DAVISA AI STUDIO · METHOD

Discovery, Pilot, Scaling. The AI Studio method.

Most AI projects fail for lack of method, not lack of technology. Davisa AI Studio applies explicit discipline in three phases, each one with a concrete deliverable, a fixed timeline and a clear decision point. This page explains exactly how we work, end to end.

Why an explicit method

Enterprise AI projects fail for three recurring reasons. The first is lack of scope clarity: the customer wants to "apply AI" and the vendor ships a proof of concept that doesn't connect with any real process. The second is magical thinking: management expects AI to solve a problem that is actually about organisation, data governance or decision rights. The third is the absence of a KPI measured before and after: without a baseline, you can't claim the project delivered anything.

Davisa AI Studio applies a method that attacks all three. The Discovery closes the scope before any development budget is committed. The prioritised roadmap expels cases where AI is not the right solution (and names them). The baseline and the KPI target are set in writing before the Pilot starts, not after.

Every phase has clear deliverables and an explicit decision point. At the end of the Discovery, the customer decides whether to launch a Pilot. At the end of the Pilot, whether to move to Scaling. There is no multi-year commitment or contract that ties you beyond the current phase. The relationship grows because both sides want it to, not because of a clause.

The method is not marketing fluff. It is the only way we have found to ship AI cases that are still working on day 200 after go-live. The following sections describe each phase with the day-by-day detail.

Phase 1 · Discovery

5 days · fixed price

Goal: understand the customer's domain, map AI-candidate processes, assess technical feasibility (data availability, quality, integrations) and size the expected ROI of each candidate. Output: a prioritised roadmap that stands on its own, whether you contract a Pilot afterwards or not.

What we do during the 5 days

Day 1 · Kickoff workshop. 3 to 4-hour session with management and operational areas. Customer process mapping, pain-point identification, sizing of team and volume involved.
Days 2-3 · Stakeholder interviews. 1-1 sessions with the owners of the candidate processes (CFO, operations director, IT director, area leads). The goal is to capture the domain data that only the people who run the process daily actually know.
Days 3-4 · Business Central data review. Technical analysis of the ERP state: tables involved, historical quality, configured analytical dimensions, active dv* extensions, existing integrations. Without clean data there is no useful AI.
Day 4 · Mapping and prioritisation. The Davisa AI Studio team builds the opportunity/impact/cost matrix for 5 to 10 candidate processes. Each candidate is evaluated with explicit criteria: estimated saving, Pilot cost, technical feasibility, available sponsor.
Day 5 · Roadmap presentation. 2-hour session with management. Delivery of the roadmap document, summary PDF for the committee and recording of the session.

Deliverables

Prioritised roadmap. 20 to 30-page document with 3 to 7 AI opportunities evaluated by impact, cost and technical feasibility.
Sizing of the recommended Pilot. For the best-fit case, functional scope, KPI target, estimated baseline, technical stack, timeline and indicative investment range.
Management presentation. 8 to 12-slide summary PDF to take to committee or board.

Price: fixed, on request. Not published on the web.

Decision point: with the roadmap in hand, the customer decides whether to launch a Pilot and which one. If they don't, they keep the roadmap. It stands on its own.

Phase 2 · Pilot

8 weeks · fixed scope

Goal: roll out into production one use case chosen in the Discovery, with KPI measured before and after against the baseline. At the end of the Pilot, the customer knows in concrete numbers what has been saved, what has improved and what decision to make for the next cases.

Typical structure of the 8 weeks

Weeks 1-2 · Detailed design. Technical architecture of the case, flow design, integration with BC and dv*, definition of prompts and models, screen mock-ups and approvals. Final scope sign-off.
Weeks 3-6 · Development and testing. Implementation of the AI pipeline, configuration of Azure OpenAI, Document Intelligence or AI Builder depending on the case, integration with BC via API, internal testing with real customer data in a sandbox environment.
Week 7 · Controlled go-live. Production rollout over one business unit, one company or a bounded perimeter. Daily support of the customer team to resolve exceptions and fine-tune confidence thresholds.
Week 8 · Measurement and adjustment. Comparison against the baseline, fine-tuning based on real usage, team training, operational documentation and closing presentation to management.

