From Big Four slideware to small, focused AI projects: how Bubble wants to replace bloated consulting with fast, measurable automation for finance teams.
When I left Deloitte after five years in audit and financial consulting, my partner was leaving UBS where he had managed equity funds for six years. We could have continued our well-established careers. Him, climbing the asset management ladder to reach the group's executive committee. Me, aiming for partner status then shareholder-partner at Deloitte. But we made a different choice: launching our own company, Bubble Invest.
Why leave? Not out of rejection of the sector or the skills we acquired. But because we saw too many projects get bogged down in endless PowerPoints, POCs (proof of concepts) that never deploy, and budgets of €200,000 for six-month missions whose final deliverable fits on three slides.
The bullshit problem isn't new, but AI has made it worse. In the Big Four as in large banks, AI has become a magic word that justifies everything — like before it the words "digital transformation," "automation," "connected tools" — astronomical billing, unreasonable timelines, bloated teams. Except that at the Big Four, nobody really knows how AI works, or worse everyone denies using it secretly and pretends to have pseudo expertise. Directors sell dreams to clients while their juniors/seniors toil like ants, between two or three unacknowledged ChatGPT searches. And in between, nothing materializes, clients wait, processes drag, deadlines extend, and it takes more and more consultants for even fewer results.
At Deloitte and KPMG, I saw entire departments not apply the advice they give (or could give) to their own clients — a technophobia and unwavering will to justify at all costs the necessity of all tasks accomplished, even the most repetitive and within reach of any algorithm. Private banking isn't far behind, endless audits and advisory services, without any reliable and automatic decision support process ever being put in place.
We decided to do things differently with Bubble Invest. No traditional "consulting" or "AI consulting." Concrete, measurable, explainable implementations. No drawn-out missions. Pragmatic work in 2-4 months maximum. No prohibitive rates. A freelance/remote model accessible to SMEs and small asset management firms. And most importantly: we act with you, we don't give you beautiful presentations before disappearing.
State of the Market
To understand what we wanted to change, we analyzed the French "AI finance consulting" market for several weeks. What we discovered confirms our intuition: it's a sector in full ferment, with lots of noise and few tangible results.
Who's selling "AI consulting" in France today?
The landscape is fairly clear:
1. The Big Four (Deloitte, PwC, KPMG, EY)
They've all launched AI offerings for finance. Deloitte has deployed Zora AI (in partnership with Nvidia) to automate invoice processing and trend analysis. EY claims its AI now assists 80,000 tax professionals and handles 3 million compliance cases per year[1]. But these tools are reserved for their own teams or their largest clients. For SMEs and mid-sized asset managers? Inaccessible.
Internally? AI tools were long ignored — ChatGPT was blocked on their systems to protect client data, but paradoxically, Azure OpenAI GPT-4o was authorized. Result: in October 2024, Deloitte Australia got caught red-handed with a AU$440,000 government report riddled with AI hallucinations — fabricated citations, fictitious federal court judgments, non-existent academic references. Discovered by a University of Sydney researcher, the scandal forced Deloitte to partially refund the Australian government. A senator compared these errors to "what a first-year university student would get in deep trouble for." The ultimate hypocrisy: banning ChatGPT internally while using AI without disclosing it to clients.
2. Pure-play AI companies
In France, a few companies have specialized in specific niches. Akur8 (founded in 2019, $120M raised in Series C) optimizes insurance pricing through machine learning[2]. Shift Technology automates fraud detection for insurers[3]. Zelros and Golem.ai work on NLP for insurance. These are software publishers, not consulting firms. Their model: sell a SaaS license, not accompany a transformation.
3. Specialized finance firms adding AI
Many audit or regulatory consulting boutiques now offer "AI services" without really having the technical skills. Result: more studies, diagnostics, roadmaps... but nothing concrete.
4. Freelancers and no-code agencies
An ecosystem is emerging around tools like Notion, Make, Zapier, Bubble. According to a study, 70% of companies will use no-code/low-code technologies in 2025[4]. These players work fast, charge less, but often lack finance expertise.
The Disturbing Numbers
Our research highlighted alarming statistics on AI projects in businesses:
- Between 70% and 85% of AI projects don't achieve their business objectives or are abandoned before deployment, according to converging studies from Gartner, McKinsey, BCG, and RAND[5].
- Gartner specifies that nearly 80% of AI projects don't produce the expected results in terms of business value. This failure rate is almost double that of traditional IT projects[6].
