AI use case prioritization framework: a scoring matrix for portfolio companies
An AI use case prioritization framework scores every candidate initiative against the same weighted criteria, here EBITDA impact, time to value, data readiness, talent need and risk, then ranks them by total. This template also scores an option most matrices leave out: licensing a portfolio company's existing records to AI buyers, alongside internal automation projects.
When to use this framework
Use it whenever a portfolio company has more AI ideas than budget and management attention: during the 100-day plan, at the annual budget, when the value creation plan is refreshed, or after a run of vendor pitches. One shared scoring method lets a head of value creation compare a distributor's collections agent with a software company's support copilot on equal terms.
The bar for what counts as worthwhile has risen. Bain's Global Private Equity Report 2026 estimates that a deal which needed 5% EBITDA growth a decade ago to reach a 2.5x return over five years now needs about 12%, and that buyout holding periods at exit are around seven years. AI initiatives have to earn their place on EBITDA, not novelty.
The scoring matrix
Score each use case from 1 to 5 on five criteria, multiply each score by its weight, add the results and divide by 100. The total is a figure out of 5. The default weights suit a mid-hold operating company; the adjustment table further down changes them for other situations.
| Criterion | Default weight | Score 1 | Score 3 | Score 5 |
|---|---|---|---|---|
| EBITDA impact | 30 | Under {low_impact_threshold} a year | Between the two thresholds | Over {high_impact_threshold} a year, or one-time cash of similar size flagged as non-recurring |
| Time to value | 20 | More than 18 months | 6 to 18 months | Within the current fiscal year |
| Data readiness | 20 | Records missing, scattered or of unclear ownership | Records exist but need cleanup or a rights review | Records exist, are structured, exportable and owned |
| Talent need (inverse) | 15 | A new team or scarce hires required | Some new skills, or a vendor plus an internal owner | The existing team or an outside specialist does most of the work |
| Risk (inverse) | 15 | Customer-facing errors, regulated data or contract exposure | Manageable with controls | Internal, reversible, little sensitive data |
Treat anything scoring below {cutoff_score} as parked for this planning cycle, and write down why.
The use case card
Fill one card per candidate before anyone scores it, so every number has evidence behind it.
Scoring data licensing as an extra row
Most matrices list only things the company would build or buy. Add one more candidate wherever a portfolio company has years of its own operating records: licensing those records to AI labs and data buyers through SourceX. It is scored on the same five criteria, read slightly differently.
| Criterion | How to score a data license | Evidence to collect |
|---|---|---|
| EBITDA impact | A one-time license payment, not run-rate; flag it as non-recurring | Interest confirmed after qualification, never an assumed price |
| Time to value | Once the company is deal-ready, buyers typically respond within about two weeks, and payment typically arrives within about 60 days of invoicing once a buyer selects the data | Inventory status and sponsor availability |
| Data readiness | The same test as automation: years of records across systems, exportable, with outcomes | A completed data inventory |
| Talent need | Low: SourceX handles qualification, buyer review and contracting; the company supplies a sponsor and an export owner | A named export owner |
| Risk | Rights, confidentiality, exclusivity and any sale process | Contract and privacy review |
Two cautions keep the score honest. First, revenue recognition for a license depends on its terms. Deloitte's ASC 606 licensing roadmap explains how a license is assessed as a right to use intellectual property, recognized at a point in time, or a right to access it, recognized over time, so ask the company's auditors before booking anything. This is general information, not legal, tax or financial advice. Second, some companies screen out entirely: check which companies cannot license their data and the data rights documentation guide before scoring risk above 3.
Illustrative scored example
Illustrative: a fictional industrial distributor with about 180 full-time employees, nine years of ERP and helpdesk history and a lean finance team. The scores are invented to show the arithmetic, not a forecast.
| Use case | EBITDA (30) | Time (20) | Data (20) | Talent (15) | Risk (15) | Weighted total |
|---|---|---|---|---|---|---|
| AP invoice exception agent | 3 | 4 | 4 | 3 | 4 | 3.55 |
| Collections and dispute triage agent | 4 | 3 | 3 | 3 | 3 | 3.30 |
| Demand forecasting model | 4 | 2 | 2 | 2 | 3 | 2.75 |
| Sales call summaries in the CRM | 2 | 5 | 3 | 4 | 4 | 3.40 |
| License order-exception and support history | 3 | 4 | 4 | 5 | 3 | 3.70 |
The license ranks first here because it uses records the company already keeps and needs little new talent, but its effect is one-time, so a sensible board pairs it with the AP agent rather than choosing between them. The two also help each other: cleaning exception codes for the agent improves the dataset. The guide to back-office AI agents in portfolio companies explains why those exception records matter to builders.
