How to evaluate a CWM platform

A packaging manufacturer with 380 people opened a platform evaluation in August. It received four proposals, watched four flawless demos and ended up with a spreadsheet of 214 features marked yes, no and partial. In the decision meeting, nobody could say which of those 214 items would change the month-end close, which was the reason the evaluation started in the first place.

A feature list compares products. It doesn't compare outcomes. What follows is a script that keeps the conversation on your company's processes, organized around the seven capabilities that define the category — plus the questions that rarely show up in an international comparison.

01The seven capabilities of the category #

The category got a formal market name when Gartner published the Magic Quadrant for Collaborative Work Management on October 28, 2025. Here we use only the category's public capability structure as an editorial script. Proper attribution: GARTNER is a registered trademark of Gartner, Inc. and/or its affiliates; Gartner does not endorse any vendor described in its publications, and it has neither evaluated nor positioned Relevanti — we do not appear in that quadrant. What's useful from an analyst is the vocabulary; the decision is yours.

1. Work planning — tasks, projects, recurring processes and continuous operations in one model, with dependencies, capacity and calendar.
2. Collaboration in context — the conversation attached to the stage, not to a parallel channel; mentions, assignment and history inside the record itself.
3. Content collaboration — documents, spreadsheets and attachments belonging to the work item, with inherited permissions and traceable versions.
4. Automation and workflow — conditional rules, approval thresholds, relative due dates, cross-team triggers and integration with the systems you already run.
5. Reporting and dashboards — metrics computed from the work data itself, not from a status spreadsheet filled in by hand.
6. Intelligent assistance — AI that summarizes, classifies, drafts and, where authorized, executes inside the process, with an audit trail.
7. Use-case accelerators — ready templates for real processes (purchasing, hiring, service desk, close) that shorten implementation.

02The checklist: what to ask to see, capability by capability #

The checklist has one rule: nothing is evaluated by description. Every item becomes a request to demonstrate on your process, with your data, done by your team — not by the pre-sales engineer.

If a vendor needs two weeks to show any of those items, you already learned something: that's a configuration project, not a product capability. Not necessarily bad — but it changes the cost and the timeline of the deal.

03Five criteria buyers in Brazil have to ask about #

International comparisons treat these as administrative detail. In a mid-size Brazilian company they decide whether the platform gets used or worked around.

1. Native language of the product and of the AI. A translated interface isn't enough. Ask which language the assistant was evaluated in, whether it understands 'nota fiscal', 'boleto' and 'quinto dia útil', and whether the Portuguese answer holds the same quality as the English one.
2. Contract currency. Pricing in dollars hands you the entire currency risk. Ask whether the contract is in Brazilian reais, what index governs annual increases, and what happens when you add seats mid-cycle.
3. Data residency and LGPD. Where the data lives, who is processor and who is controller under the LGPD, Brazil's data protection law (Law No. 13,709/2018), how deletion on a data subject's request works, and the breach notification window. Ask for the current security certification too, ISO/IEC 27001:2022 or equivalent. If there's AI in the flow, ask whether your data trains a third party's model — and the NIST AI Risk Management Framework 1.0 (2023) is a good script for that conversation. None of this is legal advice: it's what your legal team will ask anyway.
4. Who does the implementation. The vendor, a partner or you? Ask for the names of the people, the estimated hours from your own team, and exactly what the price includes. Implementation sold as 'free onboarding' usually means two recorded webinars.
5. Support time zone. Support that answers at 3 a.m. Brasília time isn't support. Ask for coverage hours in local time, the channel, the contractual first-response time, and who answers in Portuguese.

Add it all up as three-year total cost, not price per seat. In a typical evaluation at a mid-size Brazilian company (figures in Brazilian reais, no conversion), licenses of R$ 12 mil per month, implementation of R$ 60 mil and one half-time person in the first year add up to a very different number from the one on page one of the proposal.

04The pilot is worth more than the demo #

No demo reveals what a four-week pilot reveals. Pick a process that hurts: the one that jams every month, has a clear owner and involves at least two teams. Purchasing with two approvals and employee onboarding are the best candidates because they cross boundaries — and the boundary is where a platform shows what it's made of.

• Measure first: current cycle time, number of back-and-forths, how many people touch each case.
• Run four weeks with real data and real users — a pilot on fake data only tests the screen.
• Measure after, on the same basis, and record what got worse too. Something always gets worse.
• Ask the people who operated it, not the people who bought it. The sentence that matters is 'would I go back to the spreadsheet?'.

Want to see how this looks inside a real operation? Explore the platform.

And the warning from automating a bad process only speeds up the error applies: if the chosen process is badly designed, the pilot will surface the problem faster and with more people watching. That's a good outcome, even if it doesn't feel like one that day.

05What to ask Relevanti #

It would be odd to publish an evaluation guide and then slip away when it's time to apply it at home. So here are the questions you should put to us — and that we answer in writing, in the proposal.

• How many companies of my size and in my sector are live, and for how long? Ask to speak with one of them.
• What was the hardest process to implement so far, and what went wrong in it?
• What do agents execute on their own and what requires human approval, by default? Who defines scope and permissions, and where does the audit trail live?
• What's the exit path: in what format do I take my data with me if I switch vendors, and how long does it take?
• What doesn't the platform do yet that's on the roadmap — with a date, not a vague quarter?

06When Relevanti is not the best choice #

There are situations where the honest answer is 'we're not for you right now'. Recognizing them early saves everybody months.

• Your company is almost entirely unique client projects. Agency, consultancy, studio: a good project tool does the job, and the difference is spelled out in CWM is not project management.
• You need a deep vertical system. Electronic health records, retail point of sale, shop-floor MES, core banking. We're the work layer around those systems, not a replacement for them.
• Nobody can own a process. Without someone mandated to decide how work runs, the platform becomes one more place where the mess lives — now with single sign-on.
• The company has fewer than 30 people. Below that, the cost of coordinating is still low and simple tools usually suffice; the argument is in the cost of coordinating.
• You require installation in your own datacenter. We run in the cloud; if your policy forbids that, the conversation ends there, respectfully.

07How to close the decision without a feature spreadsheet #

Reduce it to three weighted criteria, defined before you watch any demo: how much the platform solves the two processes that hurt most, the three-year total cost, and implementation risk — which is people, timeline and vendor dependency. A feature that serves none of the three doesn't count, however good it looks on screen.

If you want to apply this here, start with how the modules fit together and what agents execute inside a process; if you'd rather get to the point, talk to us with a real process in hand. The track's foundational piece is worth reading too: what Collaborative Work Management is.

A good evaluation isn't the one that picks the best product on the market. It's the one that picks the product your company will still be able to operate two years from now, with the people it has — including when the answer is to stay as you are for another cycle.