Most enterprise AI business cases start with one question – how much faster can we do the work we already do? It’s a fair question. And it’s also the wrong place to stop.
A larger return may be sitting in work your organization does not do at all – processes everyone agrees are valuable, but no one runs consistently. Why? Because their workflows cross too many systems, inboxes, and spreadsheets to justify the labor.
Duvo, the Czech AI agent company co-founded in 2025 by Rohlik Group founder Tomáš Čupr, calls this “abandoned work.” Operations teams know exactly what needs to happen. They simply can’t get to all of it.
For IP leaders, this concept reframes how value is created by shifting the question from “What can we automate?” to “What have we quietly stopped doing consistently?” Ultimately, that second question points directly to budget lines most AI programs overlook.
From quarterly at best to continuous
Rohlik Group is Europe’s leading online grocer, with more than €1.5 billion in revenue and operations across the Czech Republic, Germany, Austria, Hungary and Romania. Its commercial team faced a familiar gap. Prices on commodity-linked products move constantly, and supplier terms should move with them. Tracking those shifts and pressing suppliers for adjustments was a process the team understood well – and ran quarterly at best.
A Duvo agent took over the monitoring and began initiating supplier outreach itself. That first process became a template. Rohlik now runs dozens of agent-driven processes, and within three months the company reported more than €2.8 million in annualized savings from work that had never been systematically run. Inbound delivery confirmations rose from 52% to 90%, covering every supplier daily.
None of that value came from doing an existing task faster. It came from running tasks that barely ran before – a category of return that never appears on a time-saved dashboard.
Read the numbers carefully, then change what you measure
This is a vendor-reported result, not an audited study. It appears in a customer case study published by Anthropic, whose Claude models underpin the deployments. It does not publish a full baseline or methodology. You should treat the figures as directional. Regardless, the pattern behind them deserves attention because it exposes a blind spot in how most organizations evaluate AI.
Time-saved metrics only count work that already happens. Work that is skipped, sampled, or deferred has no baseline, so it never enters the business case.
Three changes close the delta.
The first is to stop measuring only the time saved on existing work. Efficiency gains are real, but the size of today’s workload caps them. Another is to map the work your teams skip, sample, or defer indefinitely. Ask which reviews happen annually that should happen weekly, and which checks run on 10% of cases that should run on all of them. Finally, you should measure coverage, recovered value, and earlier intervention.
Seats purchased and prompts submitted really describe adoption. They say nothing about value. Organizations that measure coverage will find the opportunities their competitors’ dashboards cannot see.
What abandoned work looks like in IP
Intellectual property (IP) and R&D functions carry more abandoned work than most. Their processes span the IP management system (IPMS), patent office registers, outside counsel, inventor inboxes, product roadmaps, and licensing agreements, among many other sources. Each handoff is a reason a valuable task gets postponed.
There are some ready examples of where to leverage agentic AI to shore up processes, gain efficiency and drive better oversight.
Continuous portfolio hygiene is the most immediate. Reconciling IPMS records against national registers, assignment records and deadlines typically happens during an annual audit or a due diligence scramble. An agent can run that reconciliation every week and flag discrepancies before they become lost rights or embarrassing data-room findings.
Dormant-invention follow-up is another example. Invention disclosures marked “revisit in 12 months” are rarely revisited. Systematic follow-up with inventors and product teams can surface inventions that have since become commercially relevant – while filing is still an option.
Competitor monitoring follows the same logic. Most landscape reviews happen ahead of a launch, a negotiation or a dispute. Continuous monitoring of competitors’ filings, publications and product announcements turns that intelligence from reactive into standing.
Finally, filing and renewal assumptions deserve regular re-evaluation. Those decisions are often made once, against the business case at the time. US maintenance fees fall due at 3.5, 7.5, and 11.5 years, and European renewals arrive annually in every validated country. Re-testing each decision against current product, licensing and competitor evidence transforms renewals from a routine payment run into active capital allocation.
Additionally, each of these processes positively builds on one another. A portfolio that is cleaned, monitored and re-evaluated continuously is easier to license, defend and value – and cheaper to maintain.
Possible for the first time
The default AI question asks how to make a known process cheaper. The better question asks which valuable processes have never been economical to run at all. Every IP department has that list. Start there. AI’s largest return may not be a cheaper process – it may be making a valuable process possible for the first time.

Written by Toni Nijm
Chief Product Officer, Anaqua
You may also like…
BIP Asia Forum and Entrepreneur Day open today
Advancing innovation and global expansion November 19, 2026 – The 16th Business of IP Asia Forum (BIP Asia Forum),...
UPC confronts alleged abuse of divisional patent applications in Sandoz v. Novo Nordisk
The Unified Patent Court (UPC) has confirmed that it can take measures to prevent patent proprietors from using...
GSK wins major UPC patent ruling against Pfizer and BioNTech over mRNA vaccine technology
The Unified Patent Court has found that Pfizer and BioNTech infringed GSK's patent covering lipid-based RNA delivery...
Contact us to write for out Newsletter












