Most companies automate the wrong thing first. In a five-day audit I show you which three processes have the largest leverage, with data instead of gut feel. Calendar mining, Slack pattern scan, shadowing, tool inventory. Output is a Process Register with 40 to 80 entries, top 10 ranked by EUR per year leverage.
Function leads describe the target process. The real process lives in calendar, Slack and tool invoices. Four sources, parallel to the workshop day, fill the Miro board with reality.
Every recurring meeting in the last 90 days is a process candidate. If you sit in five hours of status calls per week, there are processes behind them that no BPMN diagram shows. This source alone surfaces the most frequent bottleneck in 80 percent of engagements.
Top 20 channels by volume. Search for three phrases: can someone, status, blocking. Each phrase marks a handoff that goes wrong or a question that is tribal knowledge. Output: hit list with channel, phrase, frequency per week.
90 minutes per function lead, on a real ticket. One question is enough: what are you doing for the third time this week? You catch processes nobody documents because they seem too trivial. Those are the ones that add up to half an FTE.
SaaS invoices over fifty euros per month. Every tool has a process behind it. The real stack is never the documented stack. Tools without API show copy-paste workflows, tools without owner show processes that fall apart when X quits.
On the detailed BPMN frame of every top-15 process we stick five colored markers. A quick glance at the board shows the most expensive box.
Team change mid-process. Every handoff costs wait time and re-briefing.
Data manually copied between systems. Typo risk, always the same person.
Approval queue, batch run at month end, slot at the senior. Sits idle, costs cycle time.
Only one person knows. Sickness or vacation kills the process.
Tool forces copy-paste. Automation fails because of this, not the model.
Three pattern pages show how Handoff, Re-Keying, Wartezeit and Tribal land together in real workflows.
Three handoffs, manual re-keying between support, Stripe dashboard and email. Wait time on the finance slot.
Status calls that decide nothing. Tribal knowledge in Slack threads, wait time between meeting and action item.
'Can someone quickly check' requests. Tribal knowledge with senior engineers, handoff between DM and wiki.
Five axes, 0 to 3 points each. The sum is the score, the score lands in a quadrant. No more gut-feel discussion, only assumptions per axis.
Honest comparison: 10 axes, decision tree, quadrant. So you do not buy the wrong tool.
Volume times ROI on the X axis, standardizability on the Y. Four cells, four recommendations. Quick wins ship first, strategic plans as a quarter, wait waits for better data, do not automate keeps the human.
Quarterly project, agent with carrier API, runbook, eval setup. No sprint, no big bang.
Week sprint, n8n plus Slack approve. First run live in two weeks.
Relationship beats tooling. Keep the human, possibly a briefing agent in layer B.
Fix data or SOP first, then re-score. No structured data means no agent.
Four artifacts. Not a PDF with strategy advice, but a concrete backlog for the next twelve weeks.
40 to 80 entries with owner, frequency, pain, volume, current tools. Top 10 ranked by EUR per year leverage and quadrant.
L1 helicopter journey, L2 stage map per AAARRR stage, L3 BPMN detail for the top 15 processes. Every top-10 row in the sheet links directly to its L3 frame.
Per prioritized process: recommended phase (pilot, hardening, rollout), fixed price, success criterion in writing. You decide phase by phase, no big-bang commitment.
German DPA template plus a briefing note for your data protection officer. The GDPR path is open from day one, not week eight.
I see these eight mistakes in almost every audit. If you have one running right now, say so in the intro call. We start with the hammer test, not with tool choice.
Big-bang RPA initiative without journey map. Result: tooling for processes nobody needs anymore.
Six-month consulting study. Something must run after four weeks, otherwise the sponsor loses the mandate.
Most expensive problem first instead of most frequent and easiest. You burn the goodwill on the first use case.
Agents without escalation path and audit log. The first error and the whole setup is on trial.
Tool choice before problem choice. Whoever starts with n8n or Tasklet bends the process to the tool, not the other way around.
Black-box agents in customer-facing flows without human in the loop. One press incident stops the program.
Optimizing what should be killed. Hammer test first: do we still need this process at all.
Tooling without eval setup. You do not know if it gets better or worse, only that it runs.
Five-day audit. Fixed price from 4,800 EUR. Output is the four artifacts above, not a PDF. DPA template and GDPR briefing note attached on day one.