- McKinsey points to data problems: 70% of AI initiatives don't succeed primarily due to poor quality or poorly structured data[7].
- In France, the situation is even more paradoxical: 91% of decision-makers consider AI important or priority, 44% have launched projects in 2025, but only 26% of French companies have concretely deployed AI[8].
- Overall, only 11% of global companies (and 26% of large ones) manage to derive real value from their AI[9].
What we learned analyzing 30+ players
Three findings stand out:
1. The eternal POC syndrome
Firms sell "pilot phases" that drag on without ever going into production. Why? Because they bill by time spent, not by results. The longer the project lasts, the better for their revenue.
2. Technical incompetence masked by jargon
Many consultants talk "machine learning," "NLP," "computer vision" without ever having coded a single line. They subcontract the tech part... which doesn't understand business issues. Result: tools that don't meet the need.
3. Opaque pricing
The Big Four charge between €150,000 and €300,000 for a 6-month AI transformation mission. But what does this envelope really contain? Teams of 3-5 consultants, half of whom do documentary research. An 80-page report, 60 of which are filler. And often, zero lines of code in production.
International Benchmark: What Works Elsewhere
We looked at what's happening in the United Kingdom, Switzerland, and the United States. Three major differences:
1. UK and US: The emergence of value-based pricing
There, consulting firms are starting to abandon the daily rate model (TJM) in favor of packages based on value created (improved conversion rate, reduced costs, generated revenue). According to several analysts, the traditional TJM model could become obsolete within 2-3 years[10].
2. Switzerland: The agility of short missions
Swiss asset management firms, accustomed to efficiency, favor 1 to 3-month missions with concrete deliverables. No 18-month strategy. Rapid deployment, test-and-learn.
3. Everywhere: The rise of specialized "AI agents"
CGI Switzerland has developed DeepContext, a technology accelerator for specialized AI agents in finance. In the US, fintechs favor agentic workflows (autonomous tasks delegated to AI) rather than "big bang" transformations[11].
In France, we're behind. We remain stuck in a logic of large projects, heavy governance, monthly steering committees. Meanwhile, others are deploying.
What We Really Do (Our Difference)
At Bubble Invest, we don't sell PowerPoint. We don't do "strategic AI diagnostics" billed at €50,000. We implement. With or without AI? Concretely, it's not the question. It's like saying "are you going to use the Internet?" In 2025, it's no longer a question to ask.
Our approach: not traditional consulting, measurable and concrete implementation
We start from a specific business problem. Not a technological fantasy. Examples of typical projects we handle:
- Monthly reporting automation for an asset manager that spends 15 hours per month compiling Excel data from three different sources. We create a workflow via Make or n8n that automatically retrieves data, normalizes it, and generates a ready-to-send PDF. Result: 15 hours saved, zero input errors.
- News monitoring for thematic funds: a manager wants to follow ESG news from 50 portfolio companies. We set up an AI agent (via Claude API or Gemini) that scrapes relevant sources, summarizes articles, and sends a weekly digest. Monthly cost: €20-30 in API. Time saved: 5-7 hours per week.
- Allocation strategy backtesting: we develop simple tools in Python or JavaScript to test risk-parity, momentum, or multi-asset strategies. No complex platform. Clear, documented scripts that the client can modify themselves.
Our tech stack: intelligent no-code/low-code
We acknowledge we don't do advanced ML engineering. We don't develop proprietary models. We compose with the best available tools:
- Notion: to structure data, create relational databases, centralize information
- Claude Code and Gemini CLI/Codex: to generate code quickly, automate repetitive tasks, create workflows; to connect APIs and automate processes. We code in JavaScript/Node.js or Python, but always with code we understand at least minimally.
This approach allows us to deliver in 2 to 4 months maximum what others take 12 months to do. Why? Because we don't reinvent the wheel. We use what already exists, assemble it intelligently, adapt it to the specific need.
Concrete examples
- Bank reconciliation automation for a fintech SME: previously, an accountant spent 2 days per month reconciling Stripe transactions and accounting entries. We created a script that does this automatically, with a 98% matching rate. Implementation time: 3 weeks. Cost: €8,000. ROI: recovered in 4 months.
