PCS Global Built a Private AI and Saves $20,000 a Year

VitalIntel Editorial Team

Practical guides from the VitalIntel team, who have been building, rescuing and modernizing software for clients since 2012.

PCS Global's team of 100+ people was using free AI tools with no rules. Some staff had shared emails and client data with outside AI models by mistake, and they were unsure what else had gone. The plan was to pay more than $25,000 a year for AI. We built a private AI on PCS Global's own network instead. It saves them $20,000 a year, and prompts stay on the company network.

The Problem: Free AI and No Rules

PCS Global is a team of more than 100 people. Staff were using free AI tools, and there were no rules about what could be shared with them. It was a free-for-all.

Some people shared emails and client data with frontier AI models by mistake. Nobody was sure what had been shared.

The plan was to pay for AI, at more than $25,000 a year. But paid cloud AI typically means company data is processed on the vendor's servers, and the bill comes back every year.

The Decision: Rent AI or Run It

The alternative was to run the AI inside PCS Global's own network.

Open-source models can handle many everyday business tasks, such as drafting, summarizing and answering questions from documents. Running them yourself means you buy hardware instead of renting a subscription, and you decide who can use the system and under what rules.

What We Built

Six pieces:

  • Hardware. Purchased and connected to the company network.
  • vLLM with open-source models. vLLM is an open-source server that runs language models on your own hardware and serves them to the whole team.
  • Single sign-on. Only people with a PCS Global email ID can sign in.
  • Guardrails and cybersecurity. Set up so the company has enterprise control over how the system is used.
  • A knowledge bank. Company information that gives the AI full context, and that users can reference while they prompt.
  • Approved email templates. Email drafts the company has pre-approved.

The assistant is reachable only from inside the network.

Control and Security

Data control is built in from the start, not added on.

Sign-in runs through single sign-on, tied to company email IDs. Guardrails set the boundaries for what the AI does. And because the system sits on PCS Global's own network, the company, not an outside AI provider, decides who has access and which rules apply.

A Knowledge Bank That Knows the Company

A general chatbot knows nothing about your clients, products or policies. So people often paste that context in by hand, which is also how sensitive data can end up in the wrong tool.

The knowledge bank holds company material in one place. The AI draws on it for full context, and any user can reference it directly while writing a prompt. That should mean better answers without copying documents into a chat window.

Approved Email Drafts

Staff also get email drafts built from templates the company has pre-approved.

The company decides the approved wording, so client emails start from text it has signed off.

Before and After

AreaBeforeAfter
AI toolsFree public AI toolsPrivate open-source models on hardware on the company network
Company dataSome emails and client data shared with outside models by mistakePrompts stay on the company network
Rules and accessNo rulesSingle sign-on with company email, plus guardrails
CostPlan: more than $25,000 a year for AISaves $20,000 a year

PCS Global now has a private AI inside its own network, with a knowledge bank and approved email templates. It saves $20,000 a year; the plan had been to pay more than $25,000 a year for AI.

Is Private AI Right for Your Company?

It tends to fit when:

  • Staff already use public AI tools and you can't say what has been shared.
  • Paid AI for a team your size can run to tens of thousands of dollars a year.
  • You handle client or financial data you are expected to protect.
  • You would rather own the system than rent it.

It has trade-offs. You buy and maintain hardware, and open-source models can trail the best paid models on the hardest tasks. For everyday drafting, summarizing and answering from company documents, they are often enough.

Not sure where you stand? Take the free AI readiness quiz, see what a private executive workshop covers, or read more about AI implementation and AI automation.

Frequently Asked Questions

What is vLLM?

vLLM is an open-source inference server. It runs large language models on your own hardware, typically GPUs, and exposes them through an API that your tools and chat interface can call.

Why not just buy enterprise AI licenses?

That is a valid route, and right for some companies. For PCS Global, the plan to pay for AI was more than $25,000 a year, and paid cloud AI typically processes data on the vendor's servers. Running it on their own network saves them $20,000 a year and keeps control with the company.

Can the AI use our own documents?

Yes. A knowledge bank holds company material. The AI draws on it for context, and users can reference it directly in their prompts.

Does everyone need to be on the company network?

The assistant is reachable only from inside the network. Access for remote staff is a network policy decision, usually handled with a company VPN.

How much does a private AI setup cost?

It depends on how many people use it, which models you want and what hardware that needs. We size it on a free triage call.

Is your team using AI without rules?

We design and build private AI setups with single sign-on, guardrails and a company knowledge bank. Start with a free 30-minute triage call.