DIAGNOSE

Quick Answer

DIAGNOSE examines the current workflow before an organization commits to an AI solution.

In the DIAGNOSE stage the team maps how work moves from the start to final output, measuring current performance, identifying delays and rework, finding bottlenecks and root causes, and checking data quality and requirements.

DIAGNOSE produces three possible outcomes:

Improve Without AI. Improve Before AI. Improve with AI augmentation.

Where is the DIAGNOSE Stage in ADVIS? 

ALIGN → DIAGNOSE → VALIDATE → IMPLEMENT → SCALE

ALIGN = Strategic Justification →
          DIAGNOSE = Diagnostic Evidence →
                    VALIDATE = Controlled Proof →
                              IMPLEMENT = Operational Proof →
                                        SCALE = Sustained & Expanded Proof

Why This Matters

Why the DIAGNOSE Stage Matters

Much of the value from AI comes from changing how work gets done.

That makes understanding workflows central to successful AI implementation.

McKinsey's 2025 global research found that redesigning workflows had the largest effect among 25 organizational practices tested on respondents' ability to report EBIT impact from generative AI. Yet only 21% of respondents using generative AI said their organizations had fundamentally redesigned at least some workflows (McKinsey & Company, 2025a). (McKinsey & Company)

More recent McKinsey research found the same pattern. Leaders were 5.3 times more likely to report enterprise value from AI when workflows were redesigned rather than left unchanged (McKinsey & Company, 2026). (McKinsey & Company)

BCG similarly reports that organizations capturing greater AI value are moving beyond tool deployment and redesigning workflows end-to-end (Boston Consulting Group, 2025a). (BCG Global)

The lesson for executives is important:

AI value depends on understanding the workflow before redesigning the work.

DIAGNOSE prevents firms from automating the wrong problem.

Consider a professional services firm where producing a major client proposal takes five days.

Management may initially conclude that proposal writing is too slow.

That observation describes the symptom. The actual causes might include:

• Information scattered across multiple systems
• Slow expert approvals
• Poor reuse of previous work
• Inconsistent templates
• Unclear responsibilities
• Repetitive drafting
• Missing client information
• Too many handoffs

AI could help with some of these causes.

Others may require a process change, clearer ownership, better information management, or conventional automation.

An AI workflow assessment separates the visible problem from the conditions creating it.

DIAGNOSE improves investment decisions by separating AI from the problems.

There is a danger in knowing that AI is a potential solution.

Experimenting and demoing potential AI solutions is fun, exciting, and potentially wrong.

The ease of creating an AI demonstration can make weak opportunities look attractive.

DIAGNOSE identifies problems that should be fixed first.

Some workflows are not ready for AI.

The underlying process may be inconsistent.

Data may be unreliable.

Knowledge may exist only in the heads of a few experts.

Roles may be unclear.

A workflow may contain unnecessary approvals or duplicate work.

Faster may not be better

Adding AI under those conditions can make a poor process faster without making it better.

A diagnosis identifies those weaknesses before they become AI implementation problems.

This is especially important when applying AI to professional services.

Consulting, accounting, marketing, legal, financial, advisory, and other professional work often combines repeatable tasks with expert judgment.

A workflow analysis should identify:

• Where professional evaluation is required
• Where approval is necessary
• Which exceptions need escalation
• Where client context changes the answer
• Who remains accountable for the result
• Where unnecessary work is performed
• Where communications slow the process and more

The goal is not to remove people wherever possible.

The goal is to design a better system that combines AI efficiency with human expertise.

What Successful Firms Do

A successful DIAGNOSE stage gives leadership a credible picture of how the work performs today and what should be changed.

The work team doing the DIAGNOSE stage should understand the workflow well enough to explain where value is being lost, where bottlenecks are, and why there are slow downs.

The team working on this workflow should know:

• What triggers the work
• What information enters the process
• Which tasks are performed
• Where decisions are made
• Which systems are used
• Who performs the work
• Where handoffs occur
• What outputs are produced
• What requirements are unnecessary
• Where workflow chokepoints are and their severity

The team should be able to identify where performance is gained or lost.

Current performance is measurable

The team has usable evidence about current performance.

Depending on the workflow, their evidence might include:

• Cycle time
• Costs
• Error rates
• Rework
• Quality
• Professional hours
• Throughput
• Client response time
• Conversion
• Capacity

Metrics established in ALIGN should be cross checked in the DIAGNOSE stage.

