Case studies, AI for business process optimisation

Twenty AI implementations by sector, with KPIs and a measurement plan.

Which strategic AI implementations are most effective for business process optimisation? Below are three to five concrete production implementations per sector, delivered by The Automation Group, the AI software company of Job van den Berg. Each one lists the process goal, the KPIs you steer on and when you measure.

Request a proposal View consultancy

Short answer: five patterns that work most often

Across these twenty programmes, five implementation patterns keep coming back. They share one trait: a bounded process with high volume, a clear owner and human final control.

1. Knowledge access with citations

An assistant answering from your own knowledge base or dataset, always with a source. Steer on handling time, traceability and answer consistency.

2. Classification and routing of inbound work

Automatically labelling and assigning mail, requests and applications. Steer on time to assignment, classification accuracy and capacity freed.

3. Document processing into ERP and data lake

Reading, validating and forwarding invoices, orders and price lists. Steer on straight-through processing, import errors and processing time per document.

4. Draft generation inside the source system

Preparing emails, reports and contracts as drafts within the system people already work in. Steer on lead time and the share of drafts sent unchanged.

5. Agents for research and internal knowledge

Continuous monitoring, filtering and summarising. Steer on research lead time and adoption, not on FTEs saved.

Insurance, member service and customer contact

Processes with high inbound volume, case files and standard correspondence. The gain sits in shorter handling time and consistent answers, not in replacing the employee.

Process: Knowledge access in member service

ANWB

A knowledge assistant that gives agents the right answer from the knowledge base straight away, including citations.

Goal

Replace searching with one sourced answer so members consistently hear the same thing.

KPIs

Average handling time per contact, Telephony and ticketing system, weekly average
Share of answers with a citation, Assistant logging
Answer consistency across agents, Sample of 50 tickets per month by a quality coach

Measurement

Baseline in week 0, first measurement after 4 weeks, monthly thereafter

View the case

Process: Customer correspondence from the case file

InShared

A writing assistant inside the case system drafts emails with the relevant customer data and guards tone of voice. The employee reviews, edits and sends.

Goal

Move the employee from writing to reviewing, within one system.

KPIs

Lead time per outgoing email, Timestamp from opening the file to sending
Share of drafts sent unchanged, Diff between draft and sent text
Deviations from tone of voice, Monthly text review on a sample

Measurement

Baseline in week 0, weekly during the pilot, monthly after go-live

View the case

Process: Email classification and routing

Meijers Verzekeringen

Incoming mail is classified and routed to the right department inside the company's own cloud environment, tested step by step and rolled out in phases.

Goal

Remove manual inbox sorting and shorten lead time.

KPIs

Capacity freed on sorting work, Half an FTE in this programme, measured through time tracking
Classification accuracy, Weekly sample against manual labels
Time to first response, Mail platform, per department

Measurement

Baseline in week 0, daily checks during testing, weekly until week 12

View the case

Retail, e-commerce and franchise

Repeat questions, many locations and many disconnected systems. AI pays off fastest on customer questions, inbox handling and knowledge sharing between sites.

Process: Customer questions in the webshop

Kitchen 247

A chatbot answering customer questions 24/7 from current product knowledge. Every new version is tested automatically before it goes live.

Goal

Answer customers instantly and relieve support, without quality risk on updates.

KPIs

Share of conversations without handover to an agent, Chat log, weekly figure
Number of support tickets, Ticketing system, compared with the same period last year
Successful deployments without rollback, Deployment pipeline test report

Measurement

Baseline in week 0, weekly in quarter 1, monthly thereafter

View the case

Process: Orientation phase and showroom conversion

Kuchenwelt

A FAQ chatbot answering 125 specific questions and guiding orienting customers towards a showroom visit, running on a fixed CI/CD stack.

Goal

Answer questions instantly while generating more showroom appointments.

KPIs

Coverage of the FAQ set, Share of questions answered within the 125 defined topics
Showroom appointments from chat, Booking system with source tracking
Lead time from new question to live answer, Date reported to date deployed

Measurement

Baseline in week 0, every two weeks until week 12, quarterly thereafter

View the case

Process: Inbox handling with client-managed templates

Hizi Hair and Kinki Kappers

Incoming mail is read and labelled, standard questions get an immediate draft reply. Templates live in an Excel file the client maintains itself.

Goal

A clean inbox where the team manages its own answers, without a developer.

KPIs

Emails still requiring manual sorting, Mail platform, daily figure
Share of standard questions with an automatic draft, Logging per template
Template changes without IT involvement, Version history of the Excel file

Measurement

Baseline in week 0, weekly until week 8, monthly thereafter

View the case

Process: Learning, onboarding and development reviews

AMI Kappers

One platform combining e-learning, onboarding and development reviews, connected to HR data, with dashboards per role and a learning assistant.

