Centribal presented NOVA July 1, 2026, combining conversational and generative AI to handle enquiries and qualify sales contacts for small and medium-sized businesses.
The announcement describes an initial contact that collects data and identifies needs before handing the conversation to sales. The company also proposes service and appointment-management uses. It supplies no named-customer results or comparison of commercial improvement.
Agree which contacts sales will accept
Assessing qualification requires a shared marketing and sales criterion. A name and phone number permit contact but do not establish that the offer fits. Record the need understood and information missing before deciding the next action. This is an assessment framework, not an observed NOVA result.
The agreement should become explicit fields and an identifier that preserves continuity. If the contact is duplicated, its origin disappears or the reply lands in an unowned inbox, automation has accelerated record production rather than business. The meaningful demonstration is an enquiry reaching a person and becoming a commercial action.
| Stage | Useful information | Outcome to examine |
|---|---|---|
| First enquiry | Declared need and context | An understood enquiry without invented details |
| Qualification | Criteria agreed with sales | An accepted opportunity or explained rejection |
| Handoff | Record, owner and next action | Effective contact with preserved context |
| Close | Purchase, cost and origin | Commercial contribution linked to the process |
Collapse notes
This is an editorial framework for assessing the workflow, not a list of features verified in a NOVA installation. Its purpose is to distinguish a pleasant conversation from an opportunity someone is prepared to pursue.
Connect contracted capacity to the remaining work
Centribal's general plans page offers conversation capacities by tier. That catalogue does not itself establish what a particular NOVA contract includes. Buyers should confirm applicable limits and how conversations are counted before calculating cost per opportunity.
Cost per accepted opportunity helps organise that discussion. The denominator should not include everything the bot labels a lead, but what sales accepts under stable criteria. The subsequent purchase and contribution must then be observed, including contacts that do not progress. That stops optimistic classification being confused with a more productive funnel.
Retain context when handing off to a person
A usage review can follow an incomplete enquiry or a service unsuitable for the need. Buyers need to establish which data reach sales and how the next step is explained to the user. That reviews handoff quality, rather than treating every opportunity label as a possible sale.
A personalized response does not erase the preference for a person
Stated preferences in Spain, EAE report published February 2024.
Tap or hover for details. All values are also available in the data table.
Collapse notes
EAE Business School: 1,000 respondents, age profile 18–70. The original does not detail fieldwork dates, method or frame. Statements overlap; they are not exclusive preferences or an experimental comparison of chatbots and humans. NOVA users are not measured. The 9.1% refers to AI recording preferences for personalised advertising, not consent to any AI use.
Explore the data table
| Category | Surveyed consumers |
|---|---|
| Advice from a real person | 56.9% |
| Relevant product notifications | 28.9% |
| Does not mind AI recording preferences for ads | 9.1% |
EAE, reported by Statista, records advice from a real person in 56.9% of statements, while 9.1% do not object to AI recording preferences for advertising. These are different statements; they do not prove a bot provides worse service. For NOVA they locate a handoff responsibility: the first conversation should explain what the assistant will resolve and when a person will continue.
A correct answer can recommend the wrong next step
In commercial information, a sentence can faithfully reflect a document yet guide the person towards an unsuitable action. A tariff may require an additional condition, an appointment may need preparation or a service enquiry may belong to another area. The proposed review would retain the question, information used and suggested next step. Reviewing wording alone is insufficient. The criterion should examine whether the user can make an understandable decision from the response, rather than assessing only fluency or whether a phrase appears in an approved knowledge document.
Internal documentation creates a currency problem. An answer may accurately reproduce a version that is no longer the available offer. The company needs to establish who maintains information, when it updates and how it identifies an answer using older content. That responsibility exists even if a supplier makes document upload easy. The launch describes an intelligent document base but publishes no change-history assessment verifying the time validity of responses for a particular customer. An apparently accurate answer therefore still needs its source and applicable conditions checked.
Continuity also needs an owner. A qualified contact loses value if nobody receives the task or if the person must explain everything again. Time between referral and human conversation should remain visible alongside enquiries left open. A smaller company's sales capacity may vary during the day and after business hours. Continuous assistant availability, which Centribal presents as an intended use, provides a complete experience only when the next step the user needs remains possible and clearly explained. A responsive opening message cannot compensate for an unavailable service presented as immediate.
Measure selection errors as well as successes
Qualification assessment should observe contacts accepted by the system and those left out. Reviewing only referrals cannot establish how many people with relevant needs were excluded. Treating every bot label as correct makes the assessment circular. A team would need an independent conversation review with explained decisions, including cases not forwarded to sales. This article proposes that framework without presenting an accuracy rate the supplier has not published. A high total of automatically labelled leads would otherwise tell readers little about the reliability or commercial relevance of those labels.
Errors have different costs. Referring an unsuitable enquiry consumes salesperson time; omitting a viable need can lose an opportunity. The economic weight depends on offer and capacity, rather than a universal percentage. A comparison should retain both outcomes and the information available when the decision was made. A later change in the person's need does not automatically make an earlier classification wrong. Preserving the timeline helps assess the system fairly and usefully, especially when a conversation develops rather than arriving with every relevant detail already supplied.
Lead generation and decisions are reported data-quality benefits
Survey of marketing decision-makers, January 2023. Three selected responses.
Tap or hover for details. All values are also available in the data table.
Collapse notes
Ascend2: 374 decision-makers surveyed online January 16–22, 2023; multiple answers. Statista labels the scope worldwide, while the original does not detail countries. Perceived benefits, not percentage improvements or quality of NOVA classifications.
Explore the data table
| Category | Decision-makers citing benefit |
|---|---|
| Customer experience | 61% |
| Lead generation | 35% |
| Overall decision-making | 33% |
In Ascend2’s study reported by Statista, 35% cited lead generation and 33% overall decision-making. That context does not certify an opportunity label. For NOVA, the information supporting both uses goes beyond name and phone number: an expressed need, a fit condition, an explained rejection and an owner for the next action. Only then can a team assess whether automation saved work or created an opportunity sales can pursue.
Cost per conversation needs a stable definition
The budget should retain supervision and document maintenance. Correcting an answer, updating a condition or reviewing a referral takes work even when the opening reply is automatic. The supplier describes implementation support as part of its proposition but publishes no complete cost per outcome here. An economic account would need those components under the actual contract and separate initial expenses from operation. Omitting tasks transferred to another team can exaggerate the saving from repetitive service work, making a buying decision depend on an incomplete account rather than the observed workload.
Follow-up needs to reach a recognised purchase and its contribution without removing contacts that do not progress. Sales acceptance is a stage, rather than a closed deal. Faster response can also coincide with changes in campaign, opening hours or staff. A comparison that does not retain those conditions establishes only association. NOVA's announcement cannot allocate income between response generation, classification and human work. They form one journey requiring observation before causality can be assigned to a single tool or an automatically reported lead count.
The site identifies Always On Technology, S.L. as its owner. Assessing the proposal still requires following actual enquiries through accepted opportunities and purchases, with the necessary supervision time. The announcement describes capabilities and proposed uses without quantifying that outcome.
In this article
Marian Vasile
Marian Vasile is a MarTech and growth professional. His work spans product marketing, SEO, paid advertising and deep online tracking, including behavioral analytics and attribution.