Skip to article
Datapath technology background for an AI adoption and governance briefing
Back to News
MARKET UPDATE • Published October 30, 2026 • 8 min read • Source: TELUS vendor case study

Burning AI Tokens? Turn Usage Into Business Value

TELUS, Rakuten, and BBVA report AI workflow gains. Datapath explains how to measure business value and scale usage with security and governance.

Dan J Sturdivant, Vice President at Datapath

By

Vice President

cybersecuritydata securitymanaged IT

Key takeaways

  • TELUS reports 100 billion monthly tokens through its multi-model Fuel iX platform; consumption alone does not prove return on investment.
  • Evaluate completed, quality-checked workflows alongside cost, turnaround, errors, and human review effort.
  • Scale useful AI work with approved data, limited permissions, budgets, monitoring, and clear responsibility for consequential actions.

Original source

TELUS vendor case study

Does burning more AI tokens mean a business is transforming?

More AI usage can support meaningful business change, but token consumption is an activity measure, not a business result. At Datapath, we recommend measuring completed work, quality, cost, and risk together. Expand a workflow when the evidence supports it, with the data boundaries and operating controls needed to keep that expansion manageable.

October 30, 2026 analysis. This briefing reviews published enterprise case studies and security guidance. The examples describe different organizations, time periods, and measurement methods; they are not a controlled comparison or a promise of equivalent results for your company.

The security conversation behind this briefing

The Saurav post supplied for this discussion concerns the Polymarket third-party script incident. It is not evidence about AI token consumption or enterprise productivity. Our separate Polymarket briefing explains the observed script behavior and the limits of the incident evidence.

Bloomberg’s October 29 X post raises a separate operational-security issue: military investigators were examining freezer failures. Reuters’ reporting, published by Military Times, describes a suspected hack and says it could not determine whether investigators had reached conclusions. This is not evidence about token consumption or an AI-caused incident. It is a reminder to include critical operational dependencies in a business technology review.

The relevant connection is operational: understand the external code, services, and permissions a workflow depends on. Monitoring browser JavaScript can support website defense, but it does not by itself secure an AI agent or establish that a business process is safe.

What does large-scale AI use look like in practice?

A vendor-published TELUS case study reports that Fuel iX processes 100 billion tokens monthly. Fuel iX provides access to multiple models, including Claude. The same case reports 57,000 active generative-AI users, more than 13,000 custom solutions, over 500,000 cumulative hours saved, and engineering teams shipping code 30% faster.

Those are company-reported results published by a supplier. The monthly token figure concerns the platform; it should not be presented as Claude-only consumption. The time savings are cumulative, not a monthly rate. The page does not provide enough measurement detail to independently calculate a causal return on each token or reproduce the results for another organization.

The useful business question is what changed in the workflow. Access to an approved platform can make experimentation easier, while ownership, training, and integration determine whether a useful experiment becomes repeatable work.

Which business outcomes have other companies reported?

In Anthropic’s Rakuten case study, the company reports feature time-to-market falling from 24 working days to five, a 79% reduction. It also describes a seven-hour autonomous coding run with occasional human guidance. These are specific reported engineering outcomes, not a general claim that every employee became 79% more productive. The case does not supply a token total that would let us calculate cost per successful delivery.

OpenAI’s BBVA case study, dated June 11, 2026, describes an internal assistant used by more than 3,000 employees in Peru, with average query-handling time reduced from approximately 7.5 minutes to around one minute. The publication also reports roughly three hours saved per employee per week across its broader deployment. These are vendor-published customer results, not an independent finding that every banking task became faster by the same amount.

ExampleReported operational signalWhat the evidence does not establish
TELUS Fuel iXHigh platform usage and reported cumulative time savingsA universal financial return per token
Rakuten engineeringShorter reported feature delivery timeThe same improvement for every role or project
BBVA internal assistanceShorter reported handling time for a defined query workflowEquivalent gains on every banking decision

These examples point toward redesigning specific work: retrieving knowledge, preparing drafts, building software, and helping employees move through an approved process. They do not justify replacing quality checks with a consumption target.

Why can useful agents consume so many tokens?

Longer tasks can involve repeated model calls, retrieving context, using tools, checking intermediate work, and trying again after an error. A token is a unit used in model processing; it is not a completed task or a consistent unit of business value across different models and workflows.

In its June 13, 2025 engineering account of a multi-agent research system, Anthropic says agents used about four times as many tokens as chat interactions in its data, and multi-agent systems about fifteen times as many. It also says the task value must justify the increased cost. These are observations from that system and period, not universal forecasts for a company’s next AI project.