The original assumptions may prove incomplete once the workflow is examined.

Bottlenecks and root causes and their severity are identified

The team diagnosing the workflow should be able to identify the problem and its causes. For example, if proposal turnaround is slow, the diagnosis should identify why. If client response time is poor, the team should know where the delay occurs.

This is one of the most important outputs of DIAGNOSE. Understanding where bottlenecks occur that slow the work and identifying the impact level of each bottleneck.

In any workflow there will be only one bottleneck at a time that throttles the work through the entire workflow. You must find each of those bottlenecks in turn, working from most critical to less critical, and use process improvement or AI systems to open each bottleneck.

Data and knowledge requirements are understood

The organization knows what data and information the workflow requires and whether it is:

• Accurate
• Available
• Accessible
• Reliable
• Current
• Relevant
• Appropriately secured

For knowledge-intensive professional work, DIAGNOSE also examines where expertise resides and how consistently it can be accessed.

Human judgment is visible

The future AI-enabled workflow should not be designed until the organization understands where human judgment, approval, discretion, client interaction, and accountability are required.

Those points later become part of the human-in-the-loop design.

DIAGNOSE should focus the scope of AI

The DIAGNOSE team has moved beyond a general idea such as:

“AI could improve proposal development.”

It now has a more specific implementation such as:

“AI may reduce proposal cycle time by retrieving approved prior work, creating a first draft from defined sources, and routing exceptions to the appropriate subject-matter expert.”

That is specific enough to move toward VALIDATE.

What Causes Failure

Automating waste

A poor workflow doesn't become a good workflow just because AI does part of it faster.

Unnecessary steps, duplicate reviews, confusing responsibilities, and poor information flows should be addressed before they are automated.

BCG makes the same point in its work on AI agents: process redesign must address the logic of the process itself, including unnecessary steps, roles, knowledge, systems, and data quality (Boston Consulting Group, 2025b). (BCG Global)

Confusing symptoms with root causes

“Proposal development is slow” is a problem statement, not a root cause.

The underlying cause could be writing time, research, approvals, information retrieval, data quality, expert capacity, or workflow design.

Different causes require different solutions.

Selecting easy AI use cases instead of valuable ones

Some AI applications are easy to demonstrate.

That does not make them the best place to invest.

A useful opportunity should combine strategic impact with realistic feasibility, manageable risk, and sufficient capability.

Ignoring data and knowledge readiness

AI results depend heavily on the information available to the system.

If critical knowledge is inaccessible, inconsistent, outdated, or poorly structured, the AI solution may never perform acceptably.

Ignoring human judgment

A workflow can appear highly repetitive when processes are normal, but exceptions and outliers cause slowdowns and bottlenecks.

Professional judgment may be required for unusual clients, incomplete information, regulatory issues, quality control, or decisions carrying business consequences.

Teams must identify where these outliers and exceptions are most likely to occur and then use human-in-the-loop interventions and judgment.

In some cases, especially with newly implemented AI systems, the human-in-the-loop review can take up the time saved by the AI automation. In these cases, the human managers of the system must include an improvement loop to examine delays and modify the AI system to handle those delays as well as the original system.

Designing around one AI tool

AI platforms, models, agents, and capabilities are changing rapidly. They are all approaching an asymptote of the same power, performance, and capabilities.

A diagnosis should define what the workflow needs before the organization locks itself into a particular product.

Keeping AI platform decisions open preserves more options during VALIDATE and IMPLEMENT.

How to Do the Diagnose Stage

DIAGNOSE begins with the workflow(s) selected through ALIGN.

The work should then be examined from several perspectives.

1. Map the current workflow

Start where the work actually begins.

Document:

Trigger — What causes the workflow to start?

Inputs — What information is required?

Tasks — What work is performed?

Decisions — Where are choices or approvals required?

People — Which roles participate?

Systems — What applications or tools are involved?

Handoffs — Where does work move between people or systems?

Outputs — What must the workflow produce?

For many professional workflows, a simple current-state map provides enough visibility to start.

2. Measure current performance

Determine how the workflow performs today.

Use existing data when it is reliable.

If measurement is weak, collect enough information to understand the current state before assuming potential improvements.

This may also reveal that the metric or baseline established during ALIGN needs adjustment.

