AI for private equity: the first 90 days
AI for private equity, in a $5M to $50M manufacturer or distributor, starts with the business you bought. In the first 90 days, you find out what its records hold and take quick wins on margin, pricing, working capital and reporting. You also set the rules that keep its data safe. This guide sets out that plan and the AI diligence questions to ask. The same plan works for a next-generation owner taking over a family business.
Why the first 90 days matter
A change of owner is a natural point to change how work gets done. You are reviewing every process anyway, and your team expects some change. Each later AI project reads the records and follows the rules you set up in these weeks.
Owners of manufacturers and wholesalers are among the most likely to plan a sale. In Statistics Canada’s third-quarter 2026 survey, 27.7% of manufacturers and 27.3% of wholesalers said they intend to sell or transfer within 10 years. Across all businesses, the figure was 20.3%.
Few of these businesses use AI in their operations. In Statistics Canada’s second-quarter 2026 survey, 13.1% of manufacturers and 7.9% of wholesalers had used AI in the previous 12 months. The survey counts AI used to produce goods or deliver services. Across all businesses, the figure was 19.2%, according to Statistics Canada’s analysis of the survey.
Private equity firms are backing this work directly. On May 4, 2026, Anthropic, Blackstone, Hellman & Friedman and Goldman Sachs announced a new AI services company for mid-sized companies. Anthropic’s announcement names mid-sized manufacturers among the companies that stand to gain from AI.
The data you inherit
In a manufacturer or distributor of this size, the records sit in a few systems and with a few people. Expect to find most of these.
| Source | What it holds | What to check first |
|---|---|---|
| ERP | Orders, shipments, inventory and purchasing, and in a plant, routings and work orders | Its version, who can change it, and whether it runs on a local server or in the cloud |
| Accounting system | Invoices, credit notes, payables and the general ledger | Whether it is separate from the ERP, and how the two are reconciled each month |
| Spreadsheets | Price lists, rebates, costing, forecasts and the reports the ERP does not produce | Which ones the business depends on, and who keeps each one up to date |
| Quotes, customer commitments, supplier terms and complaints | Which mailboxes hold customer and supplier history | |
| People | Pricing rules, customer exceptions and how jobs really run | Who would be hard to replace, and what only they know |
Find the person everyone asks, such as the estimator or the office manager who knows where every number comes from. Spend an hour with them in the first week and write down what they tell you.
Do not start by cleaning everything. Copy the records you need into one database, read-only, so nothing in the live systems changes. Then write down what your terms mean, such as what counts as a sale and which date makes an order late. Those definitions decide whether two people asking the same question get the same answer.
Derik at Thrive was instrumental in taking incredibly messy data we inherited in a business we acquired and, through using AI, organized it in record time in a way that made it reviewable by our team for final review and approval.
Rios-Karim Mercier, Belmont Capital.
Once the records sit in one place, your team can ask questions of them in plain language. AI reporting shows how that works, and how to check each answer.
Quick wins on margin, pricing, working capital and reporting
Pick wins that use records the business already keeps, where a person can check the result against a number they trust. These four suit a manufacturer or a distributor.
Margin
Build margin by product, customer and channel from sales and cost data, and keep it current. It needs invoice lines with their cost, plus credit notes, freight and rebates, which may sit in a spreadsheet. Check that the total ties to the income statement before anyone acts on it. Then look first for customers on old prices and items sold below current cost. Margin analysis covers the method, including a price, volume and mix bridge.
Pricing
Compare every active price with current material and labour cost, and draft a new price wherever the margin has slipped. Start with the customers whose prices have not moved since the last cost increase. A person approves each change, and each customer’s contract decides how much notice the change needs. Cost-plus pricing covers the formulas, and price optimization covers pricing from your quote history.
Working capital
In a goods business, start with inventory. Rank items by how long they have sat, flag stock that is still being reordered without sales, and set reorder points from real demand. Inventory optimization has the formulas. On receivables, a weekly list of overdue invoices, with the last contact on each, gives whoever collects a short list to work through. On payables, compare the terms each supplier offers with the terms you actually pay on.
Reporting
Build the weekly pack for the owner and the board from one database. The numbers then match, and each one traces back to its records. For a goods business, the pack can show sales and margin against last year, the order backlog, overdue receivables and inventory by age. Month-end reconciliations, variance notes and cash forecasts can be drafted for the controller to check, as AI in finance describes.
Ask which quick win the records support
Tell Derik what the business makes or distributes and which systems hold its records. He will tell you which quick win its data can support first.
