> ## Content Index
> Fetch the complete content index at: https://www.elevatewithall.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Before You Automate: ROUTE - Deciding What Belongs on the AI Journey
- URL: https://www.elevatewithall.com/before-you-automate-route-deciding-what-belongs-on-the-ai-journey/
- Published: 2026-09-17T23:39:00.000Z
- Updated: 2026-10-19T20:45:47.000Z
- Author: Sarabjeet Arora
- Tags: Agentic SaaS, Enterprise AI Productivity

Part 3 of 4 · [Enterprise AI Productivity series](https://www.elevatewithall.com/demystifying-enterprise-ai-productivity/)

\[Recap for readers arriving fresh: Part 1 laid out five fault lines behind 2026's enterprise AI measurement crisis and introduced CLEAR, a 5-phase loop for turning AI habit into governed capability. Part 2 operationalized it through mechanisms, quarterly cadence, and the balance sheet artifact. This piece asks a question that must be answered before any of that begins: does this process even belong on the CLEAR journey?\]

## The Question Everyone Skips

Most AI adoption conversations start by asking what can be automated, completely skipping a harder and much earlier question: should this run on agentic AI at all, or does it belong somewhere else entirely? Gartner's own June 2025 prediction is the clearest warning available that skipping this question is expensive: over 40% of agentic AI projects will be canceled by the end of 2027, driven by escalating costs, unclear business value, or inadequate risk controls. ([Gartner Newsroom](https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027?ref=elevatewithall.com)) Gartner also names the specific pattern behind a lot of that waste: "agent washing," where existing RPA bots and chatbots are rebranded as agentic AI without underlying capability, pushed by hype rather than real fit. The firm estimates only around 130 of the thousands of self-described agentic AI vendors offer genuine agentic capability.

The necessary fix isn't adopting a posture of general caution toward AI; rather, it requires applying a disciplined decision framework before entering the Clarify phase, which I call ROUTE.

## ROUTE 

**R - RPA-Ready.** Rule-based, structured, stable inputs, high volume, and low judgment. The simplest version of this decision rule, borrowed directly from automation practitioners, states that if inputs are consistent and logic remains unchanged, the process stays on RPA ([Make, "RPA vs Agentic AI"](https://www.make.com/en/blog/RPA-vs-agentic-AI?ref=elevatewithall.com)). Existing RPA investments matter here too: ripping out a working, inexpensive, and deterministic bot to replace it with a probabilistic agent purely for an "AI-first" narrative is precisely the kind of misapplication Gartner's cancellation forecast warns against.

**O - Orchestrate via Agentic AI.** High volume or high complexity, multi-step, cross-system, and judgmentlight, yet too dynamic or unstructured for a rules engine to handle over time, such as shifting approval logic, multiplying categories, or changing interfaces. This represents a genuine growth zone: Gartner projects that 33% of enterprise software applications will include agentic AI capability by 2028 (up from under 1% in 2024), and that 15% of day-to-day work decisions will be made autonomously through agentic AI by the same year (up from effectively zero in 2024). This is also the exact point where CLEAR from Part 1 takes over once a use case is routed here through Clarify, Launch, Embed, Amplify, and Review.

**U - Upskill and Augment (Human-Assisted).** Moderate-to-high judgment, where AI drafts or recommends while a human makes the final decision. This is precisely the zone mapped out by the landmark Harvard Business School/BCG field experiment: AI-assisted consultants completed 12.2% more tasks, 25.1% faster, and at over 40% higher quality inside AI's capability range, but performed measurably worse outside it ([Harvard Business School AI Institute](https://aiinstitute.hbs.edu/back-tothe-beginnings-of-ai-at-work/?ref=elevatewithall.com)). The "jagged frontier" finding from that research functions as a direct routing instruction: augment where the frontier is favorable, and build in a mandatory human checkpoint precisely because workers cannot always discern in advance which side of the frontier a given task falls on.

**T - Trust to Humans (Human-Dominated).** High-stakes, empathy-driven, exception-heavy, and possessing a low tolerance for error; a category every routing framework must protect explicitly, lest it be automated away by default momentum rather than deliberate decision. This pattern aligns with the National Institute of Standards and Technology (NIST) AI Risk Management Framework and its emphasis on human oversight for high-consequence decisions, ensuring automation speed never compromises diagnostic integrity, regulatory compliance, or human accountability.

**E - Evaluate and Evolve.** Periodic re-routing as underlying models and organizational trust mature. Today's Trust-to-Humans category may become tomorrow's Upskill-and-Augment category as AI's jagged frontier smooths out in specific domains, while today's Upskill-and-Augment use case may eventually graduate to Orchestrate once the underlying judgment is well-enough understood to be encoded. This also closes ROUTE into a continuous loop shape identical to CLEAR, where Evaluation feeds back into a fresh routing decision rather than treating any initial categorization as permanent.

