Documentation.
How Ziiro works.

Discovery framework

• Revenue model
• Customer journey
• Operations and team
• Cost structure
• Growth blockers

What to ask in each

• Revenue. How the business makes money, how it prices, which streams exist, and how payment actually arrives.
• Customers. Who buys, why they buy, how they find the business, and what makes them leave.
• Operations. The daily workflow, the team structure, the tools in use, and which steps are still done by hand.
• Finance. Cost structure, margins, cash flow patterns, and what is actually constraining growth.

The point is to be able to describe the business without using the word AI once.

Annotate every step

• Manual
• Bottleneck
• Repeated
• Error-prone
• Handoff

Process audit checklist

• Find. Manual steps, bottlenecks, repeated work, and the gaps where one team waits on another.
• Score. How often it happens, how long it takes, what it costs, and how often it goes wrong.
• Decide. Automate it, simplify it, eliminate it, or leave it alone.

Leave alone is a real answer. A step that is rare, cheap and reliable is not worth touching.

Revenue health

• Monthly recurring revenue
• Gross and net profit margin
• Average order value
• Customer lifetime value
• Customer acquisition cost
• Churn rate

A standard rule of thumb: if lifetime value is under about three times acquisition cost, the problem is structural and automation will not fix it.

Operational efficiency

• Time per task
• People per process
• Cost per operation
• Error rate
• Customer wait time
• Throughput
• Rework rate

Building the baseline

• One row per process. Time per occurrence, how often it occurs, how many people it takes, what it costs per month, and how often it has to be redone.

Taken before anything changes, because it is the only thing a later claim of improvement can be measured against.

Six questions per problem

• Can it be automated?
• Can it be predicted?
• Can it be summarized?
• Can it be classified?
• Can it be optimized?
• Can the decision be assisted?

If a problem answers no to all six, it is not an AI problem. That is a useful result, not a failed one.

Matching a problem to a solution

• Invoices, receipts, forms. Document extraction: read the file, pull the fields, validate against rules, push into the system of record.
• Repeated customer questions. Retrieval over your own documents, answering from the knowledge base and escalating when confidence is low.
• Lead qualification. A scoring model trained on which past leads actually closed, so the strongest are routed first.
• Email and follow-ups. An assistant that drafts from context, personalises outreach and pulls out the action items.
• Recurring reports. A scheduled pipeline that pulls the data, computes the measures and sends the result.
• Sales calls. Transcription plus extraction: the key points, the objections raised, and where the conversation turned.
• Finding internal information. Retrieval over procedures and policies, so the team can ask in plain language instead of hunting.
• Demand and churn. Forecasting on historical patterns, so stock, staffing and outreach move before the fact rather than after it.

How the calculation is structured

• Current monthly cost. People times hours times hourly rate, plus tool costs, plus what the errors cost.
• Cost after the change. The human hours that remain, plus subscription, plus maintenance.
• Monthly saving. Current cost minus the cost after the change.
• Implementation cost. Development, integration, training, and a buffer.
• Break-even. Implementation cost divided by the monthly saving.
• First-year return. Monthly saving times twelve, minus implementation cost.

Every figure comes from the baseline taken in phase 03, not from a benchmark or an industry average. A number you cannot trace back to the client's own operation is not evidence.

The two axes

• Return. The monthly saving from phase 05, vertically.
• Effort. Build time, integration difficulty and data readiness, horizontally.

Reading the quadrants

• Do first. High return, low effort. This is where the first build comes from, every time.
• Plan next. High return, high effort. Worth doing, but it needs the quick wins to land first.
• Nice to have. Low return, low effort. Fill-in work. Never the opening move.
• Skip for now. Low return, high effort. Revisit only if the return changes.

The matrix exists so the first build is settled by where things land rather than by who argues hardest for them.

Sequencing rules

• Quick wins first. Something has to be working early, while the appetite for the project is still there.
• Core systems next. Once the quick wins have proved the data is reachable and the access actually works.
• Advanced work last. Forecasting and decision support depend on both of the above being in place.

Every milestone carries

• A dependency. What has to be working before this can start.
• A success criterion. The baseline number it has to move, agreed before the build rather than after it.
• An owner. Who on your side signs it off.