Operating intelligence is the layer that reasons across a business’s tools and acts on the work, instead of just storing records. It sits above the systems of record, treats them as inputs, and takes the next action.

A model where you sell the outcome and do the work, rather than selling a tool and leaving the customer to make it work. AI makes the result, not the seat, the thing worth paying for.

A body of data only your business has, the record of how you win, lose and decide, that a competitor’s model cannot be trained on.

The undocumented know-how that runs a business: pricing judgement, client handling, the reasons past decisions were made. It usually lives in a few people’s heads.

The database that stores a category of business data, like a CRM for customers or a job system for work. It was the dominant software moat for twenty years.

AI built for one specific industry’s workflow, rather than a generic tool. The research shows vertical, deeply integrated AI succeeds at roughly twice the rate of generic builds.

Founder or owner dependence is a business condition where critical decisions, relationships and operating knowledge remain concentrated in the founder. The company may have staff and systems, yet important work still waits for the owner’s judgement or approval. It describes where the business’s capability sits, not how to fix it or how much capacity it has.

For an Australian service business between $2m and $50m revenue, expect three separate costs. Discovery and workflow mapping typically runs $8,000 to $25,000. A single production workflow costs $15,000 to $60,000 depending on how many systems it touches. Ongoing run cost, covering model usage, hosting, monitoring and maintenance, usually lands between $500 and $4,000 per month per workflow. Anyone quoting a single number before mapping your process is guessing.

Operational intelligence usually means real-time dashboards and IT or operations monitoring. Operating intelligence is the layer that runs the actual work of a business and owns the workflow and proprietary data underneath it.

MIT found 95% of generative AI pilots show no measurable P&L impact, and RAND puts the broader failure rate above 80%, about twice the rate of ordinary IT projects.

Most customers never extract full value from a tool. The value sits in the workflow around it, and when a model can run that workflow, a seat licence looks thin.

Because anyone can buy the same model. The scarce input is data that cannot be copied, which makes genuinely proprietary data more valuable, not less.

Institutional knowledge is a risk because undocumented decisions disappear when the people who carry them are unavailable. Work slows, errors repeat, customers receive inconsistent answers and new staff take longer to become effective. The cost is not only replacement training. It can include missed revenue, rework, compliance exposure and a business that cannot operate at the same standard when one experienced person leaves.

The reasoning layer that pulls from every system of record, becomes the one place to get context and act, and treats the underlying databases as plumbing.

Because friction creates workflow, workflow creates data, and data creates enduring value. The mess is the barrier to entry, which is what makes the business that codifies it defensible.

Buyers price in the risk of the owner leaving, so owner-dependent businesses trade at a discount or do not sell at all. It caps the multiple.

The build is the smaller number. Run cost is where budgets break. Model usage scales with volume, integrations need maintenance every time a vendor changes an API, and someone has to watch for silent failures. Budget 20 to 30 per cent of the original build cost annually to keep a workflow healthy. A workflow nobody is monitoring is not saving money, it is accruing risk quietly.

No. It is built by doing the work, codifying the workflow, and capturing the data the work throws off. The software is the easy part now. The context and proprietary data underneath it are what cannot be bought.

In deeply integrated, vertical workflows, usually in the back office. The most-funded use cases, sales and marketing pilots, tend to have the lowest return.

By doing the work and carrying its risk, not by advising from outside. Going inside the work is how you see the exceptions that let you codify it.

The exhaust of the work: why quotes were won or lost, how exceptions were handled, what was said on the call. It lives at the edges of the work and no form catches it.

AI captures institutional knowledge by reading unstructured records, extracting decisions, exceptions and context, then turning them into structured information that can be retrieved when the next task needs it. A wiki depends on someone remembering to write and organise an entry. AI can work across emails, notes, documents, call transcripts and job records, identify recurring patterns and return the relevant precedent in the flow of work.

Because AI can reason across many systems at once, so the value moves from where data is stored to what can reason across it. Investors describe the record becoming a commodity layer.

Regulated, relationship-heavy service industries like trades, aged care, allied health, accounting, construction and strata, where adoption is low and no operating layer exists.

Move the process out of the founder’s head and into a system the business owns, with the workflow captured and the judgement encoded as rules the layer can run.

Take the task's total annual cost, not the hourly rate. Multiply hours per week by 52, then by the fully loaded cost of the person doing it, which is roughly salary plus 25 to 30 per cent for superannuation, leave, workspace and management overhead. Subtract build cost and 12 months of run cost. If payback is longer than 18 months, the workflow is probably the wrong one to start with.

It moves the advantage from the tools you rent to the layer you own, and it shows up on the P&L: more revenue from a sales team armed with what closed before, lower software spend, and a marketing budget that follows the data.

Fix the workflow and the data first. Run the work on a system, capture the data, then apply the model to the step that moves the number.

No. The point of doing the work is to capture it. The service pays for the build, and the captured layer is the asset you own at the end.

You capture proprietary data as a by-product of the work rather than as a separate exercise. In practice, that means recording decisions, exceptions and outcomes in the system where the work already happens, so each job adds to the record without creating another administrative task. Over time, those records reveal how the business actually operates. They are defensible because they are specific to your customers, workflows and judgement, not generic data a competitor can buy.

A business that owns its knowledge as a system runs and scales without specific people, which is worth far more to a buyer than one that depends on them.

The question is no longer which database you sit on, but whether you own the layer that reasons across them, built by doing the work and capturing the data.

A service business is a strong vertical AI candidate when it has repeatable work, expensive exceptions, fragmented systems and enough volume for better decisions to compound. Look for an industry where the workflow is similar across customers, the operational data is hard to access and generic software leaves important judgement to people. The right starting point is the vertical with a painful, recurring pattern, not simply the most fashionable technology.

Build the operating layer so the business runs, scales and sells without the person who built it. Same revenue, higher multiple, because the key-person risk is gone.

More than the wages. Manual processes carry rework cost when someone makes an error, delay cost when work sits in an inbox, and opportunity cost when a senior person does administrative work instead of billable or business-building work. In professional services the third one usually dominates. A principal doing two hours of admin a day is a six-figure annual decision, made invisibly.

Australian providers generally price between $200 and $900 per month for a small business, either per seat or on a per-minute basis with an included allowance. Setup and integration into your booking or job management system typically adds $1,500 to $6,000 once. The per-minute models look cheaper until call volume spikes. Ask for a worked example at twice your current volume before signing anything.

A single-purpose agent handling one workflow, such as intake triage or quote follow-up, generally costs $15,000 to $45,000 to build properly and $300 to $2,000 a month to run. A multi-agent operating layer across several connected workflows runs materially higher. The cost driver is almost never the model. It is the number of systems the agent has to read from and write to, and the exception paths it has to handle.

Often not yet, and we will tell you when it is not. Below roughly 15 staff, most service businesses get a better return from documenting and simplifying the process than from automating it. Automation multiplies whatever process it is pointed at, including a bad one. The threshold is usually not revenue, it is repetition. If the same decision gets made the same way more than 20 times a week, it is a candidate.