Skip to content
AI-Driven Everywhere: How Spadoom Runs on AI, and What We Will Not Automate
Insights · ·8 min read

AI-Driven Everywhere: How Spadoom Runs on AI, and What We Will Not Automate

Dario Pedol

Dario Pedol

Founder & CEO, SAP CX Architect, Spadoom AG

Share

Spadoom AG, a Swiss SAP Gold Partner for Cloud ERP and Customer Experience in Zug, is moving every part of the company onto AI agents: software development, sales operations, meeting follow-up, internal systems, security and marketing. The rule underneath all of it is short: AI does the work, people stay accountable. This article explains why we chose that path, the rules we work by and the decisions we deliberately keep away from the machines. It is the first part of a seven-part series; the other parts go into each area in detail.

TL;DR: At Spadoom, agents do most of the routine work and people make the decisions. Every system has named human maintainers. Every result is checked by two AI models from two different vendors before a person sees it. Nothing leaves the company without a human yes. Agents never merge or deploy code and never submit to SAP on their own.

Why we went AI-driven at all

We sell and run SAP cloud projects. Our customers pay for judgement: how to cut a process, where the ERP and CRM seam will break, what to leave out. They do not pay for the hours we spend typing meeting notes, copying opportunity data into partner portals or updating a wiki page nobody reads.

For years those hours were simply the cost of doing business. In 2026 that changed. Agents became good enough to take over whole chains of work, not just single sentences. So we made a decision: instead of a few AI pilots on the side, we would rebuild how the company works, area by area, and learn in our own house before we recommend anything to a customer.

That last point matters to us. It is easy to talk about agentic AI in a slide deck. It is harder to run your own accounting, code and sales on it and live with the consequences. We prefer the second kind of advice.

The rules we work by

The rule set grew over the year. Some of it we designed up front, most of it came from things that went wrong. Today it fits on one page.

AI does the work, people stay accountable

Every system we run, internal or customer-facing, has named human maintainers. When an agent changes something in that system, a maintainer is responsible for it, exactly as if a colleague had made the change. There is no “the AI did it”. If nobody wants to be named as maintainer, the system does not get agents.

Agents ask when the decision is ours

Agents work on their own until they reach a point that is genuinely a human decision: a trade-off, a customer commitment, something irreversible. Then they ask. Questions and tasks for people land in one central inbox, not scattered across chat, mail and tickets. We answer with one click, and the answer wakes the waiting agent up. The agent continues where it stopped.

This is the piece that made the whole setup practical. Without it, agents either guessed or stalled.

Two reviewers before one human

Every result, whether code, a plan, a configuration, a guide, a deck, a text or a data run, gets an independent second review by two AI models from two different vendors before a person sees it. Findings are ranked; critical and important ones are fixed before handover, minor ones are named. The report says who reviewed and what was found. Part 6 of this series covers how that works.

At Spadoom, every result is checked by two AI models from two different vendors before a person looks at it, and the report always says what they found.

Nothing leaves without a yes

Emails, chat messages, comments, CRM notes, LinkedIn posts, calendar invites and document shares written by an agent are held until a person approves them. Part 3 explains why that matters for prompt injection, not only for typos.

Proof instead of claims

Agents make confident mistakes. An agent will tell you, in a calm and convincing tone, that the deployment worked, the test passed, the record was updated. Sometimes that is not true. So we stopped accepting claims. A result counts when there is proof: a passing test suite, a log line, a live check against the real system.

What changed in practice

A few examples, each covered in its own article later in the series:

  • Code. An agent picks up a ticket only if a maintainer of that system set the label. It works in a fresh copy of the repository, both reviewers check, the full test suite runs, and then it opens a draft pull request. It never merges and never deploys.
  • Meetings. Customer meeting recordings are matched to the calendar and the customer’s email domains. If the match is clear, transcript and audio go into the customer’s folder and a note lands in the CRM without anyone typing. If it is unclear, the system asks.
  • Sales operations. Open CRM opportunities automatically get a draft SAP partner deal registration with a data check, linked both ways. It goes to SAP only after a person says so.
  • Internal systems. In autumn 2026 we replaced our issue tracker and wiki, and our ERP for accounting, payroll, contracts, projects and time, with our own systems. Our ERP runs on SAP BTP and has been leading since 30 October 2026.
  • Demos. For pitches with a fixed date, we set up SAP demo systems with the prospect’s context in days instead of weeks.
  • Time. Time is booked after each work session, and a nightly job fills gaps from mail and calendar.

