Tokenization is an emerging topic for business software providers. Tokenization is the process of representing something valuable or complex with a simple computer-readable substitute to enhance security or improve the efficiency and usability of a system. Tokens are already widely used in commerce. It’s now common for sensitive data used in a transaction, such as a bank account number or credit card information, to be tokenized. However, there are use cases for tokenization of other transaction-related documentation, including invoices, sales orders and sales contracts to make these more useful in AI-embedded business software such as ERP systems.
In addition, tokenization might be a method of preventing fraud in factoring invoices and reducing costs in the factoring process. There are also security benefits. Unlike encryption, properly maintained tokens reveal nothing if they are stolen, and they can further enhance security by limiting their scope or duration as well as making them revokable.
Of course, at a time when a not-insignificant number of enterprises still use fax machines to exchange commercial documents, it might be a little early to consider tokenizing them. Maybe, but I coined the terms Integrated Business Planning and Continuous Accounting to highlight how technology could transform core business processes long before most enterprises were in a position to adopt them. I’ll also note that I’ve been wrong about what’s possible or likely to develop, because the gotchas or institutional barriers were not immediately apparent. Nonetheless, ISG Research asserts that by 2029, more than two-thirds of ERP systems will employ tokens to improve their performance and functionality, improving performance, increasing efficiency and lowering costs.
Tokenization of working-capital documentation is worth considering now, mainly because AI in all of its forms is evolving rapidly in a frenzy of innovation. Tokenization can come off as a
fintech buzzword, and it’s not for everything. But for managing certain commercial documents for multiple purposes, it has significant potential. Yet its future is impossible to assess and forecast with any certainty because of the permutations of potential outcomes of the changes that have already happened and those that will become reality over the next few years. Nonetheless, here are some possible ways that the technology might be used.
Tokenization in the context of an AI environment breaks down raw text into smaller, standardized units called tokens so a machine can process them mathematically, increasing the scope and speed of such processing while reducing ambiguity. This action converts unstructured data into a format that AI models can analyze, understand and learn to accurately and reliably perform tasks. In AI and agentic systems, tokens provide a way to make data more manageable and useful for a wide range of functions.
For predictive AI use cases, tokenization makes model features clearer, more stable and potentially more accurate. Eliminating inconsistent identification methods and duplicates reduces noise, enabling models to better identify and learn real patterns. This can improve the predictive quality of forecasts, trends and outcomes, or identify anomalies that require attention. Tokens enable binding counterparties and instruments across multiple systems or instances. This capability can dynamically and rapidly optimize working capital and liquidity, even down to the location or legal entity level. Since optimums in a commercial setting are fleeting, rapid forecasting cycles are useful only if the time and cost to effect changes is worthwhile. Since agentic systems can shrink latency