AI process improvement case study
Building an AI-led process improvement platform: consultant-grade analysis in days.
How Changeable designed and built a governed AI process improvement platform in house. It runs stakeholder interviews, analyses current-state processes and delivers a professional improvement report and SOP, without the cost or timeline of a traditional consulting engagement.
Project overview
The same quality of analysis a consultant delivers, without the six-week wait.
Process improvement work has traditionally meant booking a consultant, scheduling weeks of stakeholder interviews, and waiting on a report that arrives long after the urgency that triggered the engagement has passed. Changeable built this AI-led process improvement platform to compress that timeline without cutting the rigour: structured stakeholder interviews, evidence-based opportunity analysis and a professional deliverable, completed in days.
The platform runs a defined process improvement methodology end to end, drawing on Lean, BABOK, BPM CBOK and ADKAR practice. Conversational AI conducts the interviews and analyses the transcripts, and a senior practitioner reviews and approves the output at every critical stage, so it is never a fully automated black box.
Human-gated automation
AI drives the analysis; a senior practitioner reviews and approves before anything reaches a client.
Full methodology, not a chatbot
A structured nine-stage engagement, not a single conversational feature bolted onto a form.
Evidence-based recommendations
Opportunity analysis draws on real market benchmarks, not generic templates.
Built around real feedback
Testing surfaced a gap between interviews and the final report, closed with an automatic interim findings summary.
The problem
Traditional process improvement is slow, expensive and inconsistent.
Most small and mid-sized organisations know a process is broken, but fixing it properly means booking a business analyst or consultant, coordinating interviews across a busy team, and waiting weeks for a report. The cost puts the exercise out of reach for many businesses that would benefit most.
Even where budget exists, quality varies. Findings depend heavily on which consultant ran the engagement, how many stakeholders they managed to interview, and how much of the analysis was templated rather than specific to the business. Our process cost calculator shows how quickly that manual effort adds up.
What most businesses experience
- A process everyone agrees is broken, but nobody owns fixing it
- Weeks of scheduling friction just to get stakeholders interviewed
- A final report that arrives after the urgency has passed
- Recommendations that feel generic rather than specific to the business
- No clear next step from insight to actual automation or change
The methodology
A nine-stage engagement, run by AI and reviewed by a senior practitioner
The platform’s engagement model mirrors how a skilled business analyst would actually run a process improvement project, structured into defined stages with clear human review gates.
Organisational onboarding
The client uploads existing documentation and answers deep-context questions so the AI understands the business before any process work begins.
Project & stakeholder setup
The client defines the process in scope, names a process owner and registers the stakeholders who need to be interviewed.
Stakeholder interviews
Each stakeholder is invited to a private, conversational AI interview that adapts its questions based on what has already been said.
Current state analysis
Transcripts are analysed into a structured process step register and pain point list, reviewed and approved by a senior practitioner before proceeding.
Opportunity identification
Improvement opportunities are identified using real market research, not generic templates, then reviewed for feasibility.
Validation session
Findings are put back to stakeholders in plain language for confirmation or correction before the future state is designed.
Future state design
A revised process is designed, with every change documented and justified against the original pain points.
Automation assessment
Each future-state step is assessed for automation potential, with confidence levels and realistic effort estimates.
Report and handover
A professional report and standard operating procedure are generated, reviewed by a senior practitioner, then released to the client.
How it was built
A phased build, one working capability at a time
Rather than design the full platform upfront, it was built and tested capability by capability, with each stage running end to end before the next was layered on.
Start with the hardest part: a real AI-led interview
The riskiest capability was also the most important: an AI-led interview that felt like a conversation with a skilled analyst, not a form with extra steps. Authentication, database structure and document-upload context were built first, so the AI could read a client’s existing documentation before the conversation began and focus its questions on the gaps. Early testing with a real business document confirmed the approach worked: the AI extracted the organisational profile automatically and only asked about what was genuinely missing.
Connect interviews to evidence-based recommendations
With interviews working, the build moved through current-state analysis, opportunity identification using live market research, a stakeholder validation step, future-state design and an automation feasibility assessment. Each stage was tested end to end with real transcripts before being connected to the next, so the full chain, from stakeholder interview through to automation recommendation, could be validated in a single pass before any client used it.
Keep a senior practitioner in the loop, not just the AI
A fully automated report was never the goal. Explicit approval gates were built at the stages that matter most: after current-state analysis, after opportunity identification, and before the final report and SOP are released. A practitioner reviews the AI’s output, edits inline where needed, and only then approves the engagement to move forward. This is the same discipline Changeable brings to client AI governance work: automation accelerates the analysis, but a person remains accountable for what reaches the client.