## A Practical Routing Check

Before scoring a use case into CLEAR's Clarify phase, five questions decide which ROUTE category it belongs in first:

**● Volume and repeatability:** does this occur often enough, and in a sufficiently similar shape, to justify automation at all? 

**● Structure and rule stability:** does the underlying logic shift constantly, or does it remain stable long enough for fixed rules to hold? 

**● Judgment and stakes:** does a failure carry costs that demand accountability a model cannot shoulder? 

**● Error tolerance:** can a failure happen silently and get caught downstream, or does an error propagate before anyone notices? 

**● Existing investment:** is there already a working RPA solution in place, and does replacing it deliver genuine value beyond a corporate narrative?

A pattern of yes, yes, low, high, and yes points directly to RPA-Ready. A pattern of yes, no, low-to-moderate, moderate, and no points toward Orchestrate. High judgment with real stakes but clear efficiency upside points to Upskill-and-Augment. High judgment with real stakes alongside low error tolerance points straight to Trust-to-Humans, regardless of how easily automated the surrounding steps might appear.

## ROUTE in Practice

Four quick walk-throughs help clarify how these categories resolve during an actual backlog review:

**Invoice matching against a purchase order:** high volume, stable rules, low judgment; an RPA solution likely already exists. Routes to RPA-Ready. The temptation to upgrade this into an agentic workflow simply because agentic AI is currently trending represents the exact agent-washing pattern warned against by Gartner, as nothing here requires reasoning or fixing.

**Triaging inbound customer health-score alerts across dozens of systems, with shifting categories and thresholds:** high volume, unstable rules, cross-system, moderate judgment. Routes to Orchestrate via Agentic AI, and enters CLEAR's Clarify phase as a genuine candidate.

**Drafting a first-pass contract redline against a standard playbook**: moderate-to-high judgment, meaningful efficiency upside, but real financial and legal cost if a clause is missed. Routes to Upskill and Augment, where AI drafts, a lawyer decides, and the human checkpoint is the core value.

**A performance or termination decision informed partly by AI-flagged productivity data:** high stakes, low error tolerance, real human consequence. Routes to **Trust to Humans**, full stop, regardless of how automatable the underlying data-gathering steps are.

## Where Routing Meets the Balance Sheet

ROUTE does more than just dictate what to build; it establishes what standard of proof a use case owes, linking the framework directly back to the balance sheet artifact from Part 2\. An Orchestrate use case runs the full CLEAR loop and owes a validated outcome tied to a real baseline. An Upskill-and-Augment use case owes a lighter-touch quality and adoption signal rather than a full P&L line, given that the human decision-maker remains fully accountable. RPA-Ready processes continue reporting through existing operational metrics since they were never candidates for an AI P&L anyway, preventing unnecessary noise. That precise differentiation, ensuring not every routed use case owes the same kind of proof, is a design principle baked directly into the three-tier dashboard structure outlined in Part 2, where each function-level view tags a use case by category before determining the applicable accountability.

## What This Sets Up

Three pieces in, one underlying argument: enterprises in 2026 are not struggling because AI fails to work; they are struggling because adoption outran the systems needed to route it, run it, and measure it. Part 1 named the five fault lines and introduced CLEAR as the loop for turning individual habit into governed capability. Part 2 made that loop operational through management mechanisms, a rapid enterprise cadence, and a balance sheet artifact giving CFOs visibility into capability rather than just costs. This piece adds the essential prerequisite decision: not everything belongs on that journey, and identifying the correct category beforehand separates the 5% of organizations BCG terms "future-built" from the 95% still trying to trace misplaced budgets.

Naming categories and executing loops remains a theoretical exercise until given a concrete home. The balance sheet artifact from Part 2 is not hypothetical; it is a fully functioning prototype I have built featuring a board and portfolio view, a CFO ledger, and function-specific dashboards for Customer Success, Sales, and Engineering, all reconciled against a single shared data layer with every tracked use case tagged by its ROUTE category. Part 4 closes this series with a complete walkthrough of that dashboard, illustrating how CLEAR, ROUTE, and the balance sheet function together on screen rather than across separate articles.

This is Part 3 of a 4-part series on measuring enterprise AI productivity. Parts 1 (the five fault lines and CLEAR) and 2 (mechanisms, cadence, and the balance sheet) are linked above; Part 4 (a full walkthrough of the AI ROI dashboard) is coming soon.