What is an AI systems integrator? An AI systems integrator builds AI into the software your business already runs, reads and writes to your real records, and hands the working system to your team with a person approving anything that changes money. Not a chatbot bolted to your website. Not a subscription to somebody else's tool. A system built into the ERP, CRM, or accounting package you already log into every morning, doing a specific chore that used to eat hours.

Key takeaways

  • An AI systems integrator writes into the software you already run (ERP, CRM, accounting), not a separate app your team has to remember to open.
  • Every write to a live record goes through human approval; the system surfaces work, a person on your team approves it.
  • Throughline's rebate-and-margin tool inside Kelsan, a multi-state distributor running Epicor P21, has recovered over $100,000 in margin.
  • Engagements start with a fixed-fee AI Capability Audit ($7,500); the build is a separate fixed fee agreed in writing before work begins.
  • If the chore is rare, one-off, or already handled well by a person, do not hire an integrator. Build only where the work is repetitive and the records are digital.

The category, in plain terms

A systems integrator, in the older sense, is the firm that connects your ERP to your warehouse management system, or your CRM to your billing platform. They make software talk to software. An AI systems integrator does the same job with a different ingredient: a language model or extraction model that can read unstructured input (a PDF, an email, a scanned invoice), decide what it means, and propose an entry into a system of record.

The important word is "propose." A well-built system does not act unilaterally on money. It reads the document, matches it to what already exists in the ERP, and puts a draft entry in front of a human. That human clicks approve, edit, or reject. The record moves only after the click.

That single design choice separates real integration work from the demos you see on LinkedIn.

What the work actually looks like inside a business

Consider a distributor. Every month, dozens of supplier rebate documents arrive in different formats: PDFs, spreadsheets, emails with tables pasted in. Somebody has to read each one, find the qualifying invoices in the ERP, calculate what is owed, and post a credit. It is slow, error-prone, and easy to skip when the AP team is behind.

A builder walks that chore end to end. What document arrives, from whom, in what format. Where the matching invoices live in the ERP. Who approves the entry today, and what they check before approving. Then the builder writes a system that does the reading, matching, and drafting, and puts the result in front of the same approver, in the same place they already work. That is intelligent document automation applied to a specific back-office chore, not a generic "AI feature."

The output is not a chatbot. It is fewer minutes per document, fewer missed rebates, and a paper trail of who approved what.

How this differs from a consultant, an agency, or a tool reseller

Four kinds of firms will answer a call about AI right now. It helps to know which is which. We wrote a longer breakdown in how to choose an AI partner, but the short version:

An enterprise consultancy sells you a strategy deck and a roadmap. Useful if you need board-level cover; expensive if you need a working system next quarter.

An offshore development shop takes your spec and codes to it. Fine if you already know exactly what to build; risky if the requirements are still fuzzy, which they almost always are on the first AI project.

A no-code agency wires together off-the-shelf tools with Zapier, Make, or n8n. Fast for prototypes. Fragile when the workflow touches money, because you do not control the pieces and the seams break under load.

A builder-operator writes the code, connects to your ERP's real APIs, and stays on the hook for the system running in production. That is what Throughline is.

The proof: what we built inside Kelsan

Kelsan is a multi-state distributor inside our own group, running Epicor P21. The chore was rebate and margin recovery. Supplier rebate documents arrived in a dozen shapes. Matching them to invoices took hours, and things fell through the cracks.

We built a tool that reads the rebate documents, matches them against invoices in P21, and surfaces entries for a person to approve before they post. It has recovered over $100,000 in margin that would otherwise have been lost. Every entry that changes a record is approved by a human on the Kelsan team. The system does not act on money by itself.

We use it as our reference case because we lived it. You can read the fuller version in the rebate-and-margin tool we built inside Kelsan. If you are a distributor running P21, that page is the closest thing to a mirror we can hand you, and AI for distributors covers the broader shape of the work.

What an engagement looks like

Every engagement starts with a fixed-fee AI Capability Audit ($7,500). We walk the chores that eat time, look at the systems they touch, and hand back a written plan: what is worth building, what is not, and what a first build would cost. The audit is a real deliverable, not a sales meeting in disguise.

If a build makes sense, it is a separate fixed fee, quoted in writing before any code is written. Running costs (model calls, storage, hosting) are metered like any cloud service and quoted in writing before we start. No surprise invoices.

The work itself is ERP automation, back office automation, or generative AI integration, depending on what the chore needs. When the tool you need does not exist yet, it becomes custom AI solutions built for one company.

When we hand the system over, your team runs it. Your approvers keep approving. We stay available for changes.

When you should not hire one

This part matters, and most firms will not say it.

Do not hire an integrator if the chore is rare. If a task happens twice a year, a person doing it manually is fine.

Do not hire one if your records are not digital. If the invoices live in a filing cabinet, the answer is scanning and a document management system first, not AI.

Do not hire one if you are still figuring out what your business does. AI compounds process. If the process is broken, AI makes the mess move faster.

Do not hire one if the sole goal is to look modern for a board meeting. That is what strategy decks are for, and they cost less.

Hire an integrator when a real chore, done many times per week, on digital records, is quietly costing you money. That is the shape of a build worth doing.

About the author

Throughline is a small team of builders inside Keller Group. We build AI systems into our own operating companies first, then into yours.