3. Find friction

Look for places where value is lost.

Common examples include:

• Waiting
• Unnecessary reviews
• Rework
• Errors
• Repetitive research
• Duplicate data entry
• Searching for information
• Inconsistent outputs
• Expert bottlenecks
• Manual transfer between systems

These become candidates for deeper root-cause analysis.

4. Determine root causes

Ask why the bottleneck or friction occurs.

Useful categories to examine include:

Process — Is the workflow poorly designed?

Policy — Do rules or approvals create unnecessary work?

Data — Is required information missing or unreliable?

Knowledge — Is expertise difficult to access or reuse?

Capability — Do people lack required skills?

Capacity — Are scarce experts or resources overloaded?

System — Do applications fail to support the workflow?

Technology — Is an important capability unavailable?

Do not assume the answer to these causes will be AI.

 IMAGE DESCRIPTION — PHOTOREALIST

Photorealist: A mixed-race, mixed-gender group of six professional services employees in casual business clothes mapping a real workflow in a bright conference room. Include a partner or department leader, operations manager, experienced professional, junior professional, and facilitator. A large whiteboard shows a simple current-state process with sticky notes connected by arrows. Several notes identify Delay, Rework, and Expert Review. Participants are actively discussing the workflow rather than posing for the camera. Natural daylight, realistic office setting, laptops visible but secondary, no robots, no futuristic AI effects.

Image Caption: Effective AI workflow analysis begins with the people who understand how the work actually happens.

5. Examine data and knowledge

Determine what information each important task requires.

Ask:

• Where does the information come from?
• Can the right people or systems access it?
• How reliable is it?
• Is it current?
• Is it structured or unstructured?
• Is confidential information involved?
• Are there regulatory or client restrictions?
• Does critical knowledge exist only with individual experts?

This step is particularly important for AI systems that retrieve, summarize, analyze, recommend, or generate work from organizational knowledge.

6. Identify where human judgment and accountability are needed

Map where professional judgment matters.

Examples include:

• Approval
• Interpretation
• Client advice
• Unusual cases
• Risk decisions
• Quality review
• Exceptions
• Escalation

Also determine who remains accountable when AI participates in the work.

Your redesign must make it clear where responsibility lies for a good AI workflow – people, technology, or AI system.

7. Review measurement assumptions

Return to the metric, baseline, and target created during ALIGN.

Examine whether the workflow diagnosis supports those measurement assumptions.

Often DIAGNOSE reveals that in addition to a primary (results) metric, one or more operational metrics are needed.

For example, the strategic target may be improved client retention, while the workflow requires measurement of response time, resolution quality, and professional review effort.

VALIDATE will later determine formal acceptance criteria.

8. Identify AI leverage points

Now examine where AI could make a meaningful difference.

Useful categories include:

Automate

Perform repeatable tasks with limited human involvement where risk is acceptable.

Augment

Help a professional complete work faster or with better information.

Analyze

Examine documents, data, patterns, or large volumes of information.

Recommend

Provide possible actions or decisions for human review.

Generate

Create drafts, summaries, analyses, plans, or other work products.

Accelerate

Reduce waiting, retrieval, research, coordination, or cycle time.

Not every leverage point should become an AI use case.

The purpose is to create options.

9. Consider future capabilities and innovations

Do not design your improvements only around today's capabilities.

Consider future potential and opportunities:

• Model portability
• Platform changes
• Integration requirements
• APIs
• Reusable prompts
• Agents
• Data portability
• Growth in transaction volume
• Reuse across teams
• Potential cross-department application

10. Prioritize opportunities

Rank opportunities using criteria such as:

• Strategic impact
• Economic value
• Feasibility
• Workflow readiness
• Data readiness
• Risk
• Complexity
• Implementation effort
• Scalability
• Future value

It is easy to be overwhelmed by the number of potential projects. But it is critical to choose the few that will go on to the VALIDATE stage. Forwarding too many potential projects to the next stage can overwhelm your team.

Several methods can identify projects with the best potential. One of the easiest and fastest to use is the Impact – Effort Matrix. This simple matrix ranks improvement projects by how difficult they are to fix and the impact they have on the business objective.

If more than two criteria are important for selecting improvement projects, then use a Weighted Project Prioritization Matrix and enter a ranking for different criteria in a spreadsheet to produce a final weighted sort.