI: Alberto Rizzoli’s startup journey
with Alberto Rizzoli, V7 Labs
From blind vision to billion-dollar AI: Alberto Rizzoli’s startup journey
Show Notes
Alberto Rizzoli is the co-founder and CEO of V7 Labs, an AI company headquartered in Oxford Circus, London that builds AI agents for automating the administrative and back-office work that keeps people inside spreadsheets and word processors for hours every day. V7 is about 80 people, has roughly 350 customers including KPMG and major private equity and FinTech firms, raised $33 million in 2022 during one of the worst fundraising environments in recent memory, and generates approximately 70% of its revenue from the US market despite being founded as a European company.
Alberto's path to V7 runs through a previous company called I Poly - a computer vision app for the blind and visually impaired, built for the iPhone 5 in 2015 and recognized with Italy's National Genteel Prize for Science and Innovation. That experience with AI for good, and the hard lesson about market size, shaped everything about how he builds today. V7's current flagship product is V7 Go, an AI agents platform that ingests complex unstructured documents and turns them into structured, actionable outputs - with a hallucination-elimination layer called Visual Grounding that traces every AI answer back to its exact location in the source document.
What V7 Go Does: Eliminating the Back Office
The core product is a platform where companies build AI agents that process documents - leases, investment memos, tax claims, rent rolls, financial models - and return structured outputs: spreadsheets, IC reports, CSV files, status summaries. The agents work through multi-step reasoning rather than a single prompt, which dramatically improves accuracy. A financial model analysis agent might first extract all values, then check them against the company's guidelines, then determine required next steps and who needs to be notified. Each step is verifiable.
A meta-agent called Concierge sits above the individual agents and routes incoming work to the right one automatically. Users can submit tasks via email, drag-and-drop, or Slack mention - Concierge reads the request and delegates to the appropriate agent. Alberto describes this as a chief of staff for every employee in the organization, with a network of specialized agents beneath them handling the work that humans should not be spending their days on.
The critical differentiator is Visual Grounding - V7's anti-hallucination layer. Every document ingested is scanned with computer vision to map where all text lives. When an agent produces an answer, it must locate the exact position in the source document where that answer is grounded. A user reviewing 12 extracted data points can click any one of them and see immediately where in the document it came from. This matters particularly in high-stakes workflows like term sheet evaluation or tax claim processing, where human review of AI output is still required and accuracy is non-negotiable.
The Question Every AI Founder Should Ask Every Six Months
Alberto frames the V7 pivot from Darwin (data labeling) to Go (AI agents) as the application of a question he now asks formally every six months: what is the product that, if it came out today, would kill my company? The answer is what you should be building. He used this framework to justify sacrificing a double-digit percentage of V7's revenue to create V7 Go - because the arrival of GPT-3 made it clear that LLMs would replace the custom document-understanding models his customers were spending months building. Rather than wait for a competitor to take that market, V7 built the replacement itself.
He is explicit about the cost of this kind of internal rebuilding: it is a startup within a startup. A 50- or 60-person company does not have the same ungodly energy as a five-person one. People do not work weekends at the same rate. The culture of sacrifice that characterized the early days requires deliberate reinjection. His metric for whether that reinjection is needed: if you are only adding incremental features to your existing product rather than building something that could disrupt it, you are at risk.
Raising $33M in 2022: Fundraising as Sales
2022 was one of the hardest years in recent memory for venture fundraising. 2021 had been euphoric; 2023 bounced back. 2022 was the trough. Alberto raised $33M anyway. His explanation is honest about what actually works: fundraising is sales. He maintained a rigorous investor CRM throughout the process, tracked every contact through a Sankey diagram of rejections and offers, created pipeline systematically, and treated investor outreach with the same discipline he would apply to a sales motion. He allocates 90 minutes every Friday to investor CRM and communications even when not actively fundraising - a small but consistent investment in relationships that compounds.