- Performance tracking dashboard for an asset manager with €800M AUM: we create an automated and customizable client reporting tool. We built a simple webapp in JavaScript + Chart.js that connects to their back-office, retrieves data, and generates custom PDFs. Time: 2 months. Budget: €22,000 (vs. €80,000 requested by a software publisher).
Our model: freelance, remote, efficient
We don't do on-site for the sake of being on-site. We don't believe in two-hour meetings every week to "check in." Our approach is consistent with our automation vision: we come on-site only if necessary, not to blow hot air or give you a speech.
We work remotely, in agile mode. 2-week sprints, incremental delivery, rapid feedback. No big frozen specifications. We iterate, test, adjust.
And we offer post-implementation maintenance. Because an automated tool evolves. APIs change, needs evolve. We don't leave you alone after delivery. We remain available, in light monthly subscription mode for all bugs and code maintenance (€500-1000/month depending on needs).
Who It's For (And Who It's Not For)
Our ideal targets
We don't work for everyone. Our typical clients:
- SMEs with 20 to 250 employees with manual financial processes to automate (reporting, consolidation, cash flow tracking, invoicing)
- Asset management firms <€2Bn AUM wanting to improve operational efficiency without hiring three back-office people
- Organizations wanting pragmatic and fast: not 18-month digital transformation, but a structuring project delivered in 2-4 months
What we DON'T do
Let's be clear about what we don't do. This will avoid misunderstandings:
- No long digital transformation: if you're looking for a firm to accompany you for 12-18 months with governance, steering committees, change management, look elsewhere. It's not our business.
- No experimental POCs: we don't do R&D. If you want to test cutting-edge technology without knowing if it will be useful, call a research lab, not us.
- No "AI revolution" promises: we're not here to sell you dreams. We automate specific tasks. We improve efficiency. But we don't claim to revolutionize your business model overnight.
- No advanced ML engineering: if you need proprietary machine learning models trained on your data, we don't have the skills. We work with existing LLMs (GPT, Claude, Gemini) and standard APIs. Period.
Realistic budget
Our projects range from €15,000 to €30,000 for a structuring mission over 2-4 months. That's 5 to 10 times cheaper than the Big Four for a concrete and deployed result.
- Specifically:
- Diagnosis + roadmap: €3,000 - 5,000 (1-2 weeks)
- Simple implementation (1 automated workflow): €8,000 - 12,000 (3-4 weeks)
- Complete project (multiple workflows + dashboard + maintenance): €20,000 - 30,000 (2-3 months)
Compare with the €150,000 to €300,000 charged by the Big Four for 6 months of mission... of which often 80% PowerPoint and 20% partial implementation.
How We Work (Total Transparency)
Our methodology: Diagnosis → Roadmap → Implementation → (Optional Maintenance)
1. Diagnosis phase (1-2 weeks)
We spend time understanding your current processes. No PowerPoint questionnaire. We look at your Excel files, your emails, your tools. We identify repetitive tasks, sources of error, bottlenecks. We quantify wasted time.
2. Prioritized roadmap (1 week)
We don't make an 18-month plan. We identify "quick wins" (or rapid gains in consultant language) and prioritize them by impact/effort. We tell you what's possible or not. We show you what we would do in your place with the existing tools and you choose.
3. Agile implementation (4-12 weeks depending on scope)
We think, we prompt, we code, we configure, we automate. Delivery every 2 weeks. You test in real conditions. We adjust. No 3-month tunnel without visibility.
4. Optional maintenance
Once deployed, we remain available. Light monthly subscription to fix bugs, adapt workflows when your tools evolve, add small features.
Let's emphasize one point: the term "consulting" is too light
We don't just advise. We don't just deliver a strategy via PowerPoint. We act with you. We show you and we do it then we train you to use it. Our LLMs document everything so you can be autonomous.
This is a big difference from the Big Four: they tell you what to do and take too long to do it (or charge you more). We do it with you from the beginning. And we teach you to do it yourself if needed.
Remote approach: consistent with our vision
For now , we're based in Paris, but we work throughout France and around the world. Remotely. Why?
- Because it's consistent with our vision of work in 2025: we can't preach efficiency and spend 10 hours a week commuting.
- Because modern tools allow it: Notion, Slack, Figma, Loom, we don't need to be in the same room to collaborate effectively.
- Because it reduces costs: no travel expenses to charge you, no time lost in transit.
Of course, we come on-site if it's really necessary for you (kick-off, collaborative workshop, training). But not for symbolic on-site presence.
