Maybank Advisors · Professional services AI operating series
01 · The economics
The economics of the AI transition in professional services
Maybank working thesis
Professional services firms may have a limited period in which AI-assisted delivery becomes materially more efficient before client expectations, competitive pricing and renewal cycles fully absorb that efficiency.
Our current working estimate is roughly 18–24 months. It is a Maybank estimate — not an industry benchmark or forecast — and the period could be materially shorter in a fast-moving market.
The important point is not whether the window closes in month 17 or month 26. It is the economic mechanism underneath it: your firm sells professional work. AI can reduce the time and coordination required to produce some of that work. Your commercial model determines who keeps the gain.
Who this is for. Managing partners, owners, COOs, CFOs and service-line leaders at firms that make money by converting professional expertise and capacity into client revenue.
White paper · 2026
Most professional services firms still have an economic model connected, directly or indirectly, to human time.
The contract may say hourly, day rate, retainer, project, fixed fee or value-based fee. But underneath many of those arrangements sits a familiar production equation:
people × available capacity × delivery time → client work → revenue
Artificial intelligence changes one of those variables. It can reduce time spent on activities such as:
It does not remove professional judgment. But it can change how much labor and coordination sit around that judgment. That creates an economic question before it creates a technology question:
If the work becomes cheaper or faster to produce, who captures the difference?
Recovered capacity does not automatically become profit.
AI-assisted delivery efficiency is an input. Four destinations, chosen or defaulted into.
AI-assisted delivery efficiency
An input, not an outcome
Margin
Same work, lower delivery cost.
Capacity / volume
The same team handles more work.
Price
Efficiency is shared with the client through lower fees.
Absorbed time
Capacity disappears into existing operations.
AI creates capacity. Management decides whether that capacity becomes an economic result.
Maybank framework
That decision is the heart of the transition.
Legal services provide unusually good public data on the conversion of professional time into cash.
Clio’s 2025 industry data reports average US law-firm utilization of 38%, realization of 88% and collection of 93%. Applied to an eight-hour working day:
An eight-hour working day, converted to cash at published US law-firm averages.
8.0 hours worked
3.0 billable hours
38% utilization
2.6 hours invoiced
88% realization
2.4 hours collected
93% collection
Roughly 30% of the eight-hour day reaches collected revenue.
A legal-industry example, not a benchmark for every professional services sector.
Source: Clio, 2025 Legal Trends Report / 2025 law-firm KPI benchmarks. Utilization 38%, realization 88%, collection 93%.
Published research
The reason it matters is structural. A firm can recover time at several points in the operating model, but the commercial value depends on what happens after the time is recovered.
Thomson Reuters’ 2025 Future of Professionals research surveyed 2,275 professionals across legal, tax, accounting, audit, risk and trade.
Respondent expectations and Thomson Reuters estimates, compounded from the weekly figure.
5 hours
expected time saved per professional per week
240 hours
per year
~$19,000
estimated annual value per professional
~$32B
estimated combined annual impact across US legal and tax/accounting
These are respondent expectations and Thomson Reuters estimates — not measured Maybank client outcomes.
Source: Thomson Reuters, Future of Professionals Report 2025.
Published research
Even so, the expected magnitude is large enough to force a commercial question. If an hourly engagement takes fewer hours, the client may simply receive a smaller invoice. The firm did the innovation. The client captured the economics.
This is why AI efficiency and commercial strategy cannot be separated.
Neither axis is sufficient on its own.
Efficiency without capture
The work gets faster while revenue remains directly connected to time. Some of the value can move to the client automatically.
The gain can be captured
Delivery efficiency and a commercial structure capable of converting some of it into margin, capacity or strategic pricing.
Exposed and standing still
Neither captured efficiency nor revenue insulated from time compression.
Commercially protected, operationally unchanged
The fee model is less directly tied to time, but delivery has not produced meaningful new capacity.
Maybank framework
A firm with high delivery efficiency and time-linked fees has done the harder operational work and given the benefit away. A firm with outcome-oriented fees and unchanged delivery has commercial protection it is not yet using. Neither position is stable, and neither is a failure — they are different halves of the same problem.
