Back to overview

Explore AI Workshop - Client Examples

Where should AI actually start?

Most organisations already have AI ideas. The harder question is which operational problems are worth solving, what the evidence supports, and where investment should start. These two anonymised cases show how COMPUTD turns that discussion into concrete use cases, priorities and a defensible next decision.

Client example seriesManufacturing & operationsEnglish edition
From AI interest to a decisionThe workshop brings management, operations and technical specialists into one structured process: identify the problems that matter, test where AI could help, expose dependencies and decide what deserves validation first.

01 - Introduction

AI becomes useful when the organisation can choose where to focus

The two examples in this document show what that choice can look like in practice. In the first, a cross-functional team moved from day-to-day operational problems to four concrete AI opportunities. In the second, the team already had a broader set of issues and needed to determine which opportunity deserved investigation first.

The workshop creates a shared basis for that decision. It connects operational pain, expected value, available data and organisational readiness before a company commits to a build. The result can be a shortlist of opportunities, a first validation target, or a sequence of actions when the underlying process and data need attention first.

Client example 01

Opportunity Discovery

From operational challenges to four concrete AI use cases in planning, work instructions, quality and knowledge access.

Client example 02

Prioritisation & Roadmap

From a broader problem set to a first validation target, a decision sequence and clear foundation work.

01
02 - Client Example 01 - Opportunity Discovery

From operational challenges to AI opportunities

A cross-functional industrial team used the workshop to turn recognisable problems into four practical AI use cases and compare them on expected value and readiness.

Anonymised external example

02.1 - Workshop Context & Outcome

Four use cases emerged from a problem-first workshop

Engineers and operations specialists examined where AI could support production planning, knowledge retention, quality assurance and access to technical information.

Participants started with problems from their own work and from adjacent departments. They expressed these as user stories, selected the strongest themes, developed them in a Use Case Canvas and connected them to relevant AI capabilities.

Use case 01

Smart production planning

Use demand, capacity, staff availability and energy information to support more realistic planning decisions.

Use case 02

Automated work instructions

Capture operational knowledge and turn it into clear, current instructions for production and onboarding.

Use case 03

AI-supported final inspection

Detect deviations, assess products against defined criteria and learn from recurring quality patterns.

Use case 04

Interactive knowledge assistant

Give employees natural-language access to internal technical and operational documentation.

Scope of this client example: organisation-specific names, participant details, contact information and workshop photos were removed or generalised. The workshop method, use-case patterns and AI capabilities were retained.

02.2 - Workshop Approach

Start with the problem, then decide where AI is relevant

The workshop combined individual observation, group selection, structured use-case development and plenary assessment.

01

Surface challenges

Participants describe concrete problems from daily work and from broader operational processes.

02

Select

Teams compare the stories and choose the challenges with the strongest relevance or potential.

03

Develop

Selected problems are structured in a Use Case Canvas and connected to relevant AI capabilities.

04

Assess

The group discusses expected impact and how ready the organisation is to pursue each direction.

Working principle

Start with a recognisable operational problem, express it clearly, then identify the AI capability that could improve the decision or workflow.

The AI Superpower Cards acted as a bridge between business challenges and possible AI functionality. They gave participants concrete examples of what AI can do without forcing the discussion toward a specific technology vendor or product.

02.3 - Impact & Readiness

Prioritise value and readiness, then validate the assumptions

The team assessed each use case on expected business impact and current readiness, considering culture, data and infrastructure.

Impact within one year - low to high
Work instructions
Knowledge assistant
Final inspection
Planning
Ability to exploit today - high to low
Interpretation: the positions are indicative and reproduce the evaluation principle in anonymised form. They do not represent a technical feasibility study or a quantified business case.

What the workshop establishes

A shared view of the operational problem, intended users, expected value and the AI capabilities that may be relevant.

What still needs evidence

Data quality, technical feasibility, baseline performance, implementation effort, ownership, change requirements and the business case for a specific solution.

02.4 - Use Case

Smart production planning

An intelligent planning concept combines demand, capacity, staff availability and energy information so planners can make better decisions and steer production more realistically.

Planning logic

Prioritise what should happen next

Prioritisation and volume forecasting can support decisions about which jobs should run, when they should run and which resources are required.

