Rootstack and the art of making AI useful

A patient wants to change an appointment. Somewhere inside a healthcare organization, the available slot already exists. So does the patient’s record. Yet a person may still have to answer the telephone, find the information, make the change, and confirm it. Rootstack’s account of one healthcare deployment begins with that familiar queue. Its opportunity sits between systems that hold information and people trying to use it.

The idea before the software

The company’s own origin story reaches back to 2010. Alejandro Oses approached university friends Diego Tejera and Juan Daniel Flórez with an idea for a technology business. Their starting material was practical: enterprise resource planning, customer management, and websites. Within a month, they had a name.

Then came a November conversation. In a retrospective essay, the founders describe meeting someone who changed their thinking, shifting their ambitions from local services toward products and scalable web services. The lesson concerned sequence: find out whether a market exists before completing the product.

The essay is unusually candid about getting things wrong. It acknowledges mistakes in sharing an idea too quickly. Its prescription is concrete: use a landing page, interviews, and conversations to test assumptions, including what customers would pay for. The reader can copy that experiment without hiring a development team.

The appointment is the point

Today Rootstack combines custom engineering with nearshore staffing, AI consulting, and managed delivery. Its company-provided Inc. profile names customers including Heineken, Hyundai, SURA, and Global Bank. The range matters: a bank, a retailer, and a healthcare provider may all need software, but their existing systems and operating rules differ.

In the healthcare project, Rootstack connected Rootlenses Voice to telephony, appointment management, and customer records. The agent could interpret a request, check availability, execute an authorized scheduling action, and confirm the result. Connections ran through APIs, the interfaces through which applications exchange information, rather than direct access to internal databases.

Rootstack reports a six-month implementation, approximately 55–65% automation of first-level calls, and an estimated 45% reduction in patient waiting time. These are the vendor’s figures for an unnamed customer. The more instructive detail is the boundary: callers could reach a human when they asked, encountered an exception, or needed support beyond the agent’s scope.

That boundary gives a prospective buyer something useful to inspect. A scheduling demonstration should include the awkward request, the unavailable slot, and the transfer to a person. The work is complete only when the appointment system records the intended change.

An agent needs somewhere to work

A separate collections case makes the same point with money. Rootstack describes connecting voice agents to billing and CRM systems, retrieving account information before calls and recording outcomes afterward. Campaigns could vary by overdue balance, contact history, and account status. Disputes and negotiations could pass to human representatives.

The reported rollout took seven months, proceeding incrementally by collection segment. Rootstack estimates a 40% reduction in cost per contact. That figure is an operating result claimed for this deployment; it is not the price of buying the platform. The sequence is revealing: automate a defined class of conversations, observe outcomes, and keep specialist intervention available.

Rootlenses also includes Insight, which turns natural-language questions into database queries and visual answers, and MCP, a layer for connecting AI models with business tools under defined permissions. Rootnet covers more familiar business territory, including CRM and intranet applications. Together they show a services company building reusable software alongside client work.

The appeal of Insight is straightforward: a business user can ask about data without writing SQL. Its product description includes role-based access and connections to databases such as PostgreSQL and SQL Server. Access control belongs in the purchase discussion alongside the quality of the answers.

Who runs the work?

Rootstack sells several arrangements. Staff augmentation adds people to a client-led team. Managed teams bring delivery management. Managed services cover continuing operations, while discovery work defines scope and architecture before development. Choosing among them requires an honest assessment of the management capacity the buyer already has. A capable internal technical lead may want additional engineers; a business without that lead may need someone responsible for the whole delivery. The same developer can sit inside very different agreements.

Clutch lists $50–$99 hourly rates and a $50,000 minimum project. Historical reviews describe smaller engagements, so those listings should be read as current positioning rather than a universal invoice. One marketing client reported spending about $20,000 monthly during December 2020 and January 2021, using flexible development capacity.

“We started by building websites and doing visual work.”

Alejandro Oses, recalling the company’s beginnings

Reviews also expose friction. One client described team assignments taking time to settle; another said a restriction on hiring Rootstack personnel directly had influenced provider choices. Buyers seeking permanent recruits should examine those terms. Buyers seeking managed delivery should ask how replacements, testing, and accountability work.

Borrow the experiment

In the market, Rootstack overlaps with nearshore engineering firms such as BairesDev, broader consultancies such as Globant, and talent networks such as Toptal. Its offer combines engineering capacity, integration work, and products of its own. The company lists Pantheon among its partners and reports ISO 9001 and ISO 27001 certifications, relevant signals for buyers evaluating delivery processes.

The transferable idea remains the early one: test the assumption before expanding the commitment. Pick a workflow, establish its starting cost, define what the software may do, and watch what happens when it cannot finish. Rootstack’s most useful promise becomes tangible at that scale: a question answered, a record updated, an appointment changed.