Insights
How legacy systems, custom apps, and AI actually fit together
Field notes from building custom business apps: how legacy systems get connected to AI, what we built for operations teams, and what actually worked.
Where AI automation actually pays in an operations business
Most AI pilots stall because they are built beside the workflow instead of inside it. Here is where AI automation reliably pays for an operations business, where it does not, and the one architectural rule that separates the two.
- Four places AI reliably pays, and three where it does not
- The model proposes, deterministic code decides
- Why the validation layer matters more than the model
How custom software with no upfront cost actually works
A no-upfront-cost software engagement works only when one workflow, acceptance, and the post-launch subscription are defined before development begins.
- One workflow first, not an undefined platform
- The subscription begins at agreed go-live
- Data, deployment, and scope need explicit terms
How to connect a legacy system to AI without replacing it
Most legacy modernization projects stall because the team starts with the model instead of the data boundary. Here is the integration pattern we use instead, and the four ways data realistically leaves an old system.
- Keep the legacy system as the system of record
- Four extraction paths, ranked by fragility
- AI at the edges, deterministic code in the middle
Getting clean data out of PDFs, scans, and phone photos
Document intake is the highest-yield place to put AI in an operations workflow, and the easiest place to quietly poison your data. The difference is the validation layer, not the model.
- Per-field accuracy, not per-document accuracy
- Arithmetic catches what confidence scores miss
- The review screen decides the ROI
What field service software actually costs
Published benchmarks and quoted prices in this market are different by a factor of three or more. Here is what the numbers actually say, and the four costs that never appear on the quote.
- Benchmark spend by trade, from a 6,000-buyer dataset
- Why quoted prices run several times the benchmark
- The four costs that never appear on a quote
Running on paper, and what to do about it first
Most contractors shopping for software are not replacing software — they are replacing paper, a spreadsheet, and a memory. That starting point changes what you should do first.
- The data: 61% of buyers had no purpose-built system at all
- Why digitizing paper as-is is the expensive mistake
- The three records to capture before anything else
When packaged field service software stops fitting
Most contractors who tell us their field service platform does not fit are right — but only about a quarter of it. Here is how to find which quarter, and what to do about it without a rip-and-replace.
- The four failure modes, only two of which justify building
- Why per-seat pricing breaks trades businesses specifically
- The hybrid that usually wins: keep the platform, build the edge
Buyers rank the features that users never rate
There is a measurable gap between what software buyers prioritize and what daily users value. It explains most disappointing rollouts, and it is avoidable.
- 94% of buyers prioritize one thing; 49% of users rate another
- Why demos systematically mislead
- An evaluation method built on days, not feature grids
Getting the dispatch board out of one person's head
Every service business has one person who knows which tech to send. Building dispatch software is mostly the work of getting those rules out of their head — not of finding a better algorithm.
- The seven constraints that actually decide an assignment
- Why full auto-scheduling fails and assisted scheduling works
- The reschedule path is the feature, not the edge case
Signs you have outgrown your field service app
Entry-level platforms are excellent until a business crosses one of four specific thresholds. Knowing which one you crossed decides whether you migrate, upgrade, or extend.
- Four ceilings, each with a different remedy
- Why upgrading tiers often solves the wrong ceiling
- The extend-instead-of-migrate option nobody quotes you
When a spreadsheet stops being the right tool
The spreadsheet is usually the best-designed thing in the business — it just quietly became an application. Here is how to tell when it has outgrown itself, and what to do before rebuilding.
- Six diagnostic signs, not a sales pitch
- The spreadsheet is the spec — read it first
- Half-measures that are sometimes enough
Why software rollouts fail in the trades
The software usually works. What fails is the rollout — and it fails in a small number of recognizable ways that have nothing to do with the product you chose.
- Six failure patterns, none of them about the product
- The parallel-running trap
- What to delete on day one
The job you lost was a call nobody answered
A homeowner with a leak calls three numbers and hires whoever picks up. Recovering missed calls is the highest-return automation in a service business — and the easiest one to do badly.
