It is the second morning of a planning event in a rented conference hall. Eleven teams sit at eleven tables, each with a whiteboard of sticky notes and a dependency board on the far wall covered in red string. A release train engineer walks between tables with a timer.
The CFO, who approved the budget for this, does a quick sum on a napkin: 130 people, 2 days, 4 or 5 times a year. That is roughly 1,200 working days spent in this room annually, before anyone counts the consultants.
Nobody in the room is doing anything wrong. This is SAFe working as designed, and many organizations find the event worth every hour. The point of the napkin is different: scaling frameworks are rarely compared by what they cost to run.
Debates about SAFe, LeSS and Scrum@Scale tend to stay at the level of philosophy, centralized planning against descaling against fractal Scrum. The comparison below adds the bill: the roles each framework creates or removes, the recurring hours its events consume, and the one cost most budgets miss, which is the reorganization each one demands before it pays back.
Three frameworks, three theories of what breaks at scale
Each framework starts from a different diagnosis of why ten Scrum teams working on one product struggle, and the costs follow from the diagnosis.
SAFe, created by Dean Leffingwell and maintained by Scaled Agile, Inc. since its first release in 2011, assumes the problem is alignment across many teams and many layers of management. Its answer is a structure: teams of teams called Agile Release Trains, usually 50 to 125 people, that plan together at a fixed cadence, the Planning Interval of typically 8 to 12 weeks.
Around the trains it adds roles such as Release Train Engineer, Product Management, System Architect and Business Owners, and in its larger configurations a Lean Portfolio Management layer. SAFe 6.0, released in 2023, is the current major version. Industry surveys such as Digital.ai’s State of Agile reports have for years listed it as the most adopted scaling framework, largely because it maps onto existing corporate hierarchies.
LeSS, developed by Craig Larman and Bas Vodde and described in their books from 2008 onward, assumes the problem is the organization itself: too many roles, handoffs and component teams. Its answer is to remove things. One Product Owner, one Product Backlog, one Sprint and one Definition of Done for the whole product, with cross functional feature teams that can each pick up any backlog item. Basic LeSS covers 2 to 8 teams; LeSS Huge adds requirement areas with Area Product Owners for larger groups.
Scrum@Scale, created by Jeff Sutherland, co-creator of Scrum, assumes the problem is that Scrum’s structure is not repeated at higher levels. It scales Scrum fractally: a Scrum of Scrums of around five teams, a Scrum of Scrums Master, a Chief Product Owner, and at the top an Executive Action Team and Executive MetaScrum where leadership removes impediments and sets priorities. It prescribes little beyond that, describing itself as a minimum viable bureaucracy.
The diagnoses predict the bills. SAFe spends on coordination structure. LeSS spends on organizational change. Scrum@Scale spends on leadership time.
The visible cost: roles, events and training
The easiest costs to count are the new roles and the recurring events. Take an illustrative product organization of 120 people in 12 teams of 8 to 10, and look at a single year.
SAFe. One train is too large at 120 people for comfort, so assume one train with 12 teams at the upper bound. New full time roles typically include a Release Train Engineer, 2 product managers above the team Product Owners, a System Architect, and often a part time Lean Portfolio Management function. That is 4 to 5 dedicated people.
Planning events run 2 days for everyone, 4 or 5 times a year: about 1,080 to 1,200 person days. Add system demos every 2 weeks, an Inspect and Adapt workshop each interval of around half a day, and ART Sync meetings. Training is usually heavy: a SAFe Program Consultant or external coach to launch the train, and role based courses for dozens of people.
LeSS. LeSS adds almost no roles. With 12 teams the group sits in LeSS Huge territory, so expect 1 Product Owner and 2 Area Product Owners, plus Scrum Masters at roughly 1 per 1 to 3 teams. Events are Sprint Planning One with representatives of all teams, an Overall Retrospective and a joint Sprint Review each sprint.
Over 26 two week sprints, multi team events might add 3 to 4 hours per person per sprint above single team Scrum: roughly 1,000 to 1,300 person hours a year for 120 people, or about 130 to 170 person days.
