The arrival of artificial intelligence in enterprise software is the major technological event of the decade. All vendors now integrate AI functions into their sales modules, and Odoo is no exception. The latest version of Odoo enriches its CRM with predictive scoring, automatic content generation, and conversational assistance. For French executives of SMEs and mid-sized companies, understanding the opportunities and limits of these features has become a strategic topic.
Predictive scoring, the fundamental mechanism
Predictive scoring applied to sales rests on a simple idea. From a company's history of sales opportunities, a model learns the characteristics that distinguish won deals from lost ones. Size of the prospect company, industry, source channel of the opportunity, estimated amount, number of interactions carried out, seniority of the contact approached, length of the qualification phase. Each of these dimensions carries a weight in the prediction.
Once trained, the model applies this learning to each new opportunity. It calculates a score between zero and one hundred that expresses the probability, according to the model, that the deal will close. Salespeople can then prioritize their efforts on the highest-scoring opportunities, and make trade-offs on those with a more moderate score.
A promise that meets resistance
The appeal of this mechanism is obvious. It makes objective an intuition that was until now diffuse and unevenly distributed among salespeople. It frees up time by sparing salespeople from exhausting themselves on low-probability deals. It makes it easier to train new salespeople, who inherit a decision-support system instead of having to gain experience through trial and error.
This promise nevertheless meets practical resistance worth identifying. The first is the resistance of experienced salespeople, who may perceive scoring as a challenge to their expertise. This resistance is cultural more than technical, and it is managed through managerial support. Presenting scoring as a decision-support tool and not as a substitute for sales expertise is essential to overcome this resistance.
The second resistance is more subtle and structural. Scoring relies on the assumption that what worked in the past predicts what will work in the future. This assumption holds when the market and the offering are stable. It is more fragile when the company diversifies into new segments, launches radically different products, or experiences a major change in its competitive environment. In these periods of change, scoring can lead to favoring opportunities similar to those of the past at the expense of innovative opportunities that would prepare the future.
Data, a necessary condition for scoring
For a scoring model to be relevant, it needs data. And a lot of it. An SME with fewer than one hundred closed opportunities (won or lost) over the last twelve months probably does not have the volume needed for a reliable model. The scores produced would be erratic and would harm users' trust.
This volume requirement leads to a practical recommendation:
- start the CRM rollout in classic mode, without scoring, for the first twelve to eighteen months.
- During this period, you consolidate team practices, feed the database with complete structured data, and calibrate the stages of the sales pipeline.
- AI scoring can then be activated, on a sound basis.
The ethics of scoring
A little-discussed dimension of scoring is its ethical dimension. Predictive models learn the biases present in historical data. If a company, by tradition or habit, has served some categories of customers less well (for example very small businesses compared with key accounts, or certain geographic areas), scoring will replicate this bias by assigning lower scores to opportunities in the same categories.
This mechanism may seem benign. It is not always. It leads to self-fulfilling commercial decisions:
- you invest less in low-scoring opportunities, so you close fewer, so the score stays low.
- Potentially profitable segments can thus be gradually abandoned without anyone explicitly making the decision.
This reflection on bias calls for keeping a critical eye on scoring outputs. Salespeople and managers must retain the authority to over-invest in low-scoring opportunities if their business judgment deems it relevant. AI is a decision-support tool, not the decision-maker.
Generative AI and sales writing
Beyond scoring, generative AI is also transforming sales practices. Automatic generation of first drafts of personalized emails, conversation summaries, suggested follow-up actions, multilingual translation of sales pitches. These functions free up administrative time and can improve a sales team's productivity.
The risk here too is the loss of the company's unique voice. When all salespeople at all companies generate their emails with the same models, the creeping uniformity of B2B communication becomes a problem. For French SMEs and mid-sized companies that stand out through the quality of their relationships and the singularity of their voice, generative AI must be used with clear editorial discipline.
Our practical recommendation is to use generative AI for transactional communications (confirmations, cart reminders, acknowledgments of receipt) and to reserve human writing for high-value relational communications (qualifying first contact, negotiation, complaint handling). This split keeps the benefit of time saved without eroding the relational quality that makes the brand.
A transformation to manage
AI in CRM is neither a threat nor a revolution. It is a technical evolution that must be managed intelligently. For French executives who want to capture the benefits of these new capabilities without suffering the pitfalls, the right angle is to gradually integrate AI features into already structured processes, measure their impact, and adjust according to observed results.
ANOR supports this measured integration, combining Odoo functional expertise, strategic thinking on usage, and managerial support for sales teams in this transformation.