PrioritisationVolume forecastingDashboarding
Potential operational value

Capacity, cost and delivery reliability

Better alignment can reduce avoidable downtime, excess capacity and understaffing. Energy-intensive production may also be shifted toward periods when energy and machine capacity are more favourable.

Why this case matters

Demand and available resources change continuously in a production environment. A planning approach based on current operational data can improve resource use and delivery predictability. For sales, that can create more confidence when responding to rush orders or non-standard delivery requirements.

The workshop established the relevance of the problem. A next phase would still need to determine which planning decisions can be improved, whether the required data exists and whether optimisation produces a measurable improvement over current planning practice.

02.5 - Use Case

Automated work instructions

The concept captures operational knowledge, structures it and turns it into current work instructions that employees can use during production and onboarding.

Problem

Knowledge becomes fragmented or disappears

Important information is not always documented in time, and experienced employees cannot always transfer their expertise in a structured way.

Working concept

Capture, process and publish

Information can be collected and converted into reusable work instructions. Video recordings of experienced employees can supplement written instructions with practical demonstrations.

Speech to textData and text analysisLLM

Used well, the approach can support onboarding and continuity. New and less experienced employees gain faster access to practical knowledge, while the organisation becomes less dependent on individual knowledge holders. The first validation question is whether structured documentation itself solves most of the problem before introducing more advanced AI retrieval or generation.

02.6 - Use Case

AI-supported final inspection

AI can support quality control by checking products against defined criteria, identifying deviations earlier and revealing recurring patterns in quality data.

Capabilities

Automated inspection

Products can be assessed against predefined standards. Deviations can be detected and classified consistently when the data and inspection method support it.

Automated qualificationAnomaly detectionClassification
Insight

Move from inspection to quality learning

Data analysis can reveal trends and recurring causes. Dashboards can give operators and quality managers a current view of status and exceptions.

Data analysisDashboarding

Traditional final inspection often depends on manual and visual checks. Automation may improve consistency and reduce subjectivity, but the more important long-term value comes from learning systematically from deviations and process data. A validation phase should first establish which defects matter, how reliably they can be observed and what action follows detection.

02.7 - Use Case

Interactive knowledge assistant

A knowledge assistant lets employees ask natural-language questions about internal technical and operational documentation and brings relevant information forward faster.

Function

Ask instead of search

Employees ask a targeted question and receive relevant information from internal documentation without specialist search skills.

AI capabilities

Structure and qualify information

Clustering, classification, data and text analysis, and automated qualification can help organise information and improve the relevance of retrieved results.

LLMClusteringClassificationText analysis

The value comes from reducing search time, reaching well-founded answers faster and solving technical problems with the right information at hand. The practical prerequisite is a reliable document base with clear ownership and update processes.

02.8 - AI Superpowers

AI Superpowers used in the workshop

These capabilities helped participants make business problems concrete and identify which type of AI functionality could be relevant.

AI capabilityDescriptionProduction example
Detect anomaliesDetect data points that deviate from the normal pattern.Flag an unexpected temperature rise in a machine component.
Intent recognitionUnderstand the intent behind a question, message or document.Infer whether a customer request concerns a quotation, complaint or order confirmation.
Volume forecastingForecast future quantities from historical and current data.Align production and resources with expected demand.
ClusteringGroup data or objects based on similarities.Group machines with similar failure patterns.
Speech to text / text to speechConvert spoken audio to text and written text to speech.Capture operational knowledge through spoken input.
Dataset completionFill missing values or information in datasets.Fill gaps in sensor measurements to provide a more complete operational picture.
PrioritisationRank tasks or problems based on impact and urgency.Identify which maintenance or production tasks require attention first.
Process analysisAnalyse digital traces to reveal processes, delays and improvement opportunities.Identify bottlenecks in a production line.
Compliance monitoringCheck whether processes and records comply with defined rules.Flag missing or deviating process steps.
DashboardingPresent data in clear charts and key indicators.Show production, energy use, capacity and disruptions at a glance.
Automated qualificationDetermine automatically whether something meets predefined criteria.Assess products, suppliers or documents against defined standards.
ClassificationSort data into predefined categories.Classify components in images as approved or rejected.
Data and text analysisSearch large volumes of information for patterns and insights.Extract recurring problems from feedback, reports and documentation.
Large language modelUnderstand and generate natural language.Retrieve information from complex documentation through natural-language questions.

