Private AI analysis for data you can't send to ChatGPT
Some data is too sensitive for public AI tools. Think of cost sheets, customer margins, client files or contracts that ban third-party AI. For those jobs, we run AI models on infrastructure Push controls, so your data never goes to a public AI service.
Free, 30 minutes, and you'll leave with at least one clear next step.
What does private AI mean at Push?
Private AI means the AI model runs on infrastructure Push controls, and your data never goes to a public AI service like ChatGPT, Claude or Gemini. The model reads your files on our own machines, and only the results come back to you.
Most of our AI work uses cloud models, because they're the strongest and fastest option. Private AI is a side offering for cases where cloud AI isn't allowed. Push runs a variety of AI models privately, with our own internal tools and workflows around them, so we can handle many kinds of data, from spreadsheets and system exports to contracts and scanned documents.
This page covers one-off analysis projects. For AI built into daily work in HubSpot, see custom AI workflows or our AI agents service. For a wider plan, start with AI consulting.
Who usually asks for private AI?
Companies with cost and pricing data
True costs, margins by customer, rebates and price history are the business. Many owners never get time to study them because the files feel too sensitive to share.
Firms with client files
Professional services firms, agencies and others hold client documents they've promised to protect. Some clients ask point blank where their files go.
Companies bound by contract
Some customer, supplier or government contracts ban sending data to third-party AI. Private AI keeps the data away from AI companies. We check with you first that the contract lets Push handle it at all.
Regulated or security-minded teams
Some owners want a short, clear answer to where their data went. When the work stays on infrastructure we control, that answer is easy to give.
Which analysis jobs do clients bring us?
Pricing and margin review
Read years of quotes, invoices and price lists to find customers below target margin and discounts that crept in without approval.
Contract review at volume
Pull payment days, price escalation clauses, renewal dates and liability caps from hundreds of contracts into one table.
Quote and proposal history
Sort past quotes or proposals by won and lost, and find patterns in timing, size or customer type that explain why some close.
CRM and back-office exports
Study HubSpot, NetSuite or QuickBooks exports side by side, matching deals to orders or invoices and finding customers who stopped buying.
Support history
Group tickets and service notes by product or service, find repeat problems and see which issues take longest to close.
What is private AI best for, and what isn't it?
Good fit for private AI
- Sensitive data that a contract, customer or policy keeps away from AI companies
- One-off or periodic analysis, like a margin review or a contract sweep
- Focused jobs such as pulling fields, sorting records and grouping notes
- Work where waiting hours or days for results is fine
- Questions with a clear answer we can check against your source files
Better handled another way
- Data that cloud AI with the right business terms can already handle
- Live tools that sales reps or customers wait on in real time
- Hard, open-ended reasoning where top cloud models are clearly stronger
- Small jobs where setup would cost more than the answer is worth
- Daily AI inside HubSpot, which suits custom AI workflows better
Which fits your data, cloud AI or private AI?
Cloud AI (business plans or APIs)
- Uses the strongest models available
- Fast, even on large batches
- Setup often takes days
- Usually cheap for small and medium jobs
- Data goes to the AI provider under their business terms
Private AI on Push infrastructure
- Uses open-weight models, capable but generally behind the top cloud models
- Slower, so big jobs take longer to run
- More setup for data transfer and model preparation
- Costs more for the same amount of work
- Data is never sent to a public AI service
What do you give up with private AI?
You gain control over where your data goes. You pay for it in model strength, speed, cost and setup time.
Model strength. The best cloud models are still ahead of the open-weight models we can run privately, especially on long, tangled reasoning. On focused jobs like pulling fields from contracts or sorting records, the gap is much smaller.
Speed. Our infrastructure is smaller than a cloud provider's, so a job that takes minutes in the cloud may take hours. That's fine for analysis. It's not fine for a live tool reps wait on.
Cost and setup. We have internal tools and workflows for moving data in, preparing it and checking results, and they adapt to many data scenarios. Each project still needs time to agree on the data, move it safely and pick the right model for it. So a small job costs more than it would on a cloud AI service.
How does data get in and out, and who can see it?
You decide exactly what we receive. It moves over a secure channel agreed with your IT team, and only the Push people on your project can reach it. Often names, emails or account numbers can be removed first, so the model sees only what the analysis needs.
Results come back as a report, a spreadsheet or fields ready to load into your systems. At the end, we delete your raw files and working copies. If you need a specific certification or audit report from us, raise it on the first call so we can tell you right away whether we fit.
