How to Measure AI ROI at a Portfolio Company

    Bijan Sirleaf
    Last updated November 10, 2026
    6 min read

    # How to Measure AI ROI at a Portfolio Company

    Short answer: Measure AI ROI one workflow at a time. Before anything changes, record a baseline the operator already trusts, with a clear denominator such as eligible requests or open estimates. Run the workflow for a fixed window, compare the same measure, and track quality, exceptions and the full cost of running it. Then let the finance owner decide whether freed capacity or recovered revenue actually reaches the P&L.

    "What's the ROI?" is the first question any investment committee asks about AI, and most answers are not defensible. Time saved gets multiplied by a salary and called savings. Faster responses get called revenue. A pilot report that does this will not survive diligence. This guide shows a more careful method for service businesses: accounting firms, legal services companies and home services operators.

    Why is AI ROI hard to measure?

    Three reasons come up again and again:

    1. No baseline. Nobody measured the workflow before the change, so the comparison is against memory.
    2. The wrong unit. Measuring "AI usage" or "tasks automated" says nothing about the business. The unit has to be the workflow's output.
    3. Capacity is not cash. If a workflow saves ten staff hours a week but nobody's hours change and no extra work gets done, the P&L does not move.

    McKinsey's June 2026 analysis of PE-backed companies makes a related point at portfolio level. It advises PE firms to establish a clear baseline on where portfolio companies use AI and whether those uses generate sustainable value before pursuing bigger moves (McKinsey).

    What should you measure?

    Pick measures tied to the workflow, with a denominator, that the operator already recognizes. Examples:

    WorkflowPrimary measureDenominator
    Accounting document collectionPackets complete by the target dateEligible client packets
    Legal intake preparationIntake summaries ready for lawyer review within one business dayEligible inquiries
    Home services estimate follow-upOpen estimates older than 14 daysAll open estimates
    Membership renewalsRenewals completed before lapseMemberships due in the window
    Invoicing and collectionsInvoices past due over 30 daysInvoices issued

    Next to the primary measure, always track:

    • •Quality: errors found by reviewers, corrections made, complaints.
    • •Exceptions: cases the workflow escalated or could not handle.
    • •Staff time: review minutes per item, not just work removed.
    • •Volume: how many items ran through it, so you know the result is not noise.

    No measurable baseline means no defensible improvement claim.

    How do you calculate ROI from a workflow pilot?

    Keep the calculation honest and show your work:

    1. Cost side. Include implementation, software and usage, integration work, monitoring and support, and the internal time spent reviewing outputs and handling exceptions.

    2. Benefit side, by type.

    • •Recovered revenue: for example, estimates closed or renewals saved that would have lapsed. Compare against the baseline period, and note seasonality.
    • •Capacity released: hours no longer spent on the workflow. Count it as capacity, not cash, until a decision turns it into either lower cost or more work handled.
    • •Risk reduced: fewer missed deadlines or errors. Usually qualitative unless you have a cost per error.

    3. Attribution. Ask what else changed in the window: pricing, staffing, season, marketing. If you cannot separate them, say so.

    4. Decision. The finance owner decides whether the measured change affects the P&L, and how much of it to underwrite for the next company.

    Does AI improve EBITDA or exit value?

    It can, but treat both claims with care. McKinsey's analysis of 471 PE-backed companies found that companies embedding AI in products and building new AI businesses traded at higher median revenue multiples. It also found the difference between companies using AI for productivity and companies using it in their operating model was negligible (McKinsey). That is correlation in one sample, not a promise for any single company.

    For most service businesses, the defensible story at exit is simpler: documented workflows, measured before and after, that a buyer can see working and that do not depend on one person. That evidence comes from measuring each workflow properly now.

    HFS Research's 2026 survey points the same way. PE buyers said they trust specific, measurable claims, and the firms pulling ahead demand quantified baselines and repeatable delivery (HFS Research).

    How long does it take to see AI ROI?

    It depends on volume and season. Nimbus Agents uses a 30-day measuring window after setup because it gives most weekly workflows enough volume to compare against the baseline. A low-volume workflow, or one in a seasonal business such as tax preparation or HVAC, may need longer. Decide before the pilot starts what happens if the evidence is not enough. No ROI timeline is guaranteed.

    What does a good AI pilot report contain?

    • •The workflow, the company and what was excluded.
    • •The baseline, how it was counted and over what period.
    • •The result over the window, with volume.
    • •Quality and exceptions.
    • •The full cost to run it.
    • •What changed besides the AI.
    • •A recommendation: continue, revise or stop, and whether to reuse it at a similar company.

    Our private equity page shows a sample pilot report layout, clearly marked as example data rather than a client result.

    Next step

    If you need numbers you can put in front of an investment committee, start with one workflow. Talk to Nimbus Agents about scoping a measured pilot, and see how we report results on Nimbus Agents for private equity.

    Related resources

    Sources

    • •McKinsey, "Beyond productivity: How AI creates value in private equity," June 23, 2026: https://www.mckinsey.com/capabilities/business-building/our-insights/beyond-productivity-how-ai-creates-value-in-private-equity
    • •HFS Research, "Make Services-as-Software the new PE operating model for AI," September 2, 2026: https://www.hfsresearch.com/research/sas-pe-operating-model/

    Frequently asked questions

    How do you measure the ROI of AI?

    Pick one workflow, record a baseline with a clear denominator before anything changes, run the workflow for a fixed window, compare the same measure, and include quality, exceptions and the full cost of running it.

    What metrics should a PE firm use for AI at portfolio companies?

    Metrics tied to each workflow's output that the operator already trusts, such as complete packets, open estimates past a deadline, response time or past-due invoices, alongside quality, exceptions and review time.

    Are hours saved the same as cost savings?

    No. Hours saved are capacity. They become savings only if staffing or spend changes, or become revenue only if the team handles more work. The finance owner should make that call.

    How long does it take for AI to show ROI?

    It depends on volume and seasonality. A 30-day window works for many weekly workflows, but low-volume or seasonal workflows may need longer. No timeline is guaranteed.

    Does AI increase a company's valuation at exit?

    Research such as McKinsey's 2026 analysis links deeper AI adoption to higher revenue multiples in its sample, but that is correlation. A buyer will look for documented, measured workflows rather than AI claims.

    Contact us

    Tell us what you are trying to automate and we will get back to you.