Systems and AI build
Two builds under one method. First the system the work runs on: the CRM or the operating software, your data inside it, the repeating steps automated. Then AI on top of it, once there is something solid to put it on. Either way you receive a working system in production, tested before release and documented at handover.
What the build is fixing
A build earns its cost by removing a specific piece of recurring work. These are the six we are asked to remove most often.
The contact history sits in an inbox, the quote in a spreadsheet, the invoice in the accounting system. Each tool holds part of the picture and none of them holds the whole one.
Reporting assembled by hand, and two people telling one client different things.Someone copies figures between two systems, checks them, and formats the result. It takes hours and it happens on a schedule.
Time spent, and errors that surface downstream.Work moves between people by email or message, and what each person needs to receive was never defined.
Rework, and a client who noticed before you did.The lead, the renewal or the quote gets chased when a person has time, in the order they happen to open their inbox.
Revenue that was already won gets lost.The answer exists, in a document nobody can find or a head that is on holiday.
Wrong answers given confidently, and duplicated effort.A prototype proves the idea, then stalls on access, data quality, or the question of who supports it in production.
Investment spent proving something you already believed.The two builds
Both run the same method and the same three tiers. What changes is what gets built. Read the first one even if you came here for the second.
CRM, data, automation and business operating software
The system the work runs on. One place the record lives, your data inside it and trusted, the repeating steps automated, and the reporting built on top of it. This is the door most clients come through.
Most systems builds land in Tier 1 or Tier 2: $2,500 to $12,000, five to nineteen weeks.
AI workflows and agents, and FusionOS where the scope is a whole function
AI put to work inside that system, on processes structured enough for it to be safe. Built to a written specification, measured against an evaluation set, and released with the human review points already in place.
AI builds usually land in Tier 2 or Tier 3: $4,000 to $40,000 and up, ten to forty weeks.
Most clients come through the first door. The system of record goes in, the data becomes something people trust, and the work stops being manual. That is the foundation the second door needs: AI on scattered data produces confident wrong answers, and no amount of model quality fixes it. Once one system holds the truth, layering AI onto it is a shorter build rather than a bigger one. Take one, or take both in either order. This is the order that usually works.
What you receive
Scope varies with the door and the tier. This part holds through both: you get the specification the system was built from, evidence it was tested, and enough documentation to hand the work to somebody who is not us.
Price and duration
What decides the tier is how much of the business the build touches, how sensitive the data is, and how much the system decides on its own. A build spanning several departments, or one that acts without a person in the loop, carries more testing, more controls and more evidence, and that is what the band pays for.
Tier 1
$1,500 to $4,000
5 to 8 weeks
A standard build on low-sensitivity data. A configured system of record, a data migration, workflow automation, reporting, an integration between tools you already run.
Tier 2
$4,000 to $12,000
10 to 19 weeks
Several departments, integrations and custom reporting on the systems side. On the AI side, anything touching personal or business data: retrieval over your own knowledge, AI-assisted workflows, systems that recommend and a person acts.
Tier 3
$12,000 to $40,000+
20 to 40 weeks
Agentic, regulated or payment-handling work. Multiple agents coordinating, systems acting on their own conclusions, anything where a wrong output has a regulator or a bank behind it.
Governance controls appropriate to the tier are inside the build price. Running the system after it goes live is priced separately, and you are not obliged to buy it. The systems half splits these tiers into a fixed-scope band and a scoped one, with the migration and the platform question answered in full, on Systems Build.
Proof
Both took a step that used to wait on a person being free and gave it to a system. The figures are what those systems do now, on ordinary days.
A form submission became a full client profile through three agents running in sequence. What used to take an analyst a morning now completes before the prospect closes the tab.
Three agents went into a live territory covering more than 550 districts, with the human review points designed before the agents were built.
Understand it first
What to automate in a CRM, what an agent can do inside Zoho, and how to put AI on operations that already have a system of record.
Fit
Five things have to be true on day one, whichever door you come through. Where one is missing, the mapping engagement supplies it faster and cheaper than the build discovers it.
After go-live
A system in production changes what the next constraint is. The handover package includes the backlog we did not get to, which is where most of these start.
Questions
They cover workflow versus agent, how long a build takes, what an agent may touch, and what happens when the output is wrong.
A workflow follows a fixed sequence with AI on named steps. An agent decides what to do next inside a defined boundary. Workflows fit when the path is known. Agents fit when the work branches and a person would otherwise decide each branch by hand. The build names which one you are getting before work starts.
Five to eight weeks at Tier 1 for a standard build on low-sensitivity data. Ten to nineteen weeks at Tier 2 when personal or business data is in scope or several teams are involved. Twenty to forty weeks at Tier 3 for agentic, regulated or payment-handling work.
Only the systems and actions named in the design and the permission matrix. Read access, write access and which records it may change are decided before build, and the agent is tested against those boundaries.
For anything that reads or writes your records, yes. Scattered or duplicate data produces confident wrong answers. Where the system of record is unsettled, Systems Build or FusionMap comes first and shortens the AI build.
Against agreed success metrics and a set of cases that include the failures you care about. Output is scored for accuracy, review burden and the actions the system is allowed to take. Go-live waits on that evaluation.
Model choice, call volume and caching are part of the design. Usage is measured against a baseline after go-live. Tier selection already reflects data sensitivity and agent scope, which is where most cost surprises come from.
Every build includes an exception path and a named owner. Wrong output is routed to a person, logged, and used to tighten the evaluation set. Agents that write to systems sit behind human approval on the actions that need it.
Bring the process you would remove tomorrow if you could. We will tell you which door it belongs to, which tier it falls in, and what it takes to get it running.
