PSA Software vs Agents: Kantata, Certinia and the Shrinking Billable Hour

Professional services automation software was built to sell more hours, and its own customers are now buying agents designed to need fewer of them. Billable utilization across 509 services firms fell to 66.4% in 2025, the lowest SPI Research has ever recorded. PSA software vendors Kantata and Certinia spent the same period shipping agents that cut the hours their products exist to count.

The short answer

PSA software does not lose its job when agents cut billable hours, but it loses its pricing anchor. Under fixed-fee work the hour turns from revenue into cost, so the PSA tool must track cost to deliver, human and agent combined, against a price agreed upfront. Vendors that make that ledger the product survive. Vendors selling timesheets do not.

What professional services automation software actually measures

Professional services automation, usually shortened to PSA, is the operating system of a consultancy, agency or implementation partner. It holds the project plan, the resource schedule, the timesheet, the rate card and the invoice. Kantata, formed from Mavenlink and Kimble, and Certinia, formerly FinancialForce, are the two names most buyers shortlist.

Strip the modules away and one number sits underneath all of them. Billable utilization is the share of a consultant's available hours that end up on a client invoice. Resource managers chase it, finance forecasts from it, and every executive dashboard in a PSA product leads with it.

That number made sense for decades because hours were the product. A firm on time-and-materials contracts earns revenue only when a human logs time against a billable code. The timesheet is therefore not admin. It is the revenue recognition event, and the PSA system is where it happens.

The seat model followed the same logic. PSA licences are bought for the people who log time and the people who manage them. More consultants meant more seats, and more seats meant more vendor revenue. The vendor and its customer both grew by adding billable heads.

Agents break both halves of that bargain at once. They do work that used to be logged as hours, and they do it without a seat. The broader version of that squeeze is covered in the analysis of seat compression across SaaS pricing. PSA is a sharper case, because here the seat and the billed hour are the same person.

Billable utilization hit a record low before agents arrived

The SPI Research Professional Services Maturity Benchmark is the closest thing this industry has to official statistics. Its 2026 edition, published in February, covers 509 organisations employing over 245,000 consultants with nearly $63 billion in services revenue.

Its headline finding is uncomfortable for every PSA vendor. Billable utilization averaged 66.4% in 2025, the lowest point in SPI's survey history. SPI treats 75% as the healthy threshold, and the top-maturity firms in the sample exceeded 80%.

The decline is not new. A Certinia partner's summary of the report describes four straight years of falling utilization. That timing matters, and the limitations section below returns to it.

Here is the detail that makes the benchmark interesting rather than merely gloomy. The same report put average project margin at 37.7%, a five-year high. Utilization fell to a record low in the same year project margin reached a five-year high.

One year of data proves nothing on its own. It is still the first public sign that what a firm bills and what it earns on a project are starting to move independently. If that continues, a product built to maximise the first number is measuring the wrong thing. SPI also reported that 27.1% of projects used generative AI in 2025, up 40% on the prior year.

PSA AI features: what Kantata and Certinia have shipped

Both vendors have moved fast, and both have aimed their first agents at the same place. That place is the resource schedule, the part of the product closest to the utilization number.

Certinia: from one assistant to Veda's agent catalogue

Certinia's Winter '26 release in December 2025 added a Project Assistant Agent that flags missing time cards, overdue items and milestones near critical status. Those features shipped first to customers in an early adopter programme.

In April 2026 it launched Veda, which it calls "an enterprise-grade intelligent operations engine". The launch, reported by CPA Practice Advisor on 15 April 2026, included a Reallocation Agent that finds replacements when someone leaves a project. Certinia claimed up to 10 hours a month returned to resource managers and up to 20 hours to project managers. Those are vendor figures with no published method.

By September the catalogue had grown sharply. Veda went from 10 agents at launch to 24, and from 64 intelligent actions to 135. The new agents reach into selling, with an RFP Responder and an SOW Drafter, and into delivery, with a Solution Architect agent. Certinia's stated goal is more than 40 agents and over 300 actions by the end of 2026.

Kantata: from Resourcing Agent to Agent Studio

Kantata's sequence is similar in shape and different in philosophy. Its Resourcing Agent shipped on 24 September 2025 to all Kantata OX customers, detecting overallocations, misaligned timelines and budget overruns. Two weeks later it announced the Kantata Expertise Engine, which was in beta with select customers at launch.

