Nine Days Faster and So Is Everyone Else

The situation

A national commercial refrigeration and HVAC maintenance business, 240 field engineers across eleven branches, 60,000 assets under maintenance contract at 4,200 customer sites. Two AI systems have been live for just over a year. A drafting assistant turns a site survey into a priced quote. A parts recommender reads the fault description and the asset record, then returns a parts list for the van.

Finance signed off a benefits register in November. Median quote turnaround has gone from eleven working days to two. Repeat visits inside thirty days have gone from 12.6 per cent of jobs to 8.4. Two of the seven schedulers have been redeployed onto contract renewals. The register values the year at AUD$2.1m against AUD$1.4m of programme cost, every number measured rather than estimated.

The company’s first strategy hire, four weeks in and reporting to the CFO, has been asked to turn the register into the competitive-position section of the February board paper. Her first question is one nobody has asked in twelve months: which of these three does our nearest competitor get by signing the same vendor contract next quarter? Two of the four largest competitors are named as customers on the vendor’s own website. For every line, the first answer is yes.

What actually matters

An operational improvement justifies itself on payback: it cost this, it returns that, it breaks even in month eight, and that stays true whether or not a rival buys the same thing. A competitive advantage has to answer a harder question: what stops a rival from having it. A board that catches one overclaim discounts the rest of the page.

Nine days off quote turnaround is the largest number on this register and the most easily copied thing the company did all year. Operational improvements are usually the bigger, faster, more certain benefit, and they belong under cost position, argued loudly. What the board has not been told is what happens next. As the sector adopts the same tooling, a two-day quote becomes the minimum a customer expects, and the gain goes into price and service-level terms. Bank the payback quickly rather than describing it as a position.

Durability comes from four things a vendor contract does not include: data only this company holds and keeps collecting, integration into a system of record a rival does not have, physical access the AI multiplies rather than creates, and capability built up over years. Test each with one question. Could a competitor have this by the end of next quarter by writing a cheque? Where a purchase order gets it, it is operational.

A drafting assistant that saves estimators time is bought capability. Capture the corrections, label them and feed them back into a model trained on the company’s own history, and the same assistant sits on a different curve. The corrections arrive as a byproduct of doing the work, and no competitor can assemble the pile. That turns on a decision about instrumentation rather than on anything the vendor ships, so the sort runs line by line rather than tool by tool.

What we’ll filter on

Five things the method has to do before it produces a section anyone can defend in the room.

1. Separates bought from built. Does it distinguish benefits a competitor gets by signing the same vendor contract from ones they would have to assemble?
2. Names what blocks replication. For every durable claim, does it say what a rival would have to acquire, and roughly how long that takes?
3. Keeps the operational payback claimed. Does the AUD$2.1m still get argued somewhere in the paper, rather than discarded because it is copyable?
4. Estimates how long the lead lasts. Does it produce a decay view for the improvements, in quarters or in years, based on what competitors have already deployed?
5. Repeatable in-house. Can the company run this again at the next planning round without buying it again?

The landscape

Claim the register as it stands

Retitle the three lines and file them under competitive position. It takes an afternoon and survives until the first director who has read the vendor’s website asks who else runs the software. Nothing in the register is wrong; finance measured every line rather than estimating it. Filed under a heading that promises defensibility, those numbers leave the paper with no answer to the obvious follow-up, and the damage runs backwards into the measured benefits.

Drop the section and argue payback only

Present the three lines as cost position, ask for the reinvestment on the basis of the return, and leave competitive position blank. Honest, and enough for a decision about tools already bought. It does not answer what the board is weighing: did a year of AI spend change the company’s position, or only its cost base? It also leaves the two durable lines unargued, and nobody tasked with building more of that kind.

Benchmark what the sector has deployed

Assemble a picture of competitor adoption from vendor customer lists, job adverts, trade press and what customers volunteer during renewals. Cheap, quick, and it turns two names on a vendor website into a date for each of the four largest competitors. It is a snapshot of the present, which sets the half-life of the operational lead and says nothing about which parts of the year’s work are defensible. An input to the assessment rather than a substitute for it.

Buy a market-positioning study

Commission an analyst or consultancy to assess the company’s AI position against the sector: four to six weeks, AUD$90,000, board-formatted, with comparators, outside pattern knowledge and senior attention. It assesses the market, so it will not find the 2,900 estimator corrections the drafting assistant discarded last year, because that asset is invisible from outside and nobody inside has counted it as one. It does not carry the measured payback either, so the paper still needs the other half written internally.

Sort the register with a replication test, then write two sections

Take each line and ask what a competitor would have to do to have it by the end of next quarter. Everything a purchase order buys goes into one list and is argued on payback and half-life. Everything requiring data, an installed base, or capability that takes years goes into the other and is argued on what blocks replication. It costs a day of work, it uses the sector benchmark as its evidence for decay, and it can be re-run every planning round by the same person. It also produces an output nobody asked for: a short list of places where the work produces corrections that nobody is keeping.