Turn structured data into a professional deliverable
A structured process register is not a deliverable a business owner can act on. A dedicated report-generation pipeline was built to turn the approved analysis into a branded PDF report and a matching standard operating procedure, alongside an SLA-tracked admin dashboard so every engagement’s review status is visible at a glance. Client feedback during testing surfaced a real UX gap: process owners had nothing tangible to see between the interviews and the final report. An interim findings summary, sent automatically once interviews were complete, was added to keep stakeholders engaged through the analysis phase.
The AI agents
Conversational interviews, not a survey with an AI label on it.
The core of the platform is a conversational AI agent that conducts each stakeholder interview individually, adapting its follow-up questions based on what the stakeholder has already said rather than working through a fixed script.
During testing, the interview AI referenced a detail a stakeholder had mentioned earlier in the same conversation and followed up on it unprompted, the kind of adaptive probing a skilled interviewer does naturally and a static form cannot replicate. The same underlying approach powers the opportunity analysis stage, where the AI draws on current market research rather than a fixed library of generic recommendations.
Platform and delivery
A public site built to be found, an application built to work
The marketing site and the working application were deliberately split, each optimised for what it actually needs to do, with the delivery tooling built around practitioner review.
Public site for discoverability
The public-facing website was built separately from the application, giving full control over page speed, meta data and content for organic search performance.
Application built to work, not rank
The client-facing app sits on its own subdomain, focused entirely on a fast, reliable engagement experience rather than search visibility.
Report and SOP pipeline
Approved analysis is turned into a branded PDF report and a matching standard operating procedure, ready to hand to the team doing the work.
SLA-tracked admin dashboard
Every engagement’s stage, review status and turnaround is visible at a glance, so no approval gate sits waiting unnoticed.
Interim findings summary
Process owners receive an automatic summary once interviews are complete, so they see progress before the final report.
Documented design system
A formal design system covering colour, typography, imagery and tone of voice keeps every screen, report and email consistent.
Results
What the finished platform delivers
Once the full methodology chain was validated end to end, testing against a real engagement produced measurable results.
Engagement turnaround
A full stakeholder-interview-to-report cycle runs in days rather than the weeks a traditional consulting engagement typically takes.
Annual admin time identified
A test engagement identified several hundred hours of annual administrative time available for reduction through the recommended automation.
Admin burden reduction potential
The automation assessment stage identified a substantial potential reduction in administrative burden for the process under review.
Enforced at every tier limit
Stakeholder and access limits are enforced at both the interface and the database level, so a restriction cannot be bypassed by editing the page.
Generated per completed engagement
Each completed engagement produces a full report, SOP and supporting documents, gated behind practitioner approval before release.
Human sign-off where it matters
Current state, opportunities and the final report each need practitioner approval before the engagement moves on.
Questions
Questions about AI-led process improvement?
Common questions about how this process improvement platform was built, and how the same approach applies to your organisation.
What is AI-led process improvement?
AI-led process improvement uses conversational AI to gather and analyse what actually happens in a process, then identifies where it can be simplified or automated. The AI handles the interviewing and analysis at speed; an experienced practitioner checks and approves the findings before they become recommendations.
Why was the interview stage built first?
It was the riskiest and most important capability. If a conversational AI interview could not genuinely replicate the adaptive quality of a skilled analyst’s questions, the rest of the platform would not be worth building. Proving that first shaped every decision that followed.
Which methodologies does the platform draw on?
The nine-stage engagement draws on Lean, BABOK, BPM CBOK and ADKAR practice: mapping the current state, removing waste, designing the future state and planning for adoption. The same methods underpin Changeable’s process improvement consulting.
Does the platform replace a human consultant entirely?
No. AI drives the interviews and analysis, but a senior practitioner reviews and approves the output at named gates before it reaches a client. The platform accelerates the work; it does not remove human accountability for the recommendations.
How does the platform keep recommendations from feeling generic?
The opportunity analysis stage uses live market research rather than a fixed library of templated suggestions, and a stakeholder validation step puts findings back to real people for confirmation before the future state is finalised.
What does a completed engagement produce?
A professional improvement report, a standard operating procedure for the redesigned process, and supporting documents covering the current state, opportunities and automation assessment. Each is released only after practitioner approval.
Why split the marketing website from the application?
The marketing site needs full control over meta data, content and page speed to perform in organic search. The application needs to work reliably for a paying client. Splitting them let each be optimised for what it actually needs to do.
What happens if the AI gets part of the analysis wrong?
The practitioner review stage exists for exactly this. Findings are editable inline before approval, and if an error is systemic rather than a small correction, the relevant analysis stage can be re-run before proceeding.
Why did Changeable build this platform in house?
Building our own products means the advice we give clients comes from delivery experience, not theory. Running the full method through software forced every stage, gate and handover to be defined precisely, and that discipline now shapes how we run AI strategy and process work for clients.
Can Changeable build something similar for our organisation?
Yes. The same use case-led, governed approach applies in client engagements, including use case development, AI app development and AI governance.
Want to know where your processes are losing time?
Talk to Changeable about applying the same phased, governed approach to your own processes, AI product or workflow.