Scrum@Scale. With 12 teams, expect 2 or 3 Scrums of Scrums, a Scrum of Scrums of Scrums, and an Executive Action Team and MetaScrum. New roles are mostly hats worn by existing people. The Scaled Daily Scrum takes 15 minutes a day for one representative per team: about 780 person hours a year across 12 representatives, under 100 person days.
The real line item is executive time: if 5 executives spend 3 hours a week in the Executive Action Team and MetaScrum, that is close to 800 hours of the most expensive time in the company.
Read the numbers as orders of magnitude, not quotes. They show that SAFe’s recurring overhead is several times higher than the other two, and that Scrum@Scale’s overhead concentrates where it hurts most to waste it.
The hidden cost: reorganization before payback
Recurring hours are visible on a calendar. Reorganization is not, and it is usually the largest single cost of any scaling framework.
SAFe has the lowest reorganization cost of the three on paper. It accepts component teams, existing managers and a separate architecture function, and asks you to group teams into trains around value streams. That flexibility is why it lands well in banks, insurers and government suppliers.
The catch is that SAFe lets the structural problems stay. If 12 teams are split by layer, front end, back end, data and QA, the planning event will reveal the dependencies beautifully and solve none of them. The cost reappears as permanent coordination overhead.
LeSS has the highest reorganization cost. Converting component teams into feature teams means people learn unfamiliar parts of the codebase, managers lose direct ownership of components, and the first 2 or 3 sprints after the change usually deliver less.
Larman and Vodde are explicit that LeSS is organizational design, not a process overlay, and that adopting it partially gives few of its benefits. For a 120 person group, a realistic transition takes 6 to 12 months, with a visible productivity dip in the first quarter. The payoff, if it comes, is a sharp drop in handoffs and coordination roles.
Scrum@Scale sits in between. It does not mandate feature teams, but its Product Owner cycle exposes priority conflicts quickly and pushes executives to resolve them in the MetaScrum. The reorganization it forces is at the top: leaders must meet on a cadence and make backlog decisions in public. In many companies that is harder than moving engineers.
A useful rule of thumb is to estimate the dip honestly. If a reorganization costs 15 percent of delivery capacity for 3 months, in a 120 person group that is roughly 1,100 person days, about the same as a year of SAFe planning events. The question is which cost you want: a one time hit that removes coordination, or a permanent overhead that manages it.
What each framework does to product ownership
The three frameworks disagree most sharply about who decides what gets built, and this has a cost beyond meeting hours: decision latency.
In SAFe, ownership is layered. Epic Owners and Lean Portfolio Management shape large initiatives, Product Management owns the train’s backlog of features, and team level Product Owners own stories. Each layer adds a handoff. The benefit is that large organizations with many stakeholders get a clear escalation path. The cost is that a team level Product Owner often has little real authority over what the team builds, and the role drifts toward backlog administration.
LeSS goes the other way. One Product Owner owns one backlog for up to 8 teams, and in LeSS Huge each Area Product Owner owns a requirement area under that single owner. The Product Owner cannot possibly write all the details, so teams talk directly to customers and stakeholders to clarify items.
That only works if teams are trusted to do discovery and if the Product Owner is a real product leader, often a senior business person rather than a backlog clerk. When the role is filled by a junior proxy, LeSS stalls fast.
Scrum@Scale keeps team Product Owners and stacks them: team POs form a Product Owner team under a Chief Product Owner, and Chief Product Owners meet in the Executive MetaScrum. The structure is lighter than SAFe’s, but it requires the executive layer to own one ordered backlog for the organization. Many leadership teams resist exactly that, because ordering a single backlog forces explicit trade-offs between departments.
In interviews for senior PM and engineering leadership roles, this is often where a scaling question really goes. Saying which framework you prefer matters less than explaining where decisions sit and how long they take to reach a team.
An illustrative case: nine teams, two frameworks, one year
The following case is a composite, built from patterns common in mid-sized software companies, not a description of a specific firm.
A payments company has 9 teams and about 85 engineers, product managers and designers working on one merchant platform. Teams are split by component: checkout, risk, ledger, reporting, mobile and so on. Releases slip because almost every feature touches 3 or 4 teams. The COO proposes SAFe, which a sister company uses. The VP of Engineering argues for LeSS.
