Soroco and the Case of the Wrong Robot

The consumer-goods company already had an automation factory. It had projects, expertise and a preferred next target. Previous efforts, however, had produced disappointing returns. So before committing again, it brought in Soroco’s Scout to examine the work. The favorite project turned out to have the lowest return potential. An expensive queue had been arranged in the wrong order.

The favorite came last

In Soroco’s account, the unnamed customer deployed Scout on premises across roughly 400 users and eight teams. Seven processes received custom analysis. Two became automation candidates; five were to be standardized and reconsidered. The study identified 39,000 hours of potential savings. That last word matters. A forecast is a reason to investigate, not a deposit in the bank.

The interesting result is the change of mind. Management had a plausible answer before the measurement began. The measurement reordered it. If a discovery tool merely confirms the project everyone already wanted, it may be useful. A tool that forces a room to reconsider its favorite deserves a closer look.

“The process originally identified as the best candidate for automation turned out to have the lowest ROI potential.”Soroco’s consumer-goods case study

The employee is the integration

A system of record can tell you when an order was created. The work graph asks about the work surrounding that event: which applications a team used, how the sequence varied, and where manual effort accumulated. Soroco derives the graph from interactions between people and software. The promise is a map of execution rather than a diagram of intentions.

Consider an illustrative order, rather than a particular customer’s data. An employee opens an email, checks a spreadsheet, searches another application, enters the order and asks someone to resolve an exception. The order system remembers the entry. The employee remembers the afternoon. Those are rather different accounts of the same transaction.

This is why a company can buy several capable applications and still leave its staff doing the joining. Each system has a job. The person supplies the continuity. Calling the resulting delay an employee productivity problem can be a wonderfully convenient way to avoid discussing the software.

Scout belongs in process intelligence and task mining. Alternatives include Celonis, UiPath and IBM Process Mining; the categories and capabilities overlap. Soroco emphasizes interaction data across applications. Its May 2024 partnership with Apromore is revealing: the two proposed combining ERP transaction insights with human-machine interaction insights. The transaction and the detour can inform the same analysis.

A factory for white-collar work

Rohan Narayana Murty, Arjun Narayan and George Nychis founded Soroco in 2014. Their academic backgrounds span Harvard, MIT and Carnegie Mellon. The company’s central question sounds almost old-fashioned beside today’s AI vocabulary: can office work be understood with the attention that manufacturing has long given its processes?

In 2022, Murty described the absence of a white-collar equivalent to manufacturing’s measurement discipline. Reporting that year described Soroco as bootstrapped. The appeal is easy to understand: a factory manager can inspect a production line, but an office process may disappear into hundreds of screens. The screen is visible; the route through it is harder to reconstruct.

Soroco says it now serves more than 200 customers globally. Its website names Bayer, BT Group, Bosch, Nestlé, Unilever and Deutsche Bank, among others. These are organizations where a modest inefficiency, repeated often enough, becomes a budget item. Scout is aimed at operations and transformation teams with enough repeated digital work to make patterns worth examining.

The company’s published culture contains a useful little test: “So what?” Every feature, policy or artifact is supposed to answer who benefits. Another value is “Learnability beats experience.” These are statements of intent, rather than a window into every working day. Still, the first is a sensible demand to make of any analytics vendor. A handsome graph should have to earn its keep.

What six weeks exposed

In a UK telecom engagement delivered with Tech Mahindra, the customer was struggling with variable order fulfillment, billing and related operations. Its earlier approach lacked the integrated analysis needed to locate bottlenecks across teams. Soroco reports that Scout exposed the manual-processing problem within six weeks: only 40% of effort went to core business applications and activities. The rest included validation, workflow management and manual order creation.

The response involved changes to the operation, including IVR-based customer validation, rather than simply admiring the discovery. The case reports a 34% reduction in manual processing effort across 261 users. That is a vendor-published result for this engagement, not a promised result for every buyer.

The distinction between finding and fixing is the practical heart of the story. A graph cannot negotiate a new handoff between departments. It cannot make an owner accept responsibility for a broken field. It can give those conversations something firmer than recollection. After that, somebody still has to change the work.

The price of seeing clearly

Scout is enterprise software sold through licensing and a partner ecosystem. NelsonHall’s June 2023 assessment described per-user licensing and a base package of approximately $25,000. Treat that as a historical benchmark. A buying decision needs a scoped quote and a separate account of the people and implementation work required.

A sensible evaluation starts smaller than a grand transformation. Pick a workflow with a measurable problem: effort per order, repeated handling or slow completion. Agree on the baseline and what a successful change would mean. Examine variation with the people who do the work. Then choose an intervention whose result can be checked against that baseline.

Copy the decision sequence, even if you never buy Scout. Observe first. Compare candidate changes. Standardize where necessary. Automate where the economics justify it. The consumer-goods case is useful precisely because it makes the middle steps visible. An automation backlog is a collection of hypotheses, however confidently it is presented.

Now the colleague is an agent

Soroco’s current positioning extends the map to people and AI agents working together. Scout is powered by TribeScope, which Soroco describes as a foundation model built on digital interaction data. The current offering includes workflow blueprints, agent evaluation sets and analysis of handoffs and human interventions. These are advertised capabilities; their usefulness depends on the workflow being examined.

The argument is that an agent trained on written procedures inherits their omissions. Observed exceptions provide better material for specifications and tests. This is an attractive extension of the original work-graph idea: first learn how a team actually completes the task, then decide what a machine should attempt.

There are conditions attached. An incomplete sample can make an unusual workflow look typical. Work performed away from captured applications needs additional context. If the real obstacle is a policy nobody may change, a more detailed map will leave it intact. And if employees distrust the collection, the project has a problem before the first chart appears.

Soroco says collection is opt-in, users can pause it, sensitive information is scrubbed at the operating-system level, and insights focus on teams with anonymized IDs. Those promises should become concrete deployment controls and a conversation with the people whose work supplies the data. Trust belongs in the operating design.

The enduring proposition is less glamorous than a robot colleague, and more useful: businesses should know what they are asking people to do. The favorite automation project may be the wrong one. The apparent slow worker may be carrying three disconnected systems. A good map gives a company the chance to discover the difference before it spends again.