What we will not automate

Being AI-driven everywhere does not mean everything is automatic. These steps stay with people, on purpose:

  • Merging and deploying code. Agents open draft pull requests. A person merges.
  • Talking to customers and partners. Agents draft. A person sends.
  • Commitments to SAP. Deal registrations are prepared by agents, but only submitted when a person says so.
  • Creating work for others. Tasks from a meeting are created only after someone says ok.
  • Changing the rules. When a guardrail blocks something, we do not let an agent work around it. Once a guardrail blocked an action, and we did it by hand instead of loosening the rule on the spot. Rules change deliberately, not in the moment.

What went wrong along the way

We would not trust a write-up like this if it only had success stories, so here are some of ours. Agents used stale credentials and failed in confusing ways. Agents ran into rate limits. Agents reported success that a live check did not confirm. Each of these turned into a rule, and the last one became the most important: a claim without proof is not a result.

The most telling example came from our content system in November 2026. The four-eyes rule there stopped the author from approving his own 136 post revisions. That was annoying, and it was correct. We did not switch the rule off. We added an audited admin override that requires a written reason. Four eyes apply to AI work too.

Frequently asked questions

What does “AI-driven everywhere” mean at Spadoom?

At Spadoom, AI agents do most of the routine work across the company: they write and test code, prepare CRM and partner records, file meeting notes, draft texts and check each other’s results. People stay accountable. Every system has named human maintainers, and anything that leaves the company or changes a decision needs a person’s yes.

Which decisions does Spadoom not hand to AI?

Merging and deploying code, sending messages to customers or partners, submitting deal registrations to SAP, and creating tasks from meetings all stay with a person. Agents prepare these steps and ask; a human decides.

How do agents ask people for decisions without flooding them?

Questions and tasks for people land in one central inbox. A person answers with one click, and that answer wakes the waiting agent up so it can continue. Agents only ask when the decision is genuinely the person’s.

Does AI make mistakes in this setup?

Yes. Agents make confident mistakes, which is why we require proof instead of claims: tests, logs and live checks. Every result is also reviewed by two AI models from two different vendors before a person sees it.

Is this relevant for an SAP customer, or only for Spadoom internally?

The same rules apply to AI in your SAP landscape: clear ownership, a human approval point before anything leaves the system, independent review and proof. We use our own experience to set up SAP Business AI scenarios that hold up in daily work.

What this means for your SAP project

If you are planning AI in your SAP landscape, start with the rules, not the model. Ask three questions for every scenario: who is the named person accountable for what the agent does, where is the point at which a human says yes before anything reaches a customer or partner, and what counts as proof that it worked? If you can answer those, the technology part is usually the easy one.

We are happy to share what we learned in our own house, including the parts that did not work. Have a look at our approach to SAP Business AI or talk to us about your scenario.

The series: how Spadoom goes AI-driven

  1. AI-driven everywhere: the rules and what we will not automate (this article)
  2. AI, security and our ISO 27001 process (coming 21 November 2026)
  3. Nothing leaves without a yes: approvals and prompt injection (coming 28 November 2026)
  4. From meeting to CRM without typing (coming 5 December 2026)
  5. We built our own internal systems with AI, and where that is a bad idea (coming 12 December 2026)
  6. Two reviewers, one human: how every result gets checked (coming 19 December 2026)
  7. How AI answer engines see us, and what we changed (coming 26 December 2026)
AIAI AgentsBusiness AIGovernanceWays of Working
Ask Spadoom · AI assistant

Ask Spadoom

Answers drawn from what Spadoom has published on this site, with links to the pages they come from.

Try one of these

Enter to send · Shift+Enter for a new line 0 / 600
Continue with an expert Opens the contact form with your question filled in.

AI-generated answers. Verify before acting. Questions are stored anonymously, without your IP address, so we can improve our content. Please do not enter personal data.

Next step

SAP Business AI implementation partner

Spadoom is the SAP Business AI implementation partner across Switzerland, Germany, Austria and Italy. 14-week median go-live. Live customers across DACH.

Related articles

Insights 7 min read

SAP Sapphire 2026: What Matters for Swiss SAP Customers

SAP announced the Autonomous Enterprise at Sapphire in May 2026. Five months and one TechEd later we can separate what shipped from what is still planned. Our status check for Swiss SAP customers: CX, Cloud ERP, AI pricing, RISE and GROW, and what to do before the first AI promotions end in December.

Dario Pedol · 30 Oct 2026
Read article →