Goal

Make knowledge centrally available and align onboarding across all locations.

KPIs

Onboarding lead time per new hire, HR system, monthly average
Completion rate of mandatory modules, Platform dashboard per role
Use of the learning assistant, Weekly active users relative to total headcount

Measurement

Baseline in week 0, monthly during the first year

View the case

Manufacturing, agriculture and supply chain

Document flows, orders and shared inboxes. Here it is about error-free processing, ERP integration and keeping data inside your own environment.

Process: Purchase invoices and call-off orders into ERP and data lake

The Onion Group

Invoices and orders are read, validated and pushed through a secure connection into the ERP and Microsoft Fabric.

Goal

Eliminate manual entry and achieve error-free import into the ERP.

KPIs

Share of documents processed fully automatically, Processing log per document type
Import errors in the ERP, ERP log, weekly figure
Processing time per invoice, Timestamp from receipt to posting

Measurement

Baseline in week 0, weekly until week 12, monthly thereafter

View the case

Process: From shared inbox to internal ticketing

Monarch

The shared Outlook inbox was replaced by a locally hosted ticketing system that estimates urgency and assigns mail to the right colleague. Data stays internal.

Goal

Remove manual sorting and get a live overview of open tasks.

KPIs

Time to assignment of a request, Ticketing system, weekly median
Correct urgency estimation, Sample of 30 tickets per month
Requests without an owner, Daily dashboard check

Measurement

Baseline in week 0, daily in the first month, weekly thereafter

View the case

Process: Leads, data and invoices in one portal

4Everyware

Separate pipelines for leads, data and invoices, connected to the existing CRM. Staff work from one central portal instead of separate systems.

Goal

Reduce daily manual work and create one extensible working environment.

KPIs

Manual actions per lead and per invoice, Process measurement before and after, per pipeline
Systems opened per employee per day, Sample across one working week
Lead time from lead to CRM record, CRM timestamps

Measurement

Baseline in week 0, measured per pipeline at delivery, monthly thereafter

View the case

Professional and legal services

Knowledge-intensive work with high demands on explainability and compliance. KPIs here are as much about traceability as about time saved.

Process: Asking questions of your data in natural language

Wolters Kluwer

A semantic layer ensures a plain-language question is answered from the right, governed dataset, explainable and traceable.

Goal

Let business users work with data independently and reduce ad-hoc requests to the data team.

KPIs

Ad-hoc requests to the data team, Ticketing system, monthly figure
Share of answers traceable to the source dataset, Audit log of the semantic layer
Time from question to substantiated answer, Sample of 20 questions per month

Measurement

Baseline in week 0, monthly from the pilot onwards, quarterly governance review

View the case

Process: Filling M&A purchase agreements with chamber of commerce data

Act Legal

Existing software and a custom plugin were benchmarked side by side on automatically filling contracts with registry data, producing a substantiated choice.

Goal

Save hours of typing without giving up precision in legal documents.

KPIs

Hours saved per contract, Time tracking before and after
Field-level accuracy of the filled data, Every field checked against the registry extract
Corrections per document, Review log of the responsible lawyer

Measurement

Comparative measurement during testing, then a quality check every 25 contracts

View the case

Process: Meeting intelligence inside the EU

Based

Meeting audio is transcribed and structured into reports. The whole solution runs on EU infrastructure and is GDPR compliant.

Goal

Retain and search knowledge from conversations within your own compliance boundaries.

KPIs

Share of meetings with a structured report, Platform reporting, weekly
Time to available report after the meeting, Timestamp from meeting end to publication
Processing exclusively on EU infrastructure, Quarterly check by the privacy owner

Measurement

Baseline in week 0, weekly in the pilot, monthly with a quarterly audit

View the case

Process: Absence reports from Excel to Word

Van Campen Consulting

Excel data is converted automatically into a finished report. Two collaborating models guard the methodology and the writing style.

Goal

Reports in minutes instead of hours, at the same quality for every consultant.

KPIs

Lead time per report, Time tracking per assignment
Methodological deviations per report, Review checklist of the subject-matter reviewer
Style consistency across consultants, Monthly sample of five reports

Measurement

Baseline in week 0, logged per report, monthly quality review

View the case

Real estate, care and HR processes

Processes with many requests, files and people. Human final control is a deliberate design choice here, and therefore also a KPI.

Process: Day planning and tenant requests during an ERP migration

Green Real Estate

An automated meeting-room day plan in SharePoint and a flow turning tenant requests into structured work orders for Yardi, built during the ongoing migration.

Goal

Save time on planning and requests without delaying the ERP migration.

KPIs

Time spent creating the day plan, Time tracking of the facility team
Share of requests arriving as complete work orders, Yardi reporting, weekly
Migration disruptions caused by the new flows, Incident register during migration

Measurement

Baseline in week 0, weekly during migration, monthly thereafter

View the case

Process: Tailored AI skills per employee

Van Deursen Group

In six weeks, AI skills were built per employee matching their own work, from analysing quarterly reports to calculating housing valuation points.

