Short answer. A consultancy that ships puts engineers on the account from week one and measures success in production systems, not slide count. A consultancy that only advises hands over a roadmap and a PDF, then leaves the build to someone else. Ableneo shipped 34 production AI projects in 2025 across banking, insurance, telecom, and energy, with roughly 4 of 5 projects reaching production, the number that separates a delivery partner from an advisory engagement.
The word “consulting” covers two different jobs. One is strategy work: workshops, maturity assessments, target operating models, a roadmap deck presented to the board. The other is build work: data pipelines, model integration, monitoring, and a system that a bank’s operations team runs every day. Many firms sell both under one name, then staff each job with different people and different incentives.
A bank evaluating an AI partner can see the difference in three concrete markers.
Ableneo runs 34 production AI projects in a year across 4 countries and 7 industries, and 94% of them use large language models somewhere in the stack. That mix, real production count plus repeat business, average clients run 1.7 projects with Ableneo, is the pattern a bank should look for in any partner’s track record, not just the client logo list.
Under the EU AI Act (Regulation (EU) 2024/1689), a bank deploying a high-risk AI system, credit scoring or fraud detection, carries deployer obligations that do not disappear when the strategy phase ends: human oversight, monitoring, and the ability to act on serious incidents. An advisory firm that leaves after the roadmap is signed is not on the hook for any of that. The bank is left holding a document and a compliance clock that keeps running.
The Digital Operational Resilience Act, in force for EU financial entities since 17 January 2025, adds a second layer: any AI system supporting a critical function needs tested resilience, ICT third-party risk management, and a named accountable owner. A DORA auditor does not accept a strategy slide as evidence of a working control. They want to see the system, the logs, and the person who owns it. That is precisely the gap between a firm that ships and one that advises, and it is the gap a bank pays for twice if it gets the choice wrong.
A delivery partner puts engineers on the account from week one; an advisory firm sends partners and a deck.
Read the proposal for verbs. A strategy deck is full of “assess,” “define,” “align,” and “recommend.” A build plan is full of “integrate,” “deploy,” “monitor,” and “roll back.” Look at the milestones: a strategy engagement’s last milestone is a presentation to the steering committee, a delivery engagement’s last milestone is a system live with real users and a support model behind it. If the proposal cannot name the production environment, the data source, and the go-live date, it is advisory work dressed as delivery.
Ask for the production count, not the client count: how many of their engagements in the last two years reached a live system used by real staff or customers, and what share stalled at pilot. Ask who stays on the account after go-live, because ongoing monitoring and incident response only work with continuity. Ask for one reference client where you can call the person who runs the system today, not the sponsor who approved the budget. A vendor that hesitates on any of these three questions is telling you the answer.
Scale works against delivery in a specific way. A large firm’s economics reward partner-led sales and junior-staffed execution, which is efficient for strategy work billed by the slide but expensive for engineering work billed by the outcome. OECD research on enterprise AI adoption across G7 firms finds that the gap between piloting AI and running it at scale is organizational as much as technical, tied to skills, data readiness, and integration effort that a strategy phase does not touch. A large consultancy can staff the assessment. Shipping the system requires engineers who stay past the handover, and that staffing model is harder to sell at partner margins.
Embedded delivery means the same team that scoped the use case writes the code, wires the integration into the core system, and stays through the first months of production, incident response included. There is no handoff between a strategy team and a build team, because it is one team. Governance is designed into the architecture in week one, not bolted on after a pilot works, so the monitoring, the audit trail, and the human-oversight controls a bank needs under the EU AI Act exist before the first real transaction runs through the system.
A partner built for delivery should have something running against real or representative data within 8 to 12 weeks, even if narrow in scope. That first working slice is the proof that the team can move from a workshop to a system, not a promise for a later phase. If a proposal’s first deliverable past the kickoff is another round of assessment, ask what happens if the assessment simply confirms what the bank already suspected. A roadmap without a working system by week 12 is a signal to renegotiate scope or look elsewhere.
Ableneo treats the strategy phase and the build phase as one engagement, not two contracts with two different teams. Across 34 production AI projects shipped in 2025 for clients including banks and insurers running 15 to 25 year old core systems, the work that holds up is the work where the same people who scoped the use case are still accountable for it in production. That is the standard we apply to AI transformation engagements: a system a bank’s own team can run, monitor, and defend to an auditor, not a slide deck filed after the workshop.
Key takeaways
Planning AI in a regulated business? Ableneo takes systems from classification to governed production.