02.9 - Recommendation & Next Step

A practical first step toward targeted AI adoption

The workshop produced four use cases that merit a clear next step. Before investing in delivery, each case should be tested for business value, data availability, process ownership, technical feasibility and implementation risk.

The method creates a useful sequence: start with the operational problem, define the intended value, test feasibility and only then select the required AI capabilities. That gives the organisation a concrete next move rather than a general AI ambition.

Recommended next decision

Select one or two use cases for a short validation phase with clear success criteria, a named owner, an explicit operational question and the data required to answer it.

02
03 - Client Example 02 - Prioritisation & Roadmap

From AI opportunities to an actionable validation roadmap

A manufacturing team identified recurring problems in tool management, production planning, forecasting, data quality, knowledge transfer and commercial targeting, then used the workshop to choose what should be investigated first.

Anonymised external example

03.1 - Workshop Context & Outcome

A concrete first priority emerged, but validation still has to earn the investment

The workshop brought together management, engineering, CAM, automation, finance/IT and work preparation to identify operational problems and examine where AI or related data-driven technology could help.

The discussion surfaced recurring issues in tool management, production planning, time estimation, pricing, data quality, knowledge transfer, technology adoption, ownership and cross-functional decision-making. A common pattern sat underneath several of these issues: operational data existed across different systems and informal sources, but teams could not consistently access, trust or combine it when making decisions.

Participants translated selected problems into four concepts: intelligent tool management, planning intelligence, operational knowledge management and a product-customer recommender. Workshop voting placed tool management first, followed by planning. That ranking reflects participant perception of impact, feasibility and organisational readiness. It should guide what to investigate first, rather than be treated as proof that implementation is technically or economically viable.

Recommended direction

Validate tool management first. In parallel, map the planning process and begin capturing critical operational knowledge where it is concentrated in individuals. Use the evidence from those activities to determine which larger investments are justified.

03.2 - Key Problem Areas

The workshop exposed both technology opportunities and management issues

Some themes are suitable for data or AI intervention. Others concern process, ownership or organisational behaviour and need management action alongside any technical solution.

ThemeAssessment
Tool management & data gapsTeams lacked dependable visibility into tool availability, usage and remaining life. Relevant information was incomplete or distributed across systems and manual records, limiting planning and automation.
Production planningPlanning was frequently reactive and involved detailed intervention from multiple stakeholders. Last-minute changes, machine loading and delivery commitments were difficult to reconcile consistently.
Order & process-time forecastingMachining and processing-time estimates were difficult because product, material, process and tool-performance data were incomplete or inconsistent. Manual estimation remained important.
Pricing & calculationQuotation and costing logic did not consistently reflect product complexity, creating slow calculations and avoidable errors. This is primarily a process and model-quality issue before it becomes an AI issue.
Quality & process reliabilityVariation in inputs and production execution contributed to defects, rework and low trust in system data. Better data capture and process discipline are prerequisites for predictive use cases.
Customer targetingThe company lacked a structured way to connect production capabilities and product characteristics with attractive customers or markets. The opportunity depends on the quality of product, customer and market data.
Knowledge managementImportant operational knowledge remained informal and person-dependent, making onboarding, standardisation and consistent execution harder.
Technology adoptionLimited time, unclear ownership and reliance on familiar practices slowed adoption. New tools need clear owners, visible benefits and involvement from the people who use them.
Collaboration, decisions & ownershipUnclear roles and decision ownership led to repeated discussions, broad meeting participation and weak follow-through. Technology cannot resolve this without explicit governance and accountability.

How the focus areas were selected

Participants filtered the wider problem set toward issues that appeared data-driven, repetitive and scalable. One group concentrated on commercial, financial and planning questions, leading to the product-customer recommender and planning concept. The other concentrated on tool management, operational knowledge and adoption barriers, leading to tool management and knowledge-management concepts.

03.3 - AI Opportunities

The team translated the problems into four connected use cases

Each concept addresses a real problem, but each also carries a different evidence burden. The workshop helped decide what should be investigated first.