Deliverables

Functional solution in production. The use case running on the customer's tenant, integrated into BC and the daily operational flows.
Measured KPI. Concrete number of hours saved, errors avoided or revenue accelerated, compared against the baseline set in the Discovery.
Technical documentation. Architecture, prompts, model configuration, BC connections, operations runbook and maintenance plan.
Customer team enablement. Training sessions for end users and the technical team that will maintain the case day to day.

Price: fixed per case. Not published on the web.

Decision point: with the measured KPI in hand, the customer decides whether to move to Scaling, run an independent second Pilot or close out the project on this case. No continuity commitment.

Phase 3 · Scaling

ongoing

Goal: roll out the next use cases from the roadmap, optimise the ones already in production and train the customer team so AI lives inside the company without structural dependency on Davisa. Scaling is consolidation: AI stops being a project and becomes operations.

Modes

Package of N next cases. Closed scope, multi-year calendar, price per case. Useful when the roadmap has 3 to 5 clear cases ready to execute and the company wants budget visibility.
Monthly retainer with SLA. Monthly hours bucket for evolution of productive cases, incident support, fine-tuning and exploration of new cases. Useful once 2 to 3 cases are in production and continuous cadence is needed.
Ad-hoc consulting by the hour. For companies with an internal AI team that need expert backup at specific moments (architecture review, stack opinion, troubleshooting a case). No minimums.

Deliverables

Each new case delivered as a Pilot. The difference is that the architecture, BC connection, data governance and team knowledge are already in place. That's why the next case is always faster than the first.
Monthly releases on productive cases. Model improvements, new suppliers learned, new exceptions handled, new integrations.
Monthly KPI per productive case. So management sees the consolidated saving or improvement month by month, not just at the end of each Pilot.

No multi-year commitment. The relationship grows if the customer wants it to, not by contract. If at any point you decide to in-source operations, the documentation we ship makes it possible.

The principles we apply

The method runs on six principles. They are not corporate slogans. They are the rules we apply on every project and that justify many of our operational decisions.

Customer data, always on-tenant. All AI infrastructure runs inside your Azure tenant or in a managed tenant under your contractual terms. Documents, prompts and responses do not go to public OpenAI nor get used to train third-party models. If a case requires sending data out, we say so before accepting it.
Every case is measured. No KPI, no start. In the Discovery we set the KPI target and the current baseline of each Pilot candidate. We do not start a Pilot without that before/after pair. If the baseline isn't measurable, the case isn't a candidate.
Discovery independent of the Pilot. The Discovery is not a sales pitch in disguise. It is field work with a formal deliverable. If after the Discovery the honest conclusion is that no case is worth it yet, we say so in writing. The roadmap stands on its own.
Everything documented. You can carry on without Davisa. Every Pilot ships technical documentation, architecture, prompts, models and processes. If you decide to continue with another vendor or in-source the team, you can. Our goal is for the case to work on day 200, not to chain you to dependency.
Promises about hours and errors, not transformation. We don't promise digital transformation or business revolution. We promise person-hours saved per month, error percentages reduced and concrete numbers with concrete deadlines. When the KPI allows, we promise a range. When it doesn't, we say so.
Honesty about the limits. If a case doesn't fit AI, we say so. If your volume doesn't justify the Pilot, we say so. If the right call is to first stabilise data governance and then automate, we say so. We'd rather lose a project than deliver one we shouldn't have accepted.

When the method does NOT fit

There are four scenarios where the honest answer is that the Davisa AI Studio method is not for you. If you recognise yourself, better to look for another route.