Evaluation

Side by side

Approach	Separates bought from built	Names what blocks replication	Keeps the operational payback claimed	Estimates how long the lead lasts	Repeatable in-house
Claim the register as advantage	✗	✗	✓	✗	✓
Payback only, no advantage section	✗	✗	✓	✗	✓
Sector adoption benchmark	✗	✗	✓	✓	✓
Bought market-positioning study	✓	✗	✗	✓	✗
Replication sort, two sections	✓	✓	✓	✓	✓

The first two rows fail the same column for opposite reasons. One calls everything an advantage, the other calls nothing one, and neither tells a director which of the three lines a competitor can copy. The benchmark is the only bought input worth having, and it does that work inside the fifth row. The market study is the most expensive option and the worst fit for this company, whose defensible assets are internal and undocumented, invisible to any external comparison. The replication sort clears all five columns because it works on the register rather than the market: every line already carries a measured number, and the test adds one column to the same spreadsheet.

The solution

The test, in one question

Ask of each claimed benefit what our nearest competitor would have to do to have it by the end of next quarter. Signing the same vendor contract makes the lead as long as their procurement cycle. Changing how they work as well stretches it to perhaps a year, still operational. Only acquiring something a contract does not sell belongs under competitive position.

The five shapes of operational improvement

Operational improvements show up as cycle time, error rate, throughput, released capacity and consistency. Quote turnaround is cycle time, repeat visits are error rate, two schedulers redeployed is released capacity. Two more were measured and left off: jobs closed per engineer per day, and the price spread for the same job across eleven branches. All five follow the tool, and two of the four largest competitors already run it, so the AUD$2.1m is a cost-position argument with a half-life in quarters.

The four things a contract does not sell

Proprietary data the system learns on. Nineteen years of job records across 60,000 assets, each tying a reported fault to the part fitted and whether the asset came back inside thirty days. A competitor buying the same recommender gets a model shaped by the vendor’s whole customer base; this company trains a model on its own installed base.

Integration rivals do not have. The recommender reads an asset register keyed to serial numbers engineers have been scanning on site for six years, and writes back into the dispatch system. A rival with the same tool works from the fault text alone.

Distribution and access. The company holds maintenance contracts on 4,200 sites, with an engineer on most of them every six weeks. AI multiplies whatever the visits are worth, and the visits are what nobody else has.

Operating capability. Estimators who review a draft rather than accept it, a governance route clearing a use case in three weeks instead of five months, and someone keeping the labels clean. Two to three years to assemble, and the habit of keeping a shipped system working rarely survives a reorganisation.

The two lines that change category

The drafting assistant produced 4,100 quotes last year and estimators materially edited 2,900 of them, each recording the right scope and price for one asset. Every repeat visit inside thirty days is a recorded failure of a parts recommendation. Both were operational on launch day and turn durable once the corrections are captured, joined to outcomes and used. Only the repeat-visit flag was captured. The system logged that a quote had been edited but wrote the new version over the old one, so the count survives and the corrections do not, and instrumenting them is a fortnight of engineering. A growing pile of corrections is an asset only once it is large enough to change how the model behaves, tied to outcomes, and legally usable under the contracts it was collected under.

What the paper ends up looking like

Cost position carries the three valued lines plus the two that were measured and never registered, each with its payback, its half-life, and a plain statement that competitors on the same platform are getting the same gain. Competitive position carries four claims, each naming the barrier and how long a rival would need: nineteen years of outcome-labelled asset history, a serial-level register six years in the making, access to 4,200 sites, and estimators trained to work with a draft. The ask is AUD$340,000 of standing cost across three years against AUD$1.4m of project spend. Where each of those sits on the buy-versus-build line is a separate decision.

Worked example

Claimed benefit	Measured	Competitor’s route to it	Category	Where it goes
Quote turnaround 11 days to 2	median, 4,100 quotes	Signs the contract	Cycle time, operational	Cost position, compounding once corrections are captured
Repeat visits 12.6% to 8.4%	all jobs, 30-day window	Signs, then needs six years of its own scans	Error rate, operational in part	Cost position, plus a competitive claim for the history
Two schedulers redeployed	headcount, seven to five	Signs the contract	Released capacity, operational	Cost position only
Price spread across branches narrowed	price variance, same job codes	Signs the contract	Consistency, operational	Cost position, add to the register
Jobs closed per engineer per day	240 engineers, same job mix	Signs the contract	Throughput, operational	Cost position, add to the register
Nineteen years of outcome-labelled history	60,000 assets, fault to part to recurrence	Acquires an installed base	Advantage	Competitive position, standing cost attached
Serial-level asset register	six years of scans, 4,200 sites	Six years of visits it lacks	Advantage	Competitive position
Access to 4,200 contracted sites	visit records, six-week cycle	Wins the contracts away, site by site	Advantage	Competitive position
Estimators who review a draft	2,900 of 4,100 quotes edited	Two to three years of practice	Advantage	Competitive position, standing cost attached

The second row is both: operational and copyable as a benefit, and a byproduct generating labelled failures only the maintenance-contract holder can collect.

What’s worth remembering

1. Copyable means operational. A competitor signing the same vendor contract next quarter gets it, however large the number, and an overclaim discredits the whole register.
2. Operational gains take five shapes. Cycle time, error rate, throughput, released capacity and consistency, each arriving with the tool and argued on payback.
3. Four things a contract cannot sell. Proprietary data that keeps arriving, integration into a workflow or installed base, distribution, and operating capability measured in years.
4. Corrections compound, if captured. An operational gain becomes an advantage only when the work produces corrections nobody else can collect, joined to outcomes.
5. Date the half-life with evidence. Competitor adoption, read from vendor customer lists, job adverts and renewal conversations, sets how many quarters an operational lead survives.
6. Both categories, different sections. Cost position argues payback and half-life; competitive position names the barrier and its running cost.