For firms whose clients are small and mid-size businesses: accounting and bookkeeping practices, digital and marketing agencies, IT and managed-service providers, consultancies and software implementation partners. Adrian builds and runs the AI service line under your brand or alongside it: the automations your clients pay for, the systems behind them, the staff training and the controls your clients expect. The model is already live at an Irish chartered accountancy firm, where the service line went from no AI offer to a client platform, a published offer and an AI Academy.

Talk about a partnershipHow it runs

What you get

Six things a partner gets from the division, each one already running at the reference firm.

  1. Sht 01

    An AI service line with its own offer

    A named AI and automation offer on your website, a price list your team can quote and a proposal brief for each client scenario. At the reference firm this went from no AI offer to a published offer inside a year.

    Proof
    Offer published; automation clients in Ireland, the UK and the US
    You keep
    Your brand, your pricing, your client relationship
  2. Sht 02

    Automations your clients pay for

    Supplier bills read by AI and drafted in the ledger, CRM deals turned into approved invoices, client onboarding, lead routing, reporting and credit control. Delivered under your name, sold as recurring work.

    Proof
    HubSpot to Xero across 3 legal entities live since April 2026; prepayments matched 107 of 107
    You keep
    Recurring revenue per client, delivered under your name
  3. Sht 03

    Operations that run on clean data

    Client records audited and moved into one schema, recurring tasks generated from the services each client buys, billing and follow-ups on a schedule, and a dashboard for every team member.

    Proof
    587 client records rebuilt in 3,172 writes with 0 failures
    You keep
    One client list the whole team trusts
  4. Sht 04

    A team that uses AI safely

    A live session with demos for the whole team, an academy of lessons and prompts written for your own work, and an AI use policy staff can follow.

    Proof
    Full team (10+) trained at the reference firm; 8 lessons, 48 prompts
    You keep
    Lessons, prompts and a policy your staff keep using
  5. Sht 05

    Controls your clients can sign off

    EU-only AI processing where the data needs it, identity checks before an account opens, an audit trail on every action and a person approving anything that moves money. Built into the software, not left to a policy document.

    Proof
    Romania's 4th ECSP crowdfunding licence, with Adrian as lead on the audit
    You keep
    Controls your clients, insurer and auditor can read
  6. Sht 06

    One senior owner, not a vendor queue

    Scoping, build, training and support in one seat, reporting to a founder or partner. Former FinTech COO, so the operating side of a service line is covered as well as the code.

    Proof
    Client platform from first commit to production in 10 days
    You keep
    Code, accounts and runbooks in your name

How the partnership runs

Discovery, a pilot on real documents, a rollout you can sell, then a handover. No step needs the next one to be signed.

  1. Rev A

    Discovery

    Weeks 1 to 2

    A session with the founders or partners, a look at the stack (CRM, ledger, inbox and the tools in use) and three client scenarios. Out of it: a written plan, the first pilot chosen and the working model agreed: white-label, co-branded or referral.

  2. Rev B

    Pilot

    Weeks 3 to 8

    One flow live for one client, or one internal process for your team, behind approvals and a test set built from real documents. Accuracy and hours saved are measured before anyone funds more.

  3. Rev C

    Rollout

    Months 3 to 6

    The pilot becomes an offer you sell: a proposal brief, an onboarding checklist and a staff session. Each new client automation reuses the same controls and tests.

  4. Rev D

    Handover

    Any time

    Code, accounts, runbooks and the test suite sit in your name from day one. You run it with an optional retainer, or Adrian stays on as the lead of the division.

Safeguards

  • EU-only AI processing, enforced in code: EU-hosted infrastructure and EU model endpoints with data caching disabled. A non-EU model is an opt-in fallback, never a silent one.
  • An AML gate: a client account cannot be opened until the identity check is verified.
  • An audit trail on every action: who uploaded, what the AI read, who approved and what was posted, with retention rules agreed in writing.
  • Human approval before anything moves money or reaches a ledger. AI drafts; a person confirms.
  • Numbers computed, words generated: fees, VAT and journals come from deterministic rules, and AI writes only the wording.
  • Least-privilege access, tenant walls between clients, no model training on client documents, and a data processing agreement before the build.

Why it matters to you

Your name is on every output. A client document leaving the EU silently, an account opened before an ID check, or a model guessing a figure in a proposal is your problem, not a bug. The controls are built for that.

Proof from a live engagement

An Irish chartered accountancy and audit firm had no AI offer in October 2025. These are the numbers from the service line Adrian built and runs there, plus the finance work and the regulated FinTech years behind it.

