SESSION · T_3895 PACIFIC TIME BUILD 2026.05.19 NO SLIDES
We ship to our own clients before we ship to yours. Discovery is 30 minutes, no slides, no pitch deck.
Two ICPs, one problem
If you run a 50–500 person business
You've sat through demos. You've watched the rep type a prompt and get a confident wrong answer. You've seen the slide deck with the four pillars and the gradient. What you haven't seen is someone show up, look at your actual ticket queue or your actual finance close or your actual onboarding flow, and tell you which one has the highest ratio of hours-saved to integration-cost.
If you run public-sector IT
You're two people short. Your ticket volume isn't going down. Vendors keep selling you "AI for government" without understanding what your data residency rules actually require. The right partner is someone who's done the M365 + on-prem balance before, knows what Copilot Studio does and doesn't do, and will tell you when an off-the-shelf tool is the better answer than a custom build.
That's the work. Everything else is theater.
What we ship
We don't pitch a methodology. We ship one of these three, in this order, depending on what your org actually needs.
Custom agents that decide what to do next based on what came back from the last step. Built on Anthropic Claude Agent SDK, OpenAI Agents SDK, or Microsoft Copilot Studio depending on where your data lives. Shipped with an eval harness, a cost dashboard, and a named human owner — or we don't call it shipped.
How we build agents12:04:11 intake → ticket #4092 received 12:04:11 classify → category: vpn_outage (0.94) 12:04:12 retrieve → 3 KB articles, 2 prior tickets 12:04:12 draft → reply + suggested fix 12:04:13 policy → escalate (after-hours rule) 12:04:13 notify → on-call engineer paged ───────────────────────────────────────────── end-to-end: 1.8s tokens: 4,302 cost: $0.011
Deterministic workflows on n8n, Zapier, Make, or Microsoft Power Automate. We pick the tool based on who owns it after we leave, where the data lives, and whether it's allowed to leave your infrastructure. About 60% of what gets pitched as AI is actually this.
Tool selection logictool data residency owner-after-leave fit ─────────────── ──────────────── ────────────────── ────── n8n self-host on customer infra customer team ★★★★★ n8n cloud (eu) vendor cloud ours, then theirs ★★★★ Zapier vendor cloud customer team ★★★ Make vendor cloud (eu) customer team ★★ Power Automate M365 tenant customer IT ★★★★ ───────────────────────────────────────────────────────────── shipping: n8n self-host reason: data residency requirement
Document QA and internal search for orgs with 100–50,000 documents and questions that cluster. Every answer cites its source passage. Every system ships with a test set of 50–100 real questions from your team and a measured precision and recall against it.
When RAG earns its keeprecall@1 recall@5 p95 latency ───────────── ───────── ───────── ─────────── naive bm25 0.41 0.68 180 ms + rerank 0.58 0.81 310 ms + hyde 0.62 0.86 420 ms + tuned chunks 0.71 0.91 430 ms ───────────────────────────────────────────────── shipping: tuned-chunks + rerank (no hyde)
Built in our lab
AVA is the AI-powered web chat we built so our clients can reach our engineers without opening a ticket queue. It writes the Autotask ticket, categorizes it, assigns it to the right engineer, and lets us log time entries, add notes, and update status without ever leaving the chat. It runs in production on our MSP every day.
The credentialing artifact: AVA is what an agent shipped to production looks like, not a slide describing what one might look like.
Licensable to other MSPs running Autotask. Talk to us if you'd rather skip the months we spent on the integration plumbing.
Diagnose first
Ten questions across four dimensions: data readiness, workflow maturity, team capacity, use-case clarity. You get a personalized PDF the same minute you finish, with a tier-specific recommendation. The PDF is useful whether you hire us or not — that's the point.
How we work
Three steps. Real timelines. No retainer-by-default.
A 30-minute call. No deck. You bring one workflow you'd like to improve. We tell you whether AI is the right shape for it, or whether you'd be better served by automation, by RAG, or by leaving it alone for now.
One workflow. Fixed scope. Fixed price. Eval harness, cost dashboard, runbook, and a named owner inside your org. End-of-pilot is a real "keep going or kill it" decision based on a metric you and we agreed to in week one.
If the pilot earns it, we move to either an embedded fractional engineer model, a few days a week with your team owning the rest, or a fixed-scope build of the next workflow. We do not default to retainer.
Tools we deploy on engagements
Who we are
AIX Automation Lab is the AI agent and automation arm of Eaton & Associates Enterprise IT Solutions. Eaton has spent the last decade-plus running enterprise IT for Bay Area municipalities, school districts, nonprofits, and SMBs. AIX is what happens when that operator team starts shipping AI for the same kinds of clients.
We built AVA because we needed it for our own MSP. We ship the same way for clients: the goal is software that runs in production for years, not a pilot that demos well and quietly dies. We're a small team on purpose. The person you talk to on the discovery call is the same person in the pilot and the same person in production.
Read more about the teamSESSION · 092 · 30 MIN · PACIFIC TIME · NO SLIDES
Worst case: 30 free minutes of AI strategy from people who've shipped it. Best case: you walk out of the call with a clear next step you didn't have an hour ago.