Use caseProblem to solveWorking conceptWhat must be validated
Intelligent tool managementInsufficient visibility into tool availability, usage and remaining life.Combine ERP, machine/runtime and tool data to provide reliable status and, where data supports it, forecast tool requirements or remaining life.Data availability and identifiers, quality, integration effort, user decisions, and whether prediction adds value beyond a reliable dashboard.
Planning intelligenceReactive planning and limited predictability of capacity, lead times and disruptions.Improve planning visibility first, then test process-mining and predictive methods where historical data can support them.The actual planning workflow, quality of lead-time, BOM, routing and order-history data, causes of schedule changes, and achievable decision improvements.
Operational knowledge managementCritical production knowledge is informal and difficult to reuse consistently.Capture and structure high-value operational knowledge so employees can find and apply it at the point of work.Which knowledge creates the most risk when unavailable, content ownership, update process, user search behaviour, and whether AI retrieval is needed.
Product-customer recommenderSales lacks a structured way to identify promising product-customer combinations.Use product and customer data, potentially enriched with external information, to rank commercial opportunities.CRM and product-data completeness, definition of a good recommendation, availability and legality of external data, and measurable uplift over current sales selection.
Workshop priority: tool management received the strongest support, followed by planning. This is a useful ordering for discovery because the first problem is concrete and recognised across functions. The ranking does not yet establish ROI or technical feasibility.

03.4 - Impact & Readiness Assessment

Use participant voting to choose the first validation target

Participants scored the concepts on business impact, technical feasibility and organisational readiness. Tool management received the strongest combined score. Planning scored highly on impact but lower on feasibility and readiness, while the commercial recommender and knowledge-management concepts ranked behind them.

Indicative order from the workshop

01Intelligent tool managementHighest shared priority
02Planning intelligenceHigh impact, more foundation work
03Product-customer recommenderDepends on stronger data
04Knowledge managementStart as a practice

How to interpret the score

The exercise captured informed internal judgement. It did not verify technical readiness. A separate review still needs to examine source systems, identifiers, interfaces, historical coverage, security, data quality, ownership and available capacity.

Use the workshop score to decide what to investigate first, not to commit to a full implementation timeline or expected financial return.

03.5 - Recommended Sequence

Turn the workshop priority into a sequence of decisions

The strongest recommendation is to validate the first operational problem before turning the four concepts into a large transformation programme.

StageFocusDecision to earn
1. Validate tool managementInventory tool-related data, map the decisions affected by tool availability, establish a baseline and build the smallest useful visibility prototype.Is the data sufficiently reliable, and does improved visibility reduce measurable operational loss?
2. Diagnose planningMap the current planning process, sources of schedule change and required data. Use process mining only if event data is suitable.Which planning decisions can be improved, and what data or process changes are required first?
3. Capture critical knowledge continuouslyStart with high-risk, frequently repeated or person-dependent procedures. Assign content owners and a review cadence.Does structured knowledge reduce search time, variation or onboarding effort? Add AI retrieval only where it improves access.
4. Test the commercial recommender when data is readyDefine the recommendation objective, clean CRM and product data, then run a controlled ranking experiment.Do recommendations produce better qualified opportunities or conversion than the current approach?

03.6 - Validation Detail

What each next step needs to prove

1. Tool management validation

Begin with a data and process inventory covering ERP records, machine/runtime data, tool databases and spreadsheets. Define a baseline for tool-related delays, stockouts, emergency actions and downtime. Build a minimum useful view of availability and usage before committing to lifespan prediction.

A 4 to 6 week pilot can be appropriate once the required data is accessible, but the duration should follow the production cycle and the amount of evidence needed.

Candidate measures: tool availability at job start, tool-related delay or downtime, emergency tool purchases, planner or operator time spent locating or verifying tools, and prediction error if lifespan forecasting is tested. Numeric targets from the original workshop report should remain hypotheses until a baseline exists.

2. Planning validation

Map how orders move from intake to planning and production, including manual overrides and the reasons schedules change. Audit lead times, routings, BOMs, machine constraints and order history. Only then determine whether process mining, optimisation, predictive models or simpler planning-rule improvements fit the problem.

Any target for fewer last-minute changes, higher forecast accuracy or better on-time delivery should follow baseline measurement and analysis of the actual causes.