  • 10 daysfrom first commit to live production for the reference firm's 20-module client platform
  • 10+people trained, the full team, in an AI for Accountants session, plus an 8-lesson AI Academy
  • 3legal entities on HubSpot to Xero invoicing delivered under the firm's brand, live since April 2026
  • 107/107Xero prepayment journals matched against the bookkeeper's own entries, running daily
  • 99.8%field accuracy of the AI bills reader on a 64-file synthetic test corpus, under one cent per document
  • 0failures in 3,172 writes when 587 client records were rebuilt

The case studies behind the numbers

  1. DWG AC-101Live in production 10 daysfirst commit to live production

    Accountancy firm: AI service line from zero, platform live in 10 days

    An Irish chartered accountancy and audit firm offered no AI. Adrian built the service line and runs it: a 20-module client platform live in 10 days, EU-only AI processing enforced in code, an AML gate, Irish statute as software, and AI training for the whole firm.M.A. Whately & Co., Irish chartered accountancy and audit firm     99.8%AI bills field accuracy, synthetic test corpus TypeScript, Vercel, Supabase (PostgreSQL, RLS), Gemini via Vertex AI EU Case 01 / 07Read case
  2. DWG AC-104Live in production 3 entities4 currencies, live since April 2026

    HubSpot to Xero, multi-entity invoicing

    Automation that moves real money: closed deals become approved Xero invoices in the right legal entity and currency, emailed to the customer and chased until paid. Live since April 2026, about 52 invoices a month under signed approvals.Multi-entity B2B software reseller (US, UK and EU companies), delivered through M.A. Whately & Co.     212+invoices processed by September 2026 n8n, HubSpot API, Xero API (3 tenants, OAuth2), Supabase (PostgreSQL) Case 04 / 07Read case
  3. DWG AC-106Live 107/107match against the bookkeeper's manual entries

    Tour-operator Xero prepayments, 107/107

    Supplier bills read, coded and moved from prepayments to expense by arrival month, as draft journals the bookkeeper approves. Matched the bookkeeper's own entries 107 of 107 before go-live.Irish tour operator, delivered through M.A. Whately & Co.     10/10daily production runs green Python 3.12, Xero Accounting API (OAuth2), GitHub Actions, pypdf Case 06 / 07Read case
  4. DWG AC-102Licensed ECSP operator Mar 2024ECSP licence granted by the ASF

    HighCrowd: EU crowdfunding licence, as COO

    Three years as COO of a regulated real-estate crowdfunding platform. Adrian was the lead on the authorisation and audit with the ASF, built automated KYC and AML onboarding and the investor-protection flows, and the platform received its EU crowdfunding (ECSP) licence in March 2024.HighCrowd, real-estate crowdfunding platform for retail investors     45 to 8minutes to onboard an investor ECSP, Regulation (EU) 2020/1503, LemonWay (PSD2), KYC and AML, Penetration testing Case 02 / 07Read case

What clients said

  1. Upwork review

    Adrian is a total pro. Just looking at his CV you can tell he's not the average freelancer on here, and the work really proved it. The automation runs perfectly…

    Nicolas P.Workflow set-up

  2. Upwork review

    Capraru was extremely knowledgable in N8N, AI and general automation. He put together an excellent workflow that helped us achieve human scoring, automated content. We are hiring him for our next AI automation project.

    UK internet-marketing firmAI workflow automation, n8n

  3. Upwork review

    Great professional. Fast, smart, clear. Already doing a new contract with him to keep working.

    Spanish travel agencyHubSpot and WhatsApp AI automation

Questions partners ask

White-label, co-branded or referral: what is the difference?

White-label: the service carries your brand and Adrian works as your AI lead behind it, as at the reference firm. Co-branded: the offer names both. Referral: you introduce a client and Adrian contracts with that client directly. The model is agreed in discovery and can differ per client.

Who owns the client relationship?

You, by default. In white-label and co-branded work, client contact runs through you or under your brand, and your clients are never marketed to directly.

Our clients use different tools. Does this still work?

Xero, HubSpot, Supabase, n8n and Make are the deepest experience. QuickBooks, Dext, Zapier and most tools with an API follow the same pattern; anything else is checked in discovery, before any build money is spent.

Where does client data live, and who can see it?

EU-hosted infrastructure by default, least-privilege API access, no model training on client documents, an audit log of every action and retention rules agreed in writing. A data processing agreement is signed before the build.

Do our staff need technical skills?

No. Staff approve, review exceptions and use the assistant. Build, tests and monitoring stay with Adrian. The team session and the academy cover day-to-day use.

What does the EU AI Act mean for us?

Mostly lighter duties: AI literacy for staff and transparency when clients interact with AI. Higher-risk uses such as screening job applicants or scoring creditworthiness carry heavier duties from December 2027. Discovery shows which bucket each use sits in.

How is it priced?

Per partner, after discovery: a margin on resold work, a revenue share or a monthly fee, set once a pilot has produced real numbers.

Where do you work?

Remotely, with partners in Europe, the UK, North America and beyond, on Eastern European Time with long overlap both ways. Irish statutory rules are already encoded as software; rules for other countries are scoped per partner.

Start with one client and one process

Send your firm's size, its stack and one client process that costs hours every month. One email is enough to start a discovery session.

Go to the contact pageEmail Adrian