Goal

Reduce repetitive work with direct time savings and broad internal adoption.

KPIs

Active use per employee, Weekly active users relative to headcount
Time saved per skill, Before and after measurement on that task
Skills still in use after six weeks, Measured in weeks 6, 12 and 26

Measurement

Baseline in week 0, delivery in week 6, adoption check in weeks 12 and 26

View the case

Process: Preparing recruitment procedures

Domus Valuas

Incoming applications are analysed, attachments processed and a forwarding email is prepared as a draft. HR decides when it is sent.

Goal

Faster administrative processing while HR keeps full final control.

KPIs

Processing time per application, Timestamp from receipt to draft email
Completeness of processed attachments, Sample of 20 applications per month
Share of emails sent only after HR approval, Send log, must be 100 percent

Measurement

Baseline in week 0, weekly until week 8, monthly thereafter

View the case

Internal knowledge, marketing and innovation

Processes where the outcome is not time saved but knowledge advantage. Measure research lead time and adoption, not FTEs.

Process: Research and content for the marketing team

AFAS

An internal platform where multiple agents collaborate on research and content, set up as an open framework so the team can extend it themselves.

Goal

Automate research and leave behind a framework the team can extend without a vendor.

KPIs

Lead time from research question to usable draft, Project registration per assignment
Agents added by the team itself, Repository history per quarter
Share of content passing through the framework, Content calendar with source tracking

Measurement

Baseline in week 0, monthly, quarterly review on independent extension

View the case

Process: Internal research agent for the field

OpenClaw

An agent continuously monitors new frameworks and models, filters relevant developments and shares only what adds value with the team.

Goal

Keep knowledge current and adopt new technology faster.

KPIs

Time between publication and internal assessment, Agent log
Signal-to-noise ratio of shared items, Weekly team relevance score
Techniques adopted per quarter, Quarterly review with the engineering team

Measurement

Weekly log, quarterly adoption review

View the case

Process: Automation scan across four departments

Beachclub Indigo

Working sessions with administration, sales and marketing mapped the bottlenecks and translated them into a roadmap with three levels of automation.

Goal

A directly executable roadmap with quick wins per department and gradual implementation.

KPIs

Prioritised bottlenecks with an owner, Roadmap with assigned owner
Lead time from advice to first implementation, Start date of the first measure
Steps completed per quarter, Progress review per department

Measurement

Zero measurement in the working sessions, quarterly progress review

View the case

Measurement plan

An implementation without a measurement plan turns into a debate about feelings. This is the rhythm used in these programmes, from baseline to quarterly review.

Moment	What you measure	Owner	Decision
Week -2 to 0, baseline	Current lead time, volume, error rate and time spent per process step.	Process owner with a data analyst	Set baseline and target value per KPI.
Week 0, pilot go-live	Technical readiness, use-case coverage and logging in place.	Implementation team	Go or no-go for the pilot group.
Week 1 to 4, weekly	Accuracy, number of corrections, failures and user feedback.	Process owner	Adjust prompts, rules and thresholds.
Week 6, interim review	First effect on lead time and quality against the baseline.	Process owner with executive sponsor	Scale, adjust or stop.
Week 12, impact measurement	All KPIs against baseline, plus adoption per team and actual cost.	Executive sponsor	Roll out to the next department or site.
Week 26 onwards, quarterly review	Sustained use, model drift, data quality and compliance checks.	Data governance and process owner	Extend, recalibrate or retire.

Source of the implementations: The Automation Group, theautomationgroup.nl/nl/cases

Frequently asked questions

Which strategic AI implementations are most effective for business process optimisation?

In practice five patterns deliver fastest: knowledge access with citations, classification and routing of inbound work, document processing into ERP, draft generation inside the source system, and agents for research and internal knowledge. They work because the process is bounded, volume is high and human final control is preserved.

Which KPIs belong to an AI implementation in a business process?

Always combine three types: a time KPI such as lead or handling time, a quality KPI such as accuracy or number of corrections, and an adoption KPI such as active use per team. Regulated processes add a traceability KPI, for example the share of answers with a citation.

How long before an AI implementation shows measurable results?

A bounded process shows a first effect within 6 weeks and a reliable impact measurement against baseline within 12 weeks. Programmes without numbers after six months usually skipped the baseline.

Why do AI projects in business processes fail?

Usually not because of the models, but because of a missing data foundation, a process without an owner and no baseline measurement. Without a baseline you cannot prove afterwards what the implementation delivered.

Where do these case studies come from?

All implementations were delivered by The Automation Group, the AI software company of Job van den Berg. The cases are also published on theautomationgroup.nl.