- How to measure what missed calls are costing you this week
- The five things an automated responder must never say
- Why the handoff to a human is the whole design
How to evaluate an AI phone agent for a trades business
The market for AI answering services is now crowded with products that all demo well. The differences only appear on the calls that matter — and you can test for them in an afternoon.
- Eleven test calls that expose the real differences
- The calls an agent must refuse to handle
- Buy, configure, or build — and when each is right
The gap between the estimate and the invoice
Every retype between the estimate and the invoice is a chance to lose a change order. Here is how to make billing a by-product of the work instead of a weekend job.
- The five handoffs where money leaks
- Change orders as first-class objects, not notes
- Why the estimate should be structured, not a PDF
Finding out which jobs actually made money
Most contractors know their annual margin and almost nothing about which work produced it. Job costing is not an accounting exercise — it is a pricing tool, and it only works if it is current.
- The four cost buckets and the one everyone gets wrong
- Why last quarter's job costing is a history lesson, not a tool
- A cost code structure small enough that people use it
What a field crew app has to survive
Field apps do not fail because of features. They fail because they were designed on a desk for someone standing in a crawlspace with one hand free and no bars.
- Five physical constraints that outrank every feature request
- Offline-first as a data model decision, not a caching trick
- The adoption rule: every screen must give the crew something back
Building an AI agent that turns inbound messages into scheduled jobs
Inbound requests arrive on five channels and only become a job after a person retypes them. Here is how we scope an AI agent for that gap — and the four things we never let it decide.
- The agent proposes, the system commits
- Emergency detection fails open to a human
- Measured by jobs booked without retyping
Connecting a custom app to your books without breaking them
The accounting system is the one place where a bad write is expensive and permanent. Here is how we draw the boundary so a custom app can use the books without ever putting them at risk.
- One rule: the books are downstream of nothing
- Sync direction, decided per object and written down
- Reconciliation that finds drift in days, not at year end
Licenses, insurance certificates, and the renewal nobody owns
Compliance paperwork fails quietly and then all at once — at a gate, at an inspection, or when a general contractor holds your payment. It is one of the cheapest systems to build and one of the most expensive to lack.
- The four document families and who each one protects
- Why the expiry date belongs on a record, not in a filename
- Blocking assignment on expired credentials — carefully
Turning one-off jobs into a maintenance program
A maintenance program turns a seasonal business into a predictable one — but only if visits actually get scheduled and renewals actually get asked for. Both fail without a system.
- The agreement, the entitlement, and the visit — three objects, not one
- Why unscheduled visits are a liability sitting on your balance sheet
- Renewal as a workflow, not a hope
How to choose and scope a first automation project
The first project's real job is not to save the most money. It is to prove the approach works in your business, quickly enough that people believe the second one.
- A scoring rubric for choosing the first workflow
- What actually drives the cost, and what does not
- Nine questions to ask before signing anything
How to replace five SaaS subscriptions with one custom app
Subscription sprawl is not a purchasing problem, it is a workflow problem. Here is the audit we run before proposing a build, including the tools we tell clients never to replace.
- Map tools to workflow steps, not to departments
- Build the seam, buy the commodity
- The renewal calendar decides your sequence
Who a no-upfront-cost software model is actually for
The model works best when a business has one expensive, bounded workflow, accessible systems, and a team ready to adopt the result.
- A clear operational owner is non-negotiable
- The first problem must be bounded and measurable
- Not every custom software project should use this model
Custom software pricing models: upfront project or subscription?
The right software pricing model depends on who carries discovery risk, how clear the scope is, and whether the system needs to keep evolving after launch.
- Price is really a decision about risk
- Fixed scope and ongoing product work need different mechanics
- Subscription fits a system that must stay current
Tell us what tools or workflows you want to replace.
Share the subscriptions, legacy systems, or manual steps that feel too expensive or too messy. We will tell you what could become one custom app.
Get a custom app review