Adán MejíasAI · PM · Mentoring

Custom Solutions

When no standard service fits, I design one to your shape

Projects on Copilot Studio, Power Platform, Power Automate, conversational agents on Claude or GPT, and native Microsoft 365 integrations. Discovery, technical proposal with fixed price, build, and handover with training.

Tell me your case See the process

This line exists for when…

  • Your process crosses three or four systems and no off-the-shelf tool covers it end-to-end.
  • You need an internal conversational agent trained on your documentation, not a generic chatbot.
  • Your team uses five different SaaS tools and every handoff between them is manual.
  • You want to take Copilot Studio beyond the "general assistant" and build agents tailored per department.
  • There's a legacy process running on RPA where adding a generative reasoning layer is smarter than rewriting from scratch.

Honest positioning

I'm not "the AI guy". I'm a senior digital transformation consultant with ten years shipping projects in banking, pharma, energy and retail. AI is one of the layers I work with, alongside Power Platform, M365 integrations and process orchestration. The technology follows the problem, never the other way around.

What kind of projects fit here

  • Internal conversational agents — trained on your knowledge base, for specific teams (support, HR, sales).
  • Complex Power Automate workflows — multi-system flows with conditional logic, exception handling and real observability.
  • Custom SaaS-to-SaaS integrations — when there's no standard connector and the weekly manual sync keeps breaking.
  • Department-specific Copilot Studio — agents tailored to the language, flows and data of each business area.
  • RPA + generative layer — legacy processes that already work, with a reasoning layer on top to reduce manual exceptions.

From discovery to delivery, no surprises

  1. Discovery (1 week). Stakeholder interviews, technical review of your stack, mapping of the real process (not the org chart's version). We come out with a defensible scope.
  2. Proposal + fixed price. Document with proposed architecture, concrete deliverables, milestones, identified risks and a fixed fee. You accept, we adjust, or you cancel without commitment.
  3. Build (2–6 weeks). Depending on scope. Biweekly reviews with a working demo. Your team participates so the knowledge stays in-house.
  4. Delivery + training. Technical documentation, user manual, training session for the team that will operate the solution. Clean handover, not a "here you go".
  5. Support (optional). If you want ongoing maintenance or further evolution, we set it up as an hour bank or monthly retainer with a clear SLA.

Indicative investment range

Price depends on scope and integration complexity. Some references to set expectations before the discovery:

  • From €3,500 — Scoped SME project: a single Copilot Studio agent for one department, a contained Power Automate workflow, or an integration between two SaaS tools.
  • €8,000 – €15,000 — Mid-size project: multi-department agents, automations touching three or more systems, integrations with complex business logic.
  • From €25,000 — Mid-market multi-phase project: corporate Copilot Studio rollout with multiple agents, RPA + generative layer over legacy processes, internal bespoke platforms.

The budget is closed after discovery, never before. The initial 30-min discovery is free and qualifies whether the case fits.

When compliance blocks off-the-shelf AI, custom architecture is the answer

Many companies cannot enjoy standard generative AI because their regulatory or privacy framework forbids it. Banking, pharma, health, defense or any organization with sensitive data cannot send information to public ChatGPT or Claude. But they can, with conscious design, use generative AI inside their limits.

Some typical scenarios solved with custom architecture:

  • Data that cannot leave the tenant. Copilot Studio deployment inside the client's Microsoft 365 tenant: information never leaves, models consume local context, logs stored on their own infrastructure.
  • Mandatory European data residency. Azure OpenAI with EU endpoints, encryption in transit and at rest, no transfer to services outside the European Economic Area.
  • Nervous DPOs. Architectures with pre-model anonymization (PII detection + masking) and full audit log of every interaction.
  • Strict AI Act and GDPR. Auditable models, explainable decisions, no personal data sent to public services. Inspection-ready documentation.
  • Regulated sectors with periodic external audits. Design from day one with what the auditor will ask in mind: traceability, context separation, granular access control.

It's not magic. It's picking the right architecture before building. Most compliance incidents with generative AI I see in the market could have been avoided with two or three technical decisions taken on day one.

FAQ

When does a custom project make sense over a standard service?

When the problem is specific enough that off-the-shelf solutions don't fit: processes that cross three or four systems, business logic with many exceptions, integrations between SaaS without a standard connector, or department-specific needs that require a tailored experience.

How does the discovery work?

A free 30-minute call to understand the problem. If it fits, I run a one-week discovery with interviews and technical review. At the end I deliver a proposal with scope, architecture and a fixed price. If it doesn't fit, I point you to who can help.

Fixed price or hourly?

Fixed price. After the discovery you get a proposal with scope, deliverables and a fixed fee. Changes are handled as scoped change requests, not surprise invoices.

Which technologies do you build on?

Copilot Studio, Power Platform (Power Apps, Power Automate, Power BI), agents on Claude or GPT when the case justifies it, native Microsoft 365 integrations, and RPA combined with a generative layer for legacy processes. The criterion is always the problem, never the tool.

What happens after delivery?

Delivery includes technical documentation, a user manual and a training session with the team that will operate the solution. Ongoing support is optional, with clear rates depending on expected volume.

How long does a custom project usually take?

Between 2 and 6 weeks of development depending on scope, plus a week of discovery. Mid-market multi-phase projects can extend to 8-12 weeks with intermediate milestones.

Can you integrate with software vendors we already use?

Yes, as long as they have a documented API or are a SaaS with a stable connector. Custom integrations are one of the most common lines: connecting ERP with CRM, syncing tickets between systems, automating back-office steps between tools that don't talk to each other.

Got a case in mind?

Tell me the problem in 30 minutes. If it fits, we run a discovery. If not, I point you to who can help.

Tell me your case Prefer to write first