Key Takeaways: Most AI roadmaps start with the technology — what models are available, what’s trending, what competitors claim to be doing. That’s backwards. The highest-ROI AI projects don’t start with the tech stack; they start with a specific process that’s costing someone real hours every week. The prioritization framework that works scores opportunities on three things: effort, impact, and data readiness — and only the last one is specific to AI. One high-confidence win, delivered in under 60 days, does more for AI adoption than a 12-month strategy deck ever will.
The Workshop Question That Usually Surprises People
When we run an AI discovery workshop with a new Odoo client, we open with one question: What do you spend time on that you wish you didn’t?
Not “What AI use cases have you heard about?” Not “What does your competitor’s AI do?” The bottleneck question.
What comes out usually surprises people — not because the answer is complicated, but because it’s mundane. Someone is exporting a report from Odoo, reformatting it in Excel, and emailing it to three people every Monday morning. A finance manager is manually matching bank statements to journal entries for an hour each day. A sales rep is copying notes from a Zoom call into crm.lead because no one set up a process to do it automatically.
These aren’t glamorous. They’re not featured in press releases. But they’re real, they’re costing measurable hours, and unlike the hypothetical benefits of a Grand AI Strategy, fixing them has a number attached.
That’s where the roadmap starts.
Why Technology-First Roadmaps Fail
The typical AI roadmap built by a vendor or consultant who hasn’t seen your actual operations will list things like: “LLMA neural network trained on vast amounts of text data to understand and generate human language. LLMs use the Transformer architecture and can perform a wide range of tasks — summarization,…-powered customer service”, “AI-driven demand forecasting”, “intelligent document processing”. Each sounds useful. None connects to anything specific in your business.
The problem isn’t that these ideas are bad — it’s that they’re disconnected from your actual bottlenecks and data. An AI-driven demand forecasting module is a good idea if your stock.move history is clean, if you have 24+ months of data, and if you have someone who can interpret and act on the predictions. If any of those conditions are false, you’ve built something that runs quietly in the background and changes nothing.
We’ve written before about how AI can’t shortcut 20 years of ERP schema design. The same principle applies to roadmaps: the AI investment is only as solid as the operational foundation under it. A technology-first roadmap assumes that foundation. A bottleneck-first roadmap checks it.
The Three-Axis Prioritization Framework
After surfacing the bottlenecks — usually through a 2-3 hour workshop with operations, finance, and technical staff — we score each candidate use case on three dimensions:
Impact: How much time, money, or error rate does this bottleneck cost right now? Be specific. “Manual invoice matching takes our finance team 1.5 hours per day” is scoreable. “Better AI in our processes” is not.
Effort: How much work does the solution actually require? A simple n8n workflow that polls Odoo via JSON-RPC and sends a Slack notification can be live in days. A custom MLA subfield of artificial intelligence where systems learn from data to improve performance on tasks without being explicitly programmed. ML algorithms identify patterns, make decisions, and generate… modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or… that predicts churn from crm.lead history takes weeks of data preparation alone. Both might have the same impact score — effort is what separates them on the priority list.
Data readiness: Does the data needed for this solution exist, in the right shape, inside Odoo? This is the axis most roadmaps skip, and it’s the one that kills the most projects. If Odoo fields are well-populated and consistently used, data readiness is high. If critical fields are blank, inconsistently filled, or living in external spreadsheets, it’s low — and no amount of engineering changes that without a data cleanup project first.
Score each candidate 1-3 on each axis. High impact, low effort, high data readiness: start there. High-potential projects with low data readiness: put them in month three or four, after cleanup. This isn’t a novel framework. What’s specific to AI is the data readiness axis — traditional automation projects rarely need to ask it. AI projects always do.
What a 6-Month Phased Roadmap Actually Looks Like
The goal of the first roadmap isn’t to list everything you’ll ever do with AI. It’s to give you a sequenced plan for the next six months that ends with something running in production that the team trusts.
Month 1-2: One high-confidence win. Pick the opportunity with the best combined score and deliver it. This is usually something rule-based or low-complexity AI — an automated alert, a report generated and emailed without manual intervention, a simple classification that routes records to the right person. The n8n + Odoo uninvoiced sales order alert is a good example of what this looks like: not glamorous, immediately useful, trusted by the team within a week.
Month 2-3: Data cleanup on the next tier. While the first win is running, start preparing the data for the next tier of projects. This means understanding which fields in account.move, sale.order, or stock.quant are actually reliable, identifying gaps, and building the cleanup process. Don’t skip this. Teams who skip it spend month four debugging a modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or… that performs poorly because the trainingThe process of exposing a machine learning model to labeled or unlabeled data so it can learn patterns. During training, the model adjusts its internal parameters (weights) to minimize a loss… data was dirty.
Month 3-4: Second project with a learning component. Once you’ve seen one project run successfully in production, you know more about your team’s capacity to adopt and maintain automated systems. The second project can be slightly more complex — a prediction modelA mathematical function trained on data that maps inputs to outputs. In ML, a model is the artifact produced after training — it encapsulates learned patterns and is used to make predictions or…, a classification step in a workflow, a document extraction pipeline. It should still connect to a specific, measurable bottleneck.
Month 5-6: Review, extend, or refocus. By month five, you have real production data on two projects. You know which assumptions were wrong, which were right, and what the team can actually sustain. This is the point at which the roadmap gets updated — not the original six-month plan. Most clients come in expecting a 12-month strategy. We push for six months with a built-in review gate, because the first 60 days of production will teach you things no discovery workshop can.
Why One Win First, Every Time
There’s a version of AI adoption where an organization announces five initiatives simultaneously, runs pilots on all of them, and eight months later has five half-finished projects and a team that has lost confidence in the whole program. We’ve seen it.
The counterintuitive lesson — and the Excel revolution parallel is relevant here — is that AI adoption follows the same pattern as every other enterprise software adoption. People trust what they can see working. They build habits around tools that save them time today. They don’t build habits around tools that will theoretically save time once the pilot is over and the data has been cleaned up.
One win, running in production, visible to the team, doing something measurable: this is what changes the organizational calculus on AI. It answers the implicit question everyone is asking — is this real or is it a consulting project? — in the clearest possible way.
The grand AI strategy comes after. Built on evidence, not aspirations.
The Discovery Workshop in Practice
The actual session runs two to three hours across three groups: operations and finance (who feel the bottlenecks), technical staff (who know what data exists and how it’s structured), and leadership (who decide what success looks like).
The output is a prioritized list of five to eight candidate use cases, scored on the three axes, with a rough effort estimate for each. Not a slide deck. Not a technology evaluation. A list with numbers attached.
From there, we work with the client to select one item for immediate implementation and two for the second wave. Everything else goes into a parking lot to revisit in three months — when we actually know more about what works in their environment. Composable ERP architecture helps here: designing each AI layer to be modular means you’re not locking in one approach across everything. You can replace components as you learn.
The constraint is intentional. Scope pressure is real, and “just add one more thing” is how six-month roadmaps turn into ongoing strategy engagements that never deliver anything. The bottleneck is the starting point. The win is what makes the next one possible.
At Trobz, our AI discovery workshops end with a deliverable — a scored opportunity list and a first-phase action plan — not a strategy deck. If you’re trying to figure out where to start with AI in your Odoo environment, we’d be glad to work through it with you.