Before expanding an agent workflow, review whether repeated calls are doing useful work. Investigate duplicated retrieval, unnecessary parallel work, retry loops, and outputs that staff must repeatedly repair. A larger bill may reflect more valuable work, more activity, or more waste; the invoice alone cannot distinguish them.

What should a business measure instead of token burn?

Use a scorecard that links usage to an accepted outcome. Start with a baseline for the existing process, including staff effort, elapsed time, quality checks, and rework. Compare a limited pilot against that baseline using similar tasks rather than only its most successful demonstrations.

A useful operating measure is cost per successful, quality-checked workflow. Include model usage, tool charges, relevant infrastructure, and human review or correction effort. Keep the acceptance test stable enough that a lower cost does not simply reflect weaker standards. Record failures and incomplete work rather than quietly excluding them from the denominator.

For an illustrative support-drafting pilot, success might mean a correct draft approved by an employee, not merely a generated response. Track turnaround time, corrections, escalation frequency, and the final resolution alongside usage. This is a proposed measurement approach, not a Datapath customer case or a financial projection.

Which controls make scaling safer and easier?

OWASP’s excessive-agency guidance recommends limiting tools and permissions, enforcing authorization in downstream systems, and requiring human approval for high-impact actions. Put these decisions into the workflow design. An assistant that drafts a message does not automatically need permission to send it, change a record, or approve a payment.

Give each integration a named owner and the access required for its job. Separate draft preparation from consequential execution. Validate actions against business rules outside the model, and test attempts to use another employee’s information or exceed the allowed task. Written instructions are useful, but they do not replace application-level authorization.

OWASP’s unbounded-consumption guidance addresses resource abuse through controls such as quotas, rate limits, timeouts, and monitoring. Define workflow budgets and escalation thresholds before scaling. Confirm how limits are enforced; an alert about spending is not necessarily a hard stop on future calls. Prepare a safe fallback for interrupted work.

Our incident response runbook guidance explains how to assign escalation and recovery responsibilities before an urgent event.

Also decide which information may enter the system, where it is retained, who can retrieve it, and which vendor terms apply. Keep useful audit records without turning logs into an unnecessary collection of confidential prompts or customer data. Train employees on the approved route and the point at which they should ask for help.

NIST’s Generative AI Profile offers voluntary risk-management guidance. It can help structure the work; following a framework does not automatically establish legal compliance, certify an implementation, or eliminate risk.

Why does operational resilience belong in an AI plan?

AI workflows may depend on systems that support facilities, logistics, or physical operations. Map those dependencies and their owners before authorizing an integration to make changes. A tool that summarizes a maintenance request and one that can alter equipment settings need different access decisions. These are illustrative design questions, not claims about the reported freezer investigation.

NIST’s operational technology security guide emphasizes the performance, reliability, and safety requirements of systems that interact with the physical environment. Bring the responsible operational and security teams into the review. AI governance, browser-script monitoring, and equipment security address different parts of a business workflow; a useful rollout connects their owners and response plans.

How can Datapath help turn a pilot into dependable work?

Datapath helps connect AI adoption with accountable owners and practical implementation. Our AI governance consulting addresses readiness, approved tools, information handling, vendor review, policy ownership, and implementation priorities. Our vendor risk management services help organize supplier evidence, access review, incident handoffs, and remediation tracking.

Bring one workflow, its current cost and turnaround, the information it uses, and the actions an assistant would be allowed to take. We can help organize the questions and responsibilities needed to evaluate a bounded rollout. Start with a measurable outcome and expand when the results and controls support the decision. High usage is useful when it serves the business; governance helps make that decision repeatable.

Ready to scale useful AI work?

Talk with Datapath about workflow ownership, data boundaries, vendor risk, and a practical AI rollout plan.

Plan your next step

Frequently asked questions

Is higher token usage evidence of better AI return on investment?

No. It measures consumption, not the value of accepted work. Compare costs with completed outcomes, quality, human review, and rework. Different models and workflows can use tokens differently, so a raw token total is not a universal productivity benchmark.

Do the reported enterprise results prove what our business will achieve?

No. TELUS, Rakuten, and BBVA describe specific deployments in vendor-published case studies. Their reported results can inform a pilot, but they do not establish the same outcomes, costs, or causal effects for another organization. Measure your own workflow against a defined baseline.

Does monitoring third-party JavaScript secure an AI agent?

No. Browser script monitoring supports one part of application defense. AI deployments also need appropriate data access, limited tool permissions, downstream authorization, evaluation, budgets, and response ownership. The controls should match the actual system and its allowed actions.

Disclaimer: This news summary is intended for informational and marketing purposes only, and nothing presented here is contractually binding or necessarily the final opinion of the authors.

Need to turn industry change into an IT plan?

Datapath can help translate security, compliance, and infrastructure signals into practical next steps for your organization.

Book an IT Consultation