3. Product-customer recommender validation

Define what a useful recommendation means for sales: a new account, cross-sell opportunity, product-market match or another commercial outcome. Align CRM and product data around that objective. Test recommendations against historical or controlled sales outcomes before introducing external enrichment at scale.

4. Knowledge-management validation

Start knowledge capture early rather than waiting for a separate platform project. Focus on procedures where missing knowledge creates delays, errors or dependency on specific employees. A wiki or existing collaboration platform may be sufficient initially. The system needs clear content owners, review dates and usage measures before adding an AI interface.

03.7 - Integration Principles

Share foundations without forcing one large architecture

The four concepts share data and knowledge, but they do not need to become one programme before their individual value is proven.

Tool-management data may later improve planning. Planning can produce capacity information that supports commercial decisions. Knowledge management can document rules and procedures that emerge from both initiatives. These dependencies are plausible and worth designing for where inexpensive, then confirming during validation.

Practical integration principle

Standardise shared identifiers, ownership and documentation early while keeping each use case independently testable.

This reduces duplicate data work without creating a large transformation programme before the organisation knows which concepts deserve investment.

03.8 - Foundation for Execution

A small set of disciplines matters more than a generic AI-readiness programme

The organisation can start with targeted governance tied directly to the selected use cases.

Data ownership

Define the data needed for a decision

Assign owners to the data domains used in each pilot. Define minimum quality rules for the decision being supported and measure compliance. Avoid broad data-cleaning programmes without a use-case requirement.

Change & adoption

Design the new workflow with the users

Identify who makes the decision today, who will use the new output, who owns the process and what behaviour must change. Measure whether the new workflow replaces manual work or improves the decision.

Ownership

Name accountable business and technical owners

A small cross-functional group can coordinate dependencies, but it should not replace explicit ownership. Review pilots against operational outcome, adoption, data quality, technical effort and expected economics.

Immediate next step

Run a short discovery and validation sprint

Produce a documented problem baseline, source-data map, integration assessment, prototype scope, success measures, named owner and a go or no-go recommendation.

04 - What the Two Examples Show

The same workshop should not produce the same answer

Both examples use a problem-first Explore AI approach, but the outputs differ because the organisations started with different problems, data conditions and levels of decision readiness.

Client example 01

Opportunity Discovery

  • Useful when the organisation needs to understand where AI could be relevant.
  • Creates a shared shortlist of concrete use cases.
  • Helps different departments compare problems and intended value.
  • Ends with a clear choice of which use cases deserve validation.
Primary outcome: a reasoned set of AI opportunities and a clear choice of what to validate next.
Client example 02

Prioritisation & Roadmap

  • Useful when several opportunities are already visible but the order is unclear.
  • Makes dependencies in data, process and ownership explicit.
  • Turns participant judgement into a testable first priority.
  • Separates immediate validation from longer-term ideas.
Primary outcome: a prioritised first hypothesis and a practical validation path before larger investment.

05 - The Explore AI Workshop

What you can expect from the workshop

COMPUTD brings together the people who understand the work, makes the operational problems visible and helps the team decide where AI deserves attention. The workshop is designed to improve the next decision, not to produce a longer list of technologies.

Outcome 01

A shared problem view

Management, operations and technical specialists compare how the same process looks from different roles and agree on the problems that matter.

Outcome 02

Concrete AI use cases

Promising problems are translated into specific applications with an intended user, value hypothesis and relevant AI capabilities.

Outcome 03

Priorities and dependencies

The team distinguishes attractive opportunities from ideas that first require better data, clearer process ownership or further investigation.

Outcome 04

A responsible next step

The workshop ends with a defined validation question, so the organisation can test assumptions before committing to a larger build.

06 - From Workshop to Next Decision

Which operational problem should your organisation investigate first?

A COMPUTD Explore AI Workshop gives your team a structured way to answer that question together. We bring the people who understand the work into one room, make the competing problems and assumptions explicit, and leave with a small number of opportunities and a clear next decision.

Useful when AI interest is broad

Move from general ideas and individual requests to a shared view of where AI may create operational or commercial value.

Useful when there are already too many ideas

Compare opportunities, expose dependencies and decide what should be tested first instead of starting several disconnected experiments.

Start a conversation with COMPUTD

These client examples are based on existing workshop reports. Personal names, client identification and organisation-specific details have been removed or generalised. No project results have been invented.