In June 2026 Kantata launched what it calls the industry's first AI superagent, explicitly arguing against selling single-purpose agents "by the dozen". On 22 September 2026 it followed with Agent Studio, where a user describes the job they want an agent to handle in plain language. A Basware executive quoted in that release wants consultants spending time with customers rather than "entering their billable time".

That quote is the whole tension in one line. A PSA vendor's flagship customer story now celebrates less time spent in the timesheet module.

Compiled from each vendor's press releases and pricing pages, read 13 October 2026. Both vendors were named Leaders in IDC MarketScape assessments of AI-enabled PSA applications, which is why this table does not try to rank them.

Kantata vs Certinia pricing: neither publishes a number

Open Kantata's pricing page and you find a form titled "Request Pricing". The copy says plans are tailored "based on your company size, goals, and the way your team works". Certinia's pricing URL resolves to a demo request page. Neither lists a tier, a per-user rate or an AI add-on price.

Third-party sites quote per-user figures for both. None of the ones checked for this post traced to a vendor document, so none is repeated here. If you have a signed order form, you know more about PSA software pricing than any comparison site does.

The one disclosed pricing structure is Certinia's description of Veda. It says a single Veda subscription gives access to the whole AI suite through "a transparent consumption model that pairs per-user fees with usage-based scaling". That is a hedge, and an honest one. The per-user fee protects today's revenue. The usage component is the bet on a future where agents, not consultants, do the counted work.

My view is that the hedge is correct and the split is the thing to negotiate. A buyer whose headcount is flat or falling should push weight toward the usage side, where cost tracks the work actually done. A buyer growing headcount should do the opposite. The same logic applies to any vendor whose value metric has to survive automation.

The exposure is easy to state. If a firm delivers the same revenue with fewer billable consultants, its PSA seat count falls even as its use of the product rises. Certinia's chief product officer, Raju Malhotra, put the new problem plainly to diginomica in September 2026: "It is also about how we take into account the AI cost, the token cost, the credit cost". A PSA system that cannot see those costs cannot price a project.

What public consulting firms disclose about fixed-fee work

Vendor messaging tells you what PSA companies hope. Annual reports from the firms that buy services software tell you what is happening to the billing model. Three filings are worth reading closely.

EPAM: time-and-materials is shrinking as a share

EPAM's 10-K for fiscal 2025 states plainly: "We generate the majority of our revenues under time-and-materials contracts". The disaggregation note shows how that majority is changing. Time-and-materials revenue was 87.0% of revenue in 2023, 82.5% in 2024 and 80.0% in 2025. Fixed-price revenue grew about 33% in 2025, against about 12% for time-and-materials.

The risk factors spell out why. EPAM writes that "If comparable services can be performed less expensively using AI, clients may seek other service providers or expect price concessions". It also says its industry "is adopting, and our clients are expecting, new pricing models, especially as AI tools evolve".

Cognizant: fixed price is already the largest bucket

Cognizant crossed over earlier. Its fiscal 2025 10-K shows fixed-price contracts at $10.0 billion of $21.1 billion in revenue, about 47.4%, up from 44.8% in 2023. Time-and-materials fell from about 45.9% to 43.3% over the same period.

Its outlook section is blunt: "As AI-based technologies or other forms of automation evolve, demand for some services that we currently perform for our clients may be reduced". A firm writing that sentence into a filing is not planning to grow by selling more hours.

Accenture: the largest firm does not report the hour at all

Accenture's fiscal 2025 10-K, covering about 779,000 employees, contains no revenue split by contract type. Instead it states: "Management does not track standard measures of unit or rate volume." It defines pricing as "the contract profitability or margin on the work that we sell."

That is the most important sentence in this whole section. The largest consulting firm in the world already manages itself on margin per contract, not on hours sold. Its risk factors add that results suffer "if our clients do not accept new pricing or commercial models that reflect the value of these AI-enabled solutions".

Billable hours and AI: who keeps the hour an agent saves

The contract type decides who captures an agent's productivity. This is the mechanism under every number in the previous section, and it is simple enough to put in a table.

Under time-and-materials, an agent that saves a consultant 10 hours removes 10 hours from the invoice. The firm has paid for the agent and earned less. No services firm will run that trade for long, which is why the filings above show the mix moving.

