The most common question we get is: where do we start?
It comes from founders who know AI is reshaping their industry. It comes from ops leads drowning in manual work. It comes from CTOs who have already tried ChatGPT wrappers and want something that actually sticks.
Our answer is always the same: you start with an audit. Not a slide deck. Not a "discovery workshop" that produces a PDF nobody reads. A four-week engagement that ends with a working prototype and a roadmap you can actually execute on.
Here is exactly what that looks like.
Week 1 — Observe and interview
We do not start with technology. We start with your people.
In the first week our team embeds with yours. We sit in on standups, shadow operators, watch how support tickets get triaged, how invoices get processed, how data moves between systems. We are not looking for what people say is slow — we are looking for what actually is.
We call this follow the friction. Friction shows up as copy-paste between tabs, as "let me just check that in the other system," as the spreadsheet that one person maintains because the integration never got built. These are the moments where AI can create step-change improvements, not incremental ones.
Alongside observation, we run structured interviews with people at every level — from the exec sponsor down to the person doing the work. Leadership tells us what matters strategically. The people on the ground tell us where time actually goes. The gap between those two stories is where the real opportunities live.
What we are looking for in Week 1:
- Repetitive, rule-based tasks that consume skilled people's time
- Decision points that rely on pattern recognition across documents or data
- Handoffs between teams where context gets lost
- Processes that scale linearly with headcount — every new customer means more manual work
By the end of the week, we typically have 15 to 30 candidate workflows. Most companies are surprised by how many there are.
Week 2 — Map and score
Week 2 is where observation turns into analysis. We take every candidate workflow from Week 1 and run it through a structured scoring rubric.
Each workflow gets evaluated on five dimensions:
- Impact — how much time or cost does this process consume today?
- Feasibility — can current AI capabilities handle this reliably, or are we betting on models that do not exist yet?
- Data readiness — is the data this process needs accessible, clean, and in a format AI can work with?
- Risk tolerance — what happens if the automation gets it wrong? Is there a human in the loop, or does a mistake hit a customer?
- Strategic alignment — does automating this move a metric the business actually cares about?
We score each dimension from 1 to 5 and weight them based on your priorities. A startup burning cash on manual onboarding will weight impact and feasibility heavily. An enterprise in a regulated industry will weight risk tolerance higher.
The output is a ranked backlog — a single, prioritised list of automation opportunities with clear reasoning behind every score. No black boxes.
We also map each workflow visually: current state (how it works today) and proposed state (how it would work with AI in the loop). This makes it concrete for stakeholders who need to see what changes before they approve it.
Week 3 — Prototype the top candidate
This is where most consulting firms hand you a report and walk away. We do the opposite.
We take the top-ranked workflow from the scored backlog and build a working prototype in five days. Not a mockup. Not a Figma file. A running automation that processes real inputs and produces real outputs.
What "prototype" means depends on the workflow:
- For a document processing task, it might be an AI pipeline that extracts structured data from unstructured PDFs, validates it against business rules, and pushes it into your existing system.
- For a customer support triage workflow, it might be a classifier that reads incoming tickets, categorises them, drafts a response, and routes to the right team — with a human approval step before anything goes out.
- For a data reconciliation process, it might be an agent that pulls from two systems, identifies discrepancies, and surfaces them in a dashboard with recommended resolutions.
The point is not to ship production software in a week. The point is to prove the concept works with your data, your edge cases, and your constraints. It also gives your team something tangible to react to — people give much better feedback when they can use a thing rather than read about it.
We demo the prototype to the project team at the end of Week 3. Every demo we have ever run has changed the final roadmap, because seeing a working system always surfaces requirements that interviews miss.
Week 4 — Deliver the roadmap
The final week is about turning everything into a plan your team can execute — with or without us.
The roadmap is a phased implementation plan, typically covering three horizons:
- Phase 1 (0-3 months): Quick wins. Automations that use proven patterns, need minimal integration work, and deliver measurable ROI fast. These build internal momentum and justify further investment.
- Phase 2 (3-6 months): Higher-complexity workflows that require deeper integration, custom model tuning, or changes to existing processes. These are where the bigger gains live.
- Phase 3 (6-12 months): Strategic bets. Workflows that depend on data infrastructure improvements or cross-team coordination. We map these out so the foundational work starts now.
Each phase includes cost projections, expected time savings, technology recommendations, and team requirements. We are specific: not "you will need AI engineers" but "this phase requires 1 ML engineer and 0.5 FTE from your data team for 8 weeks."
We also flag what not to automate. Not every process benefits from AI, and some are better solved with conventional software engineering, better tooling, or just a process change. Honesty about this is what makes the roadmap trustworthy.
What you walk away with
Here is a summary of the concrete deliverables from each week:
| Week | Deliverable | Format |
|---|---|---|
| 1 | Friction log and workflow inventory | Notion doc or shared spreadsheet |
| 1 | Interview summaries and key findings | Written brief |
| 2 | Scored automation backlog | Weighted scorecard (spreadsheet) |
| 2 | Current-state and proposed-state workflow maps | Visual diagrams |
| 3 | Working prototype of top-ranked workflow | Running code, deployed to staging |
| 3 | Prototype demo recording and feedback notes | Video + written notes |
| 4 | Phased implementation roadmap | Slide deck + detailed doc |
| 4 | Cost projections and team requirements | Spreadsheet model |
| 4 | Technology recommendation matrix | Written brief |
Everything is yours. We do not hold deliverables hostage behind ongoing contracts. If you want to take the roadmap and execute internally, go for it. If you want us to build it, we are ready to start the next week.
Why this works
We have run this process across logistics companies, fintech startups, professional services firms, and e-commerce operators. The structure stays the same. The details change every time.
The reason it works is simple: we do the work before we make recommendations. We watch before we prescribe. We build before we present. By the time you see the roadmap, it has already survived contact with your actual data and your actual team.
Most companies that go through the audit end up automating 3 to 5 workflows in the first six months. The median time saved is 20 to 40 hours per week across the team — not because any single automation is revolutionary, but because the compound effect of removing friction from multiple processes changes how the whole operation feels.
If you want to run this for your team, get in touch. We will tell you honestly whether an audit makes sense for your situation — and if it does, we can usually start within two weeks.