Every project starts with the smallest set of data that answers the question
The less data that moves, the less there is to protect.
Stage 1: Scope the question and the data
We agree the question, the files that answer it and the fields to leave out.
We start with the decision, such as which customers to reprice or which contracts renew next year. From there, we list the exact exports needed, cut unused fields and note what your contracts say about handling the data.
You get A written scope of question, files, fields and retention.
Stage 2: Move the data securely
Your team sends the agreed files over a secure channel, and we check they arrived complete.
We agree the transfer method with your IT lead, and your team masks personal details where possible. We check row counts and dates on arrival, so the analysis isn't built on a partial export.
You get Confirmation of exactly what we received.
Stage 3: Test the model on a sample
We run the model on a small sample and check its answers by hand.
We try the job on a sample, compare the answers with what a person would say, and adjust the instructions or model. If private AI can't do the job well enough, we tell you here, before the bigger spend.
You get A short accuracy note with examples.
Stage 4: Run the full analysis
The model works through the full set on our infrastructure, and we review the output.
We spot-check results across the whole set and send unclear cases to a person. Numbers that drive decisions, like margin, are checked against the source files.
You get Checked results, with unclear cases flagged.
Stage 5: Deliver results and delete the data
You get the findings, then we delete your data and confirm it.
We walk your team through the findings. Results come as a report, a spreadsheet or fields ready for HubSpot or your back-office system. Once you have them, we delete your raw files and working copies.
You get A findings report, result files and deletion confirmation.
Can private AI work with HubSpot or back-office data without moving everything?
Yes. We pull only the records and fields the question needs, usually as a filtered export, and send results back as a few fields. Your HubSpot account and other systems stay where they are.
A margin review might use two years of closed deals from HubSpot and invoice lines from your accounting system, with contact names removed. Back in HubSpot go one or two company properties, like a margin band or a reprice flag, for reps and reports to use.
Connecting HubSpot to back-office systems is regular work for us. We tied one client's sales process to its ERP, for example. See our guide to connecting HubSpot to an ERP.
When is cloud AI with the right agreements enough?
For most business data, a cloud AI business plan or API with the right contract terms is enough, and you get better models for less money. Major providers offer business terms on how long data is kept and whether it's used for training. Check your provider's current terms.
Choose private AI when a contract, a key customer or your own policy says the data can't go to an AI company. Or choose it when the data is so central to the business that you want it kept off AI companies' servers.
If you're not sure which side you're on, an AI readiness assessment checks whether your data and team are ready for AI. A Friction Point Analysis looks wider, at every place work slows down across the business.
Examples by industry
Sensitive data looks different in every business. These pages show a few cases.
- Manufacturing: drawings, bills of materials and true unit costs.
- Distribution: account pricing, rebates and margin by customer.
- Ecommerce: supplier costs and customer order history.
Private AI questions
Where does our data live during a private AI project?
On infrastructure that Push controls. The model runs there too, so your files never go to a public AI service. Only the Push people on your project can reach the data.
What happens to our data when the project ends?
We agree how long we keep it during scoping. When the work is delivered, we delete your raw files and working copies and confirm it with you. You keep the results.
Which AI models do you use for private AI?
We run a variety of AI models, so we don't force every job onto one. We test on a sample of your files and pick the model that handles your kind of data best. Our internal tools and workflows let us work with many data scenarios, from exports to contracts.
Is a private model as accurate as ChatGPT?
Not on everything. Top cloud models are still stronger at hard, open-ended reasoning. On focused jobs like pulling contract terms or sorting records, a good open-weight model can come close. We test on your files first.
Can private AI connect to our HubSpot account?
Yes, without moving everything. We analyze a filtered export of the records the question needs, then write a few result fields back to HubSpot, such as a margin band or renewal flag.
Is private AI more expensive than cloud AI?
Usually, for the same amount of work, because each project needs more setup and runs on smaller infrastructure. When cloud AI is ruled out, that cost buys analysis you couldn't otherwise do. We scope after the first call.
How do we find out if private AI fits our data?
Book a 30-minute call. We'll talk through your data and the question you want answered, and tell you whether private AI is a good fit or whether another approach would serve you better.
Related
More in AI services
Book a call to talk through your data
In 30 minutes, tell us about the files and the question you want answered. We'll talk it through and tell you honestly whether private AI fits, or whether cloud AI with the right terms is enough.