The deck worked because it was opinionated. Not “another workflow automation tool” but a specific, clearly differentiated product with a unique angle, customer proof of switching from competitors, and strong retention data. The investors who said yes believed that something was coming around the corner that would prevent an AI winter - a bet that turned out to be correct. The investors who said no were right that 2022 was hard; they were wrong about the trajectory.
Treat the US as Your Market Regardless of Where You Are
V7 was founded to be a strong European AI company. Alberto now tells founders to treat the US as their market no matter where they are headquartered. The evidence is in his own numbers: 70% of V7's revenue comes from the US, and the average US sales cycle is roughly half the length of a comparable European deal. European companies move slower, have smaller AI project budgets, and even large enterprises that are technically UK-headquartered often execute their AI purchasing decisions through US teams.
This does not require being physically in the US from day one. It does require orienting your go-to-market messaging, pricing, and sales process toward an American enterprise buyer - and accepting some late Zoom calls in the process. The tradeoff is unambiguously worth it for a software company.
Where the AI Hype Cycle Actually Stands
Alberto's view on the current moment is measured. We are at the beginning of AI's economic impact, not the peak. Most hospitals and governments are still not running AI in production. The prototyping is happening; the deployment is not. The performance plateau being observed in new frontier models is not alarming - even with today's transformer-based capabilities, the gains available in white-collar work and coding remain largely unrealized.
The genuine risk he identifies for AI startups is the “why wouldn't the model provider just build this?” question - the modern version of “why wouldn't Google crush you?” His answer: focus on the top 5% of quality requirements in your domain. The majority of consumers will accept “good enough” from a native model capability (image generation from GPT-4o, file retrieval from Claude). The 5% who need genuinely excellent, domain-specific AI - processing complex financial documents, evaluating term sheets, managing tax claims at scale - will not. Build for that 5%, and model commoditization is not a threat to your business.
He also names the hidden risk inside current AI success stories: churn. Many companies showing early AI traction are seeing users fail to deploy anything in production and then churn. The only reliable solution is human-assisted customer success and solutions engineering - which is unglamorous and expensive, but currently unavoidable for serious AI work deployed inside organizations.
Tools & Resources Mentioned
- V7 Labs / V7 Go - Alberto's AI agents platform for document processing and back-office automation; v7labs.com
- V7 Darwin - V7's original product; data labeling platform for AI researchers, still active
- Visual Grounding - V7's proprietary anti-hallucination system; every AI answer is pinpointed to its exact location in the source document
- Cursor - AI coding tool Alberto cites as a meaningful productivity increase for V7's engineering team
- Granola - meeting notes tool; Alberto describes it as a “potato peeler” - a single-purpose tool so good you would never use anything else for its job
- Framer - V7 switched from Webflow to Framer for their marketing site; cited as an example of tools reducing marketing team overhead
- Meditations by Marcus Aurelius - personal diary of a Roman emperor, recommended as an unusually unfiltered historical document about the emotional experience of leadership
- I Poly - Alberto's prior company; AI accessibility app for the blind, winner of Italy's National Genteel Prize for Science and Innovation
Frameworks
What Product Would Kill My Company?
Ask every six months: if a competitor released the most threatening possible product today, what would it be? Then build it yourself before someone else does. V7 used this to pivot from data labeling to AI agents, sacrificing existing revenue to avoid being disrupted. The question forces honest assessment of whether you are building incrementally or building what the market actually needs next.
Fundraising as Sales: Build the Pipeline
Fundraising is a sales process and should be managed like one. Maintain a CRM, track every investor contact, create pipeline systematically, block calendar time for outreach even when not actively raising (Alberto does 90 minutes every Friday). VCs want relationships, but any great salesperson will tell you that relationships alone do not close deals - pipeline does. Be opinionated in the deck: not another tool, but a specific angle backed by customer proof.
Build for the Top 5%
As model providers embed more capabilities natively (image generation, file retrieval, voice), the majority of users will accept ‘good enough.’ The startup opportunity is in the 5% of users who need genuinely excellent, domain-specific AI - complex financial documents, medical records, legal analysis. Build for that group and commoditization by foundation models is not a business threat.