If you want a quick 2-3 hour Discovery. Ours is 5 days of work with your team, not an extended sales meeting. If you need a shorter workshop format, there are generalist AI consultancies that offer it. Not us.
If you expect a Pilot in 2 weeks. The realistic timeline to deliver a productive AI case with KPI measured against a baseline is 8 weeks. Speeding it to 2 or 3 means skipping phases (training, testing, fine-tuning) and shipping something that doesn't hold up by day 60. If you have a structural time pressure, look for another vendor with another method.
If you can't assign 1-2 part-time people to the Discovery. We need the customer's domain data: how your process works, what exceptions it has, what decisions are made at each step. Without someone of yours available for interviews and data review, the Discovery falls short and the roadmap loses value.
If the decision is made by a committee without an executive sponsor. AI pilots without an executive sponsor die along the way. When it's time to decide whether the model's proposal is accepted, whether an operational process is changed, whether budget is assigned to Scaling, someone with authority needs to be at the table. If there's only a committee, the Pilot dilutes.

Frequently asked questions

How much does the Discovery cost?

The 5-day Discovery is a fixed-scope, fixed-price package. We don't publish the figure because it depends on the number of processes to map, the required technical depth and the sector. The typical range for an SMB with BC in production is reasonable as an entry to an AI project and we always confirm it in writing before starting. Request a proposal from the /en/ia/discovery/ page and we'll send it within 48 business hours.

What happens if after the Discovery I decide not to launch a Pilot?

You keep the roadmap. It is a 20 to 30-page document with 3 to 7 AI opportunities evaluated by impact, cost and technical feasibility, sizing of the recommended pilot and indicative investment ranges per case. It stands on its own: you can use it internally to build your own roadmap, take it to a committee or ask other vendors to quote against that scope. We don't sell Discovery to sell Pilot: if the roadmap concludes the timing isn't right, we say so.

Is the Pilot fixed-price or time-and-materials?

Fixed price per use case. The scope is set at the end of the Discovery and signed before the Pilot starts: deliverable, KPI target, baseline, timeline (8-week standard) and rate. We don't bill by the hour during the Pilot. If a scope change arises during the Pilot, it is treated as a formal change order and quoted separately, not buried in a monthly timesheet.

What KPI do you measure per case?

It depends on the case. For supplier invoice automation we measure person-hours saved per month in the accounts payable team and percentage of lines extracted correctly without manual intervention. For industrial scrap prediction we measure basis points of scrap avoided against the monthly baseline. For the executive close summary we measure CFO writing hours saved. In every case the KPI is set in the Discovery, the current baseline is measured before starting the Pilot and re-measured at close. Without that before/after pair there is no honest KPI.

Do you work on any ERP or only Business Central?

Our value proposition is native integration with Microsoft Dynamics 365 Business Central and the dv* extensions. We work exclusively on BC. If your ERP is another (SAP, Sage, Odoo, custom), we can talk, but you lose most of our edge: knowledge of sector processes on BC, the specific tables, the FacturaE and Verifactu e-invoicing flows already integrated. In that scenario the honest answer is to point you to an AI consultancy that specialises in your ERP.

How many cases does a company usually have in the roadmap?

Between 3 and 7 opportunities evaluated in the document, of which 2 to 4 are usually clear pilot candidates in the first year. The rest tend to be adjacent cases that unlock once the first pilot's architecture is in place (same data, same integrations, same team knowledge). An SMB with a mature BC usually consolidates 3 to 5 AI cases in production within 18-24 months if it decides to scale.

Next step

Already a Davisa customer?

The AI layer lands as an extension of the relationship you already have with BC and the dv* extensions. Your usual advisor coordinates with the AI Studio with no new onboarding.

Talk to the team →

New to Davisa?

We start with the 5-day Discovery. In one week you have a prioritised roadmap, a sized Pilot and an informed decision to take to committee.

Request AI Discovery →