How to adjust the weights
Weights always add up to 100. Change them per company, not per use case, so every candidate in one company is judged the same way.
| Company situation | Weights: EBITDA / time / data / talent / risk | Why |
|---|---|---|
| Mid-hold default | 30 / 20 / 20 / 15 / 15 | Balanced view for an established company |
| First 12 months of the hold | 35 / 20 / 20 / 15 / 10 | Run-rate gains compound over the remaining hold |
| Exit expected within 18 months | 30 / 30 / 15 / 10 / 15 | Only results visible before the data room opens count; check how any exclusive license fits the sale |
| Regulated data, such as healthcare administration or financial services operations | 30 / 15 / 15 / 10 / 30 | Privacy and contract exposure can outweigh savings |
| Thin finance or IT team | 30 / 15 / 15 / 30 / 10 | Projects stall without owners |
| Platform mid-integration | 30 / 15 / 30 / 10 / 15 | Records are moving between systems, so readiness decides what is feasible |
Review cadence
- Score at the start of the planning cycle with the CEO and CFO, and record the evidence on each card.
- Re-score quarterly alongside the board pack, changing a score only when the evidence changes.
- Pause any use case that stays below {cutoff_score} at two consecutive reviews.
- Re-run the data readiness column after every system migration, because migrations can destroy history.
What never to put in the matrix
- Vendor ROI claims that have not been tested against the company's own baseline.
- A data license counted as recurring EBITDA or as a synergy in the base case.
- An assumed license price or a partner reward amount; neither is known until terms are agreed.
- Samples of customer records, tickets or contracts. The matrix describes systems, not their contents.
- Use cases with no named executive owner.
Next step
Score the top five use cases plus the licensing row for one portfolio company this quarter. If the license scores well and the company meets the baseline on who qualifies, register as a partner and introduce the CEO or CFO; the company can also apply at sourcex.si/apply. For wider context, the AI value creation playbook covers the buy side and how to sell data to AI companies covers the supply side.
- Step 1Share your linkSend your personal link to a company you know.
- Step 2Company appliesThe company applies itself at /apply.
- Step 3Buyer selects and paysThe buyer selects and pays for the data and SourceX receives its fee.
- Step 4You get your rewardYour share of SourceX fees becomes payable.
Common questions
How many AI use cases should a portfolio company run at once?
As many as it can staff with a named executive owner and the people to do the work, which for most mid-market companies is a small number. The talent column in the matrix usually sets the limit. Running fewer use cases to completion produces evidence for the board; running many in parallel tends to produce a pile of pilots.
Should data licensing sit inside the value creation plan?
Yes, as a scored option with a clear non-recurring flag. Leaving it out means the board never compares it with automation projects that use the same records. Keeping it out of run-rate EBITDA and the synergy case keeps the plan honest, because a license is a one-time payment whose timing depends on terms and buyer selection.
Who should do the scoring?
The head of value creation facilitates, but the CEO, CFO and the functional owner of each use case should score independently first and then reconcile differences in one meeting. Independent scores expose optimism quickly. Record the evidence for each score on the use case card so a later reviewer can see why a number was chosen.
How do you estimate EBITDA impact before a pilot?
Start from a measured baseline, such as hours spent on invoice exceptions or days sales outstanding, and apply a conservative improvement range rather than a vendor's headline figure. Score the low end of the range. After a pilot, replace the estimate with observed results and re-score; a use case that only works at the high end should drop down the list.
Can one matrix compare use cases across a whole portfolio?
Yes, if every company uses the same criteria and scoring definitions. Weights can differ by company situation, so compare weighted totals within a company and raw criterion scores across companies. A portfolio summary listing each company's top three use cases, plus whether its licensing row scored well, gives the investment committee a quick view.
Related pages
- Which portfolio companies are not a fit for data licensing?
- How to document data rights and provenance before licensing data for AI training
- Back-office AI agents in portfolio companies: the records that train them
- Which US businesses are a fit for a SourceX data licensing introduction
- AI value creation in private equity: a playbook for operating partners
- How to sell data to AI companies
Free resources
- NPV calculator — Net present value with a discounted cash flow table.
- Time value of money calculator — Future and present value with optional regular payments.
- Business DSCR calculator — Debt service coverage from cash flow and loan terms.
- All free tools · MCP resource center
By SourceX Partnerships Team · Published 2026-10-09 · Updated 2026-10-09
Know a US company with valuable proprietary data?
Become a referral partner from anywhere we support, get your link and introduce an owner or authorized decision-maker.
Refer a company →I own a business
Explore licensing your company's data to AI developers worldwide. Start a short assessment; no uploads needed.
Start an assessment