Startup Within a Startup
Major AI pivots often require rebuilding from near scratch - a startup within the startup. The difficulty is that a 50-person company does not have the energy of a 5-person one. Founders need to deliberately reinjection startup-level urgency into the rebuild. A good diagnostic: if you are only adding incremental features rather than building something disruptive, you may already be at risk of being disrupted.
This Too Shall Pass
Alberto's stoic operating principle for founders, drawn from the King Solomon ring story: a ring is requested that makes the king sad when happy and happy when sad. The ring maker returns with one inscribed ‘this too shall pass.’ In a startup, every peak will end and every trough will end. Calibrating expectations to that reality - rather than riding emotional extremes - is what allows founders to persist through both.
FAQ
What does V7 Go do and who is it for?
V7 Go is an AI agents platform for companies that process large volumes of complex documents - financial models, leases, tax claims, investment memos, insurance documents. The agents ingest unstructured documents and return structured outputs (spreadsheets, reports, CSVs) by working through multi-step reasoning rather than single prompts. A meta-agent called Concierge routes incoming tasks via email, drag-and-drop, or Slack to the correct specialized agent. Primary customers include KPMG, private equity firms, banks, and FinTech companies.
What is Visual Grounding and why does it matter?
Visual Grounding is V7's anti-hallucination system. Every document processed by V7 is scanned with computer vision to map the location of all text. When an agent produces an answer, it is required to locate the exact position in the source document where that answer is grounded. Users reviewing AI output can click any data point and see immediately where it was sourced. Standard RAG can still hallucinate an answer while providing a page or document reference; Visual Grounding eliminates that ambiguity by requiring pixel-level precision.
How did V7 raise $33M in 2022 when VC markets were largely frozen?
Strong product traction, low churn, and clear customer differentiation were the foundation. The deck was opinionated - a specific angle on a specific problem, not a generic workflow automation tool - backed by stories of customers leaving competitors to switch to V7. Alberto treated fundraising as a systematic sales process: investor CRM, pipeline creation, consistent outreach cadence. Some major funds had stopped deploying entirely; he tracked rejections and offers on a Sankey diagram and kept moving. The raise required volume, discipline, and a thesis that AI was not heading into another winter - which the investors who committed were willing to bet on.
Why does Alberto recommend treating the US as your market regardless of headquarters?
V7 was founded to build a strong European AI company. It generates 70% of its revenue in the US. US sales cycles are roughly half the length of equivalent European deals. US enterprise buyers have larger AI project budgets and make faster decisions. Even technically European-headquartered companies often run their AI purchasing through US teams. For a software company, the US market is simply more efficient to sell into - and the friction of later time zone Zoom calls is a very small price for that efficiency.
What is the AI churn problem Alberto describes?
Many AI companies show strong early traction - signups, pilots, initial usage - but then see high churn as customers fail to successfully deploy AI in production and stop using the tool. Getting AI from proof-of-concept to production use inside an organization requires hands-on customer support: solutions engineering, onboarding, and genuine human assistance. Even as AI UX improves rapidly, this human layer remains necessary for serious enterprise deployments. AI companies that are not investing in customer success are hiding a churn problem.
What AI company would Alberto build today to hit a million dollars fastest?
He cites the YC ‘service as software’ framing: find a service business that is still being done manually, and replace it using existing AI tools with very little human involvement. His example is R&D tax claim filing in the UK - a mandatory, heavily manual annual process for most companies, where an AI-powered service could deliver the outcome end-to-end with minimal overhead. The opportunity is that almost everyone is building SaaS AI tools, leaving the service-delivery layer underserved.
What did Alberto learn from building I Poly that shaped V7?
I Poly proved the technology and the emotional appeal of AI for good - it won a national science prize and had genuine viral resonance. What it failed was the venture market test: accessibility is not a market that can generate the compounding revenue growth a VC-backed company needs to deliver returns. The lesson: the product can be excellent and the mission can be important, and it can still be the wrong market for the entity you are building. Market size and market growth rate are independent of product quality, and both must pass before committing to a venture-scale build.