Why maintenance is essential
An AI/automation project is never "finished." APIs evolve. Regulations change. Your needs become clearer.
That's why we systematically offer post-deployment maintenance. Not an obligation. An option. But frankly recommended.
- Specifically: a subscription of €500 to €1,500/month depending on complexity, which covers:
- Updates when an API changes
- Bug fixes
- Small evolutions (adding a field, modifying a format)
- Reactive support via email/Slack
The Big Four leave you alone after delivery. If something breaks, you have to launch a new mission. Back to square one: quote, negotiation, delays. We maintain what we built.
Conclusion
Transformation is inevitable, might as well ride the wave
AI isn't a fad. It's here. It's already transforming finance. Audit firms are automating their processes. Banks are deploying conversational agents. Asset managers are testing algorithmic trading.
You have two options:
- Wait for it to "settle down," for tools to mature, for use cases to clarify. But meanwhile, your competitors are moving forward. They're automating, gaining efficiency, reducing costs.
- Act now, but intelligently. No big transformation projects. Small concrete steps. An automated workflow. An automatically generated report. News monitoring. Quick wins that accumulate.
We're starting out, but we really act
We won't lie to you: Bubble Invest is just starting. We don't have 500 client references. We don't have an office overlooking the Champs-Élysées. We don't have 50 consultants at your disposal.
But we have something many firms no longer have: really acting instead of recommending. We use our own tools to manage our personal investments. We automate our own processes. We test what we recommend.
And frankly, we just want to build a company in 2025 that makes sense: pragmatic, transparent, accessible. A company that delivers what it promises. That bills for what it does. That remains available after delivery.
Contact to discuss your project
If you have a financial automation project, an idea to explore, or just want to discuss what's feasible (or not), write to us at contact@bubbleinvest.org.
We'll respond. Really. Not an auto-responder. Not a 15-question form. Just a direct exchange.
And if we think we're not the right fit for your need, we'll tell you. We prefer to be honest than to land a project we can't deliver.
References
[1] Emerj. (2024). "AI in the Accounting Big Four - Comparing Deloitte, PwC, KPMG, and EY". Retrieved from https://emerj.com/ai-in-the-accounting-big-four-comparing-deloitte-pwc-kpmg-and-ey/
[2] Finance Innovation. (2024). "Le Grand Témoignage d'Akur8". Retrieved from https://finance-innovation.org/le-grand-temoignage-dakur8/
[3] Shift Technology. (2024). "About". Retrieved from https://www.shift-technology.com/about
[4] NoCode Factory. (2025). "10 Formations Essentielles pour Freelances 2025". Retrieved from https://www.nocodefactory.fr/blog/formations-freelance-2025
[5] Forum des Compétences. (2024). "Un taux d'échec des projets IA inquiétant". Retrieved from https://www.forum-des-competences.org/un-taux-dechec-des-projets-ia-inquietant/
[6] Le Monde Informatique. (2024). "Gartner prédit l'abandon de 40 % des projets d'IA agentique d'ici 2027". Retrieved from https://www.lemondeinformatique.fr/actualites/lire-gartner-predit-l-abandon-de-40-des-projets-d-ia-agentique-d-ici-2027-97251.html
[7] Institut de l'Entreprise & McKinsey. (2024). "L'IA et l'évolution des compétences en France". Retrieved from https://www.institut-entreprise.fr/wp-content/uploads/2025/01/IDEP-McKinseyA5-8bis_Calameo.pdf
[8] Squid Impact. (2025). "Échec de l'IA en entreprise : les raisons et les clés du succès en 2025". Retrieved from https://www.squid-impact.fr/ia-entreprise-echec-reussite-france-2025/ [9] BM&A. (2024). "Projets d'IA : pourquoi échouent-ils ?". Retrieved from https://bma-groupe.com/lettre-d-actualites-techniques/eviter-les-echecs-projets-ia/
[10] Xerfi. (2024). "L'IA pourrait mettre un terme au tarif journalier moyen des cabinets de conseil". Retrieved from https://www.xerfi.com/blog/L-IA-pourrait-mettre-un-terme-au-tarif-journalier-moyen-des-cabinets-de-conseil_2259
[11] CGI Suisse. (2024). "IA Agentique et finance : la révolution des agents autonomes". Retrieved from https://www.cgi.com/suisse/fr-ch/blog/banque/ia-agentique-et-finance-revolution-des-agents-autonomes