Start a conversationA 90-day plan
The plan below fits a manufacturer or distributor with one ERP, one accounting system and a handful of key spreadsheets.
| Days | Work | What you have at the end |
|---|---|---|
| 1 to 30 | Map the systems and the people who run them. Copy the records read-only into one database, write the first definitions, and give staff one approved AI tool with a one-page policy. | A map of the data, a first margin report and written rules for AI use |
| 31 to 60 | Review prices against current cost, rank slow-moving stock and start the weekly overdue-invoice list. Train each desk on its own work. | Price changes ready to approve, a stock list to act on and a collections list |
| 61 to 90 | Automate the weekly reporting pack, and put one routine job, such as order entry, into daily use with a person approving each draft. Review the AI register with the board. | A weekly pack that runs on its own, one job in daily use and a register the board has seen |
Keep the order even if the dates slip, because every later step reads the same records and follows the same rules. For the approved tool in the first month, Claude for business covers the plans and the settings to change.
What to check in AI diligence
AI diligence asks what AI the business already uses and what its records could support. Ask before closing if you can, or in the first two weeks after.
| Question | Why it matters | A good answer |
|---|---|---|
| Which AI tools do staff use, and on which accounts? | On personal Claude plans, each person decides whether their chats can train models, and the company has no admin control | One company account, with personal accounts closed or moved into it |
| What company or customer data has gone into those tools? | Under clause 4.1.3 of PIPEDA, a business stays responsible for personal information it transfers to a third party for processing | A list of tools, the data each one received and the terms that apply |
| Who owns the code and automations that contractors built? | Under section 13(4) of the Copyright Act, an assignment of copyright is valid only in writing, signed by the owner | Signed assignments, with code and accounts in the company’s name |
| Which software and AI contracts hold company data? | They set retention, training use and where data is stored | Terms on file for every vendor |
| Does anything send, order or change records without a person approving it? | An automation with no owner can keep running after the person who built it leaves | A named owner and an approval step for each automation |
| Do the seller’s documents claim AI? | Claims need checking on the business’s own records | A demonstration on live records, checked against a known answer |
| Where does the know-how sit? | Pricing rules and customer exceptions can live in one person’s head or one spreadsheet | Written rules, or a plan to capture them before that person moves on |
The AI readiness assessment scores the records side of this in twelve questions you answer in a browser. To find tools that staff use on personal accounts, see shadow AI.
Governance in a portfolio company
At this size, AI governance fits on a few pages. It needs four things, each with a named owner.
- A register of every AI tool and automation: what it does, which data it reads and who approves its output. AI governance shows a one-page version.
- A policy that names the approved tools and the records that stay out of them, starting from the AI policy template.
- Accounts in the company’s name, with single sign-on and the settings in the secure AI checklist.
- An approval step for anything that reaches a customer, a supplier or the ERP, as human in the loop describes.
Each quarter, give the board one page that shows:
- The tools in use and the data each one reads.
- Each job’s result against its baseline.
- Any output that was wrong, and any data that went where it should not.
- The next job on the list.
Where the data sits is a board question too, and private AI for business covers the options for keeping records in Canada.
What can go wrong in the first 90 days
- Too many projects at once. Each project needs the same few people who know the records. Run one at a time until the database and the definitions are in place.
- Numbers nobody checked. A report can look clean and still be wrong. Tie every new report to a figure the controller trusts before it reaches the board.
- Company data in personal accounts. Without an approved tool, staff may paste company records into personal AI accounts. Give them a company account in the first month, with the policy beside it.
- Losing the person who knows. Write down what the key people know in the first month, and involve them in testing the new tools.
New owners and next-generation successors
A private equity team arrives with a playbook and learns the records as it goes. A son or daughter taking over the family business knows the customers and the team, and inherits systems that grew around the founder. Both start with the same two steps: put the records in one place, then take quick wins from data the business already has.
A successor should add one step. Write down the founder’s pricing rules and customer exceptions while the founder is still there, and check them against past quotes.
Sellers care how the change lands. Among owners who intend to sell or transfer within two years, 51.7% told Statistics Canada that protecting employees was very important. Introduce AI as help with routine work, with your people approving what it drafts.
How ThriveAI helps
ThriveAI is an AI engineering company in Ottawa. It builds private AI systems on a company’s own data, for businesses that make, move or sell physical goods. It works with new owners and private equity teams running $5M to $50M businesses. Its quick wins are the reports your ERP won’t give you, margins, prices that follow your costs, a faster month-end close and demand forecasts.
Stage one is the job costing you the most, built for a fixed price and running on your own systems within weeks. If you stop there, it is yours to keep. The platform is designed to keep each client’s data on its own server in Canada. You choose a model on that server or a hosted model under a written zero data retention agreement. A hosted model may process requests outside Canada. Derik Lawlis, the founder, leads every project and stays close to the build.