This issue is not theoretical in every sector.
ABA Formal Opinion 512 addresses generative AI in legal practice and discusses, among other obligations, the reasonableness of fees. For hourly work, lawyers generally bill for actual time spent rather than time that would have been required without the technology. The opinion also discusses how AI-assisted efficiency can affect the reasonableness analysis for other fee structures.
That guidance applies to lawyers operating under the relevant professional rules — not to professional services generally. But it demonstrates the larger point:
The commercial consequences of AI-assisted efficiency are already entering professional standards, not merely technology discussions.
The 18–24 month estimate rests on four conditions that we believe are temporary.
The conditions the 18–24 month estimate rests on. Conceptual, with no statistics attached.
Protection 1
Pricing changes in cycles
Many client relationships reprice through renewal, annual rate review or new scope rather than continuously.
Protection 2
Buyers have incomplete visibility
Most buyers do not yet have a standardized method for measuring how AI changes the cost of professional work.
Protection 3
Competitive repricing is uneven
Different firms and sectors are adopting, pricing and disclosing AI-assisted delivery at different speeds.
Protection 4
Quality and risk still matter
Clients may value efficiency, but they also remain concerned about accuracy, confidentiality, professional judgment and accountability.
These conditions will not disappear simultaneously. They can also accelerate one another.
Maybank framework
Each protection fades rather than expiring on a date. No band endpoint is empirically measured.
Conceptual Maybank model, not a forecast. Any of these conditions could change earlier, and the shaded region is a judgment about when several are likely to have weakened together.
Bar opacity represents remaining protection, not a measured quantity.
Maybank estimate
The answer should be calculated from the firm’s own economics.
A useful starting formula for capturable delivery value is:
Annual fee base × delivery-cost ratio × efficiency improvement × period before repricing
Then subtract implementation cost, new technology cost, retained capacity, reinvestment and any price shared with clients.
Assume an annual fee base of $10M, delivery cost at 50% of revenue — $5M — and a delivery-efficiency improvement of 20%. The annual economic value of the released delivery capacity is $5M × 20% = $1.0M. If fee levels held for 18 months, that is $1.5M.
That is not automatically $1.5M of EBITDA. It is $1.5M of theoretical delivery capacity or cost opportunity, before implementation costs and before management decides whether the value becomes margin, volume, price or reinvestment.
$10M annual fee base · 50% delivery-cost ratio · fees assumed unchanged during the modeled period.
| Delivery efficiency | 12 months | 18 months | 24 months | 30 months |
|---|---|---|---|---|
| 10% | $0.50M | $0.75M | $1.00M | $1.25M |
| 20% | $1.00M | $1.50M | $2.00M | $2.50M |
| 30% | $1.50M | $2.25M | $3.00M | $3.75M |
| 40% | $2.00M | $3.00M | $4.00M | $5.00M |
This is theoretical delivery-capacity and cost opportunity before implementation costs and before management decides whether the value becomes margin, volume, price or reinvestment. It is not profit, EBITDA, a forecast or an expected return.
Run the Engagement Margin Model →
Prefer the argument narrated?
The executive briefing connects the market research to the Pyxl operating case and the practical question to ask inside your own firm.
Watch the 18-minute briefingReplace every assumption with your own economics before relying on it.
Maybank illustrative model
Maybank first tested the operating logic inside Pyxl, a professional services company with real clients, delivery teams and a real P&L.
One affiliated company · not a forecast for your firm
August 2025 → Q2 2026 · approximately nine months
+1,300 bps
Gross-margin expansion
Flows through EBITDA
~500 bps
Additional operating leverage below gross profit
+1,800 bps
EBITDA-margin expansion
Pyxl Intelligence was deployed as part of a broader change to how work was produced, coordinated and managed. The 13-point gross-margin expansion reflects stronger delivery economics and therefore flows through to EBITDA. EBITDA margin expanded by 18 points in total, indicating approximately five additional points of operating leverage below gross profit.
Internal Pyxl operating result from one company. It is not a forecast, benchmark or promise of results for another firm. Detailed causal attribution remains subject to Controller confirmation.