They compromise on a 2 quarter pilot. Five teams adopt LeSS: they merge into 4 feature teams under one Product Owner, the former checkout PM, with a shared backlog. Four teams keep their structure and run SAFe style planning events every 10 weeks with a part time release train engineer.
After the first quarter, the LeSS group looks worse. Throughput, measured as backlog items finished per sprint, drops about 20 percent while engineers learn unfamiliar code. The SAFe group looks better: the planning event surfaces dependencies early, and predictability, measured as committed objectives delivered, rises.
By the end of the second quarter the picture changes. The LeSS group’s throughput is back to baseline and still rising, and the number of cross team dependencies on its board falls from 30 to 7, because most features now fit inside one team.
The SAFe group’s predictability holds, but its dependency board still has about the same number of items; they are just better managed. The company extends LeSS to the remaining teams and keeps one practice from SAFe: a quarterly half day joint planning session with business stakeholders.
The lesson is not that LeSS wins. It is that the comparison must run long enough for a reorganization to pay back. A 3 month evaluation would have picked SAFe for the wrong reason.
Mistakes that inflate the cost of any scaling framework
Most cost overruns come from how a framework is adopted, not from the framework. The recurring ones:
- Scaling before one team works. If a single team cannot ship a tested increment every sprint, no framework will fix it at 12 teams. It will multiply the problem.
- Buying SAFe’s full configuration for one train. Portfolio and Large Solution layers are designed for many trains. Adding them to 100 people creates roles with no work to coordinate.
- Running LeSS with component teams. Keeping teams split by layer while adopting one backlog gives you LeSS meetings and none of its benefits.
- A proxy Product Owner. In LeSS and Scrum@Scale, a Product Owner without authority over priorities turns the single backlog into a negotiation forum.
- Executives skipping the MetaScrum. Scrum@Scale without leadership participation is just a Scrum of Scrums, which most organizations already had.
- Measuring adoption instead of outcomes. Counting certified people, trains launched or events held says nothing about lead time, quality or customer value.
- Ending the pilot too early. Any structural change looks worse in its first quarter. Judge it after at least 2.
- Outsourcing the whole adoption to consultants. External coaches help launch a train or a first LeSS sprint, but if they still run the events after 6 months, the organization has rented a process it does not own.
- Ignoring technical practices. Twelve teams integrating into one product need continuous integration, automated tests and trunk based development. Without them, every framework spends its event time discussing merge conflicts and broken builds.
A decision tree, and a verdict on the trade-off
Work through these questions in order and stop at the first clear answer.
- Are fewer than 3 teams working on the product? If yes, use plain Scrum or Kanban with a shared review; no scaling framework pays for itself.
- Is the product genuinely one product, with one set of customers and one backlog that could be ordered? If no, scale each product separately before adopting anything across them.
- Is leadership willing to change team structure and reporting lines? If yes, consider LeSS; it has the highest upfront cost and the lowest running cost.
- Are you in a regulated or contract heavy environment with fixed governance, many vendors and annual budgeting that will not change? If yes, SAFe’s structure will be easier to fit, and its overhead buys compatibility.
- Will executives commit to a weekly cadence for prioritization and impediment removal? If yes, Scrum@Scale gives you the lightest structure with a working top layer.
- If none of the above is a clear yes, start with a Scrum of Scrums and a joint quarterly planning session, measure dependencies and lead time for 2 quarters, and decide then.
The verdict comes down to where you prefer to pay. SAFe charges a high, predictable, permanent fee in roles and event hours, and in exchange it fits almost any organization as it is. LeSS charges a large one time fee in reorganization and a real risk of a failed transition, and in exchange its running cost is the lowest of the three. Scrum@Scale charges in executive attention, which is cheap in hours and expensive in willpower.
For a single product with up to 8 teams and leaders willing to restructure, LeSS usually costs least over 2 years. For many products in a governed enterprise, SAFe is often the only option that gets approved. And for a company whose real problem is unmade decisions at the top, Scrum@Scale is the honest choice, provided the people at the top show up.