Under a fixed fee, the same saving becomes margin. One example from SiliconANGLE's coverage of Certinia at Dreamforce in September 2026 shows the sequence. Platinum River Innovations rewrote its client contracts from time-and-materials to fixed-bid before it introduced AI tools, so it absorbs delivery risk and keeps the efficiency. SiliconANGLE's parent network was a paid media partner for that event, which is worth knowing when weighing it.

Notice the order of operations. The contract changed first and the tooling second. That is the right order, and it is the one I would recommend to any firm reading this. Agents bought into a time-and-materials book cut your own revenue.

For PSA software, the fixed-fee world changes what the timesheet is for. Hours still matter, because they are the main cost input against a fixed price. But they become a cost-accounting record rather than a billing event. The commercial mechanics of the outcome row are covered in the breakdown of outcome pricing contract clauses, so this post will not repeat them.

Can PSA become the system of record for outcomes?

This is the real question for the category, and the honest answer today is that the data does not exist yet. The Resource Management Institute study commissioned by Kantata, released in April 2026, found that only 7.3% of respondents had outcome data easily accessible and used in staffing. Half were unsure whether their firm would meaningfully adopt outcome-based pricing at all.

The sample is 44 organisations and the sponsor sells PSA software, so treat these numbers as directional. Still, the gap they describe matches everything else in this post. Firms are moving to fixed prices without a ledger of what each delivery actually achieved.

That gap is the opening. A PSA product that records, per engagement, the agreed price, the human hours, the agent runs and their token cost, and a verified result, owns something no general ledger or CRM holds. That record is what lets a firm price the next fixed bid with confidence. It is also what a services firm needs to sell services as software on outcomes without guessing.

What has to change in the data model

Three changes matter more than any single agent. First, agents need to be first-class resources in the schedule, with a cost rate, the way contractors already are. Second, cost to deliver has to be a live number on every fixed-fee project, not a month-end reconciliation. Third, the outcome has to be a field someone signs off, not a status colour.

Certinia's September release points in this direction, with ledger actions exposed through the Model Context Protocol and agents callable from Slack, Claude and Microsoft Copilot. Kantata's Agent Studio points toward firms encoding their own delivery methods. Neither, from the public material, yet treats an agent's cost as a line on the project P&L the way a consultant's cost is. Pricing the agent side of that ledger is its own problem, covered in the enterprise agent cost model.

Where this argument is weakest

The thesis here is that agents push services firms off the hour and PSA vendors must follow. Three pieces of evidence cut against it, and each deserves a fair hearing.

Utilization was falling before agents could explain it

Four straight years of decline means the slide started around 2021, well before agentic tools were in production at services firms. Softer demand, post-pandemic overhiring and longer sales cycles explain a falling utilization number at least as well as AI does. Treating 66.4% as an agent effect would be a mistake, and this post does not.

The SPI figures here also come through Certinia, Deltek and partner summaries of a paid report. Two of those sources sell PSA software. The 509-firm sample is published by SPI itself, but the detail sits behind a paywall.

Cheaper projects can mean more projects

Platinum River's chief executive, Mike Piehl, made the counter-case in the same SiliconANGLE piece. Labour per project is falling, he said, but price points are falling too and demand is rising. "So, more projects coming in, and the net effect is we need more people." If that pattern generalises, PSA seat counts hold up even as hours per project shrink.

I think this is the strongest objection, and I do not think it is settled. It is one firm, quoted at a vendor event. It is also a coherent economic argument with long precedent: when the cost of a service drops, buyers often consume more of it.

The contract shift is slow and partly acquired

EPAM's 2025 revenue grew 15.4%, and acquisitions closed in late 2024 contributed 9.2 points of that. Some of the fixed-price growth may be inherited rather than chosen. Cognizant's shift is under 3 points across two years. Accenture does not disclose a split at all. Nothing here shows a cliff, and anyone selling the death of the billable hour by 2027 is ahead of the evidence.

Frequently asked questions

Where to start

Pull last year's revenue and split it by contract type, the way EPAM and Cognizant do in their filings. If more than half is still time-and-materials, decide which engagements move to fixed fees before you buy another agent. Agents bought into an hourly book shrink your own invoices.

Then, before your next PSA renewal, ask your vendor one question in writing: how will the contract price agent usage, and does the per-user fee fall if your billable headcount does? The answer tells you whether the vendor is building an outcomes ledger or defending a timesheet. For the wider platform question, see whether Salesforce is a platform or roadkill in the agent era, which matters directly if you run Certinia.

Related reading