Internal operating evidence
Margin for its own sake is not a strategy.
Recovered economics can fund something that is less exposed to time compression.
Five destinations, each less exposed to time compression than billable hours.
Captured delivery value
Productized services
Defined scope, repeatable delivery and a fee tied more closely to the outcome than the hours.
Recurring revenue
Monitoring, managed services, governance, assurance or other continuing value.
Owned intellectual property
Methods, data, benchmarks, knowledge systems and proprietary tooling.
Capability investment
Talent, technology, acquisitions or new service lines.
Strategic price
Improving client value or competitiveness without automatically giving all of it away.
Decide what the recovered economics will fund before ordinary operations absorb them.
Maybank framework
Thomson Reuters’ 2025 research shows a useful tension.
Four independent questions from one survey. Not stages of a funnel.
80%
Believe AI will have a high or transformational impact on their profession within five years
53%
Believe their organization is already experiencing at least one benefit from AI adoption
38%
Expect high or transformational change in their own organization this year
30%
Believe their organization is moving too slowly in AI adoption
Each bar is measured against 100% of respondents independently.
Source: Thomson Reuters, Future of Professionals Report 2025, n=2,275.
Published research
The pattern is clear: the profession broadly believes the change is material. The harder question is whether firms are converting that belief into an operating and commercial system quickly enough to capture the economics.
The failure is rarely that leadership did nothing. It is that activity occurred in the wrong order.
A common sequence in which every decision is reasonable and the order is wrong.
Months 0–3
Evaluate tools
Months 3–6
Run enthusiastic pilot
Months 6–9
Discover inconsistent delivery method
Months 9–12
Discover contract / confidentiality issue
Months 12–15
Resolve governance and ownership
Months 15–18
Begin real deployment
A better sequence starts with one service line and resolves the operating, governance and commercial questions together.
Every individual decision can be reasonable. The sequence can still consume most of the commercial advantage.
Maybank illustrative failure sequence
Not the next two years. The next quarter.
Ten actions in four groups. Not the next two years — the next quarter.
Size
Economics and work
Size the economics
Use actual fee, labor, utilization and margin information from one or two recurring service lines.
Map the work
Identify where time and coordination are consumed.
Decide
Gain, contracts and owner
Decide where the gain lands
Margin, capacity, volume, price or reinvestment. Do not let the current fee model answer by default.
Read the executed agreements
Identify what the work and information permit before choosing the architecture.
Pick one service line
Start where the work repeats and the economics matter.
Name an owner
One person with authority over the operating decision.
Build
Service line, standard and governance
Build the standard
Document the method before scaling the AI layer.
Put review and governance inside the work
Do not bolt them on later.
Measure
Baseline and allocation
Measure the before state
Delivery time, review, rework, cost, capacity and relevant commercial metrics.
Write down what the value will fund
Captured efficiency without an allocation decision tends to disappear into ordinary operations.
Maybank framework
A firm does not need to believe Maybank’s exact 18–24 month estimate to accept the underlying exposure.
If AI materially reduces the cost or time required to produce professional work, firms eventually have to answer three questions:
How will the work change?
How will the commercial model change?
What will the firm build with the economics it captures in between?
Waiting is also an answer. It leaves the existing delivery and fee models to make those decisions by default.
The 18–24 month window is a Maybank working estimate. The four temporary protections, the sensitivity matrix, the efficiency and fee-model matrix, the deliberation sequence and the capital-allocation framework are Maybank analytical models, not external datasets.
Pyxl figures are internal operating results from one company, confirmed against its own accounts. They are not a forecast, benchmark or promise of results for another firm.
Find out where your firm stands
The readiness assessment looks across delivery, data, tooling and governance. The Engagement Margin Model lets you replace our illustrative assumptions with your own fee and cost structure.
Continue the series →
02 · The delivery system
The Prompt Library Playbook
How recurring professional work becomes a governed AI operating asset.
03 · The control system
Governance for Professional Practices
Using AI while keeping the work reviewable, accountable and defensible.
Maybank Advisors · Nashville · Charleston