How to automate 70% of your operation in 8 weeks

Why 8 weeks and not 18 months

Automating 70% of a mid-market operation is feasible in eight weeks if the method is executed with discipline. The figure is not marketing. McKinsey estimates that between 60% and 70% of work activities contain tasks that are technically automatable with current technologies, including generative AI (McKinsey Global Institute, “The Economic Potential of Generative AI”, 2023). Deloitte documents that implementations fail not for lack of a model, but for excess scope and lack of cadence (Deloitte AI Institute, “State of Generative AI in the Enterprise”, Q4 2024).

The method we used at Cootradecun, a credit union with more than 40 thousand members and nine coordinated areas, reached 85% of tickets resolved without escalating to a human in its first cycle. The difference against a traditional project is not the technology. It is the cut of the scope and the control of the cadence.

Week 1 to 2: a measurable diagnostic

The first mistake of long projects is to start building before measuring. The diagnostic applies an AI Readiness instrument to the real operation. Eight organizational dimensions are measured (informational, environmental, infrastructure, participants, process, customers, data and technology), following the academic body synthesized in Nortje and Grobbelaar (2020) and later extended by Holmstrom (2022). The deliverable is not a presentation. It is a heat map with candidate processes, available data, connectable systems and regulatory risks.

The operational output is three to five processes prioritized on a double criterion: volume and reversibility. Volume because ROI arrives fast. Reversibility because the first deployments must allow correction without damage. Banco Interamericano de Desarrollo recommends this dual prioritization in its digital transformation guides for LATAM (BID, “Servicios públicos digitales”, 2022).

Week 3 to 5: the first agent in production

The rule is direct. One agent. One area. One channel. At Cootradecun, the first agent handled the member service conversational queue over WhatsApp Cloud API, routing to subagents specialized in certificates, memberships and PQR. The architecture follows three principles:

• On top of what already exists. No migrations first. The agent sits on the current CRM and core.
• Explicit decision bands. The agent has bounded authority. Anything outside the range is escalated to a human with full context.
• Audit ledger from turn one. Every action emits a signed, immutable event. This satisfies SARLAFT, Habeas Data (Ley 1581 de 2012) and the requirements of the Superintendencia Financiera (SFC, Circular Externa 029 de 2022).

See conversational architecture for the technical detail of the router and the subagents.

Week 6: extension to operational back office

With the first agent stable, the scope widens to a back office operation that depends on documents: certificates, validations, simple reconciliations. This is the operations layer, where a multi-agent orchestrator executes over the document corpus with strict RAG. Gartner reports that organizations combining conversational agents with back office workflows obtain 2.4 times more ROI than those deploying chatbots alone (Gartner, “Hype Cycle for Generative AI”, 2024).

Week 7: intelligence over the operation

The audit ledger accumulated over six weeks already has useful volume. An intelligence layer is built with BigQuery and semantic modeling in dbt. Dashboards with role-based views: the manager sees load; the area coordinator sees case types; the team sees productivity. Stripe describes this dynamic in its annual reports as a “feedback economy”: the agent’s data goes back to tune the bands (Stripe, “The Hidden Economy of API Calls”, 2024).

Week 8: handoff and quarterly cadence

The eighth week does not close the project. It delivers it. The customer’s team is left operating with runbooks, monitoring and the quarterly adjustment ritual. World Economic Forum identifies this pattern, early delivery and continuous adjustment, as the common trait of organizations that sustain adoption beyond the first year (WEF, “The Future of Jobs Report”, 2025).

The 70% is not a promise of technical coverage. It is the fraction of tasks that a well bounded agent resolves today in an average mid-market operation. What determines whether that fraction is captured in 8 weeks or in 18 months is scope discipline.

Why long consulting engagements fail

BCG and Deloitte report that the main predictor of failure in AI transformations is initial over-architecture (BCG, “Where’s the Value in AI?”, 2024; Deloitte, 2024). Designing for an 18-month horizon assumes the stability of an environment that changes quarterly. The consultancies that sell those horizons end up delivering documentation. The implementations that produce ROI deliver turn by turn.

The Cootradecun case documents the full cycle with public numbers. The methodology is replicable when the customer’s team keeps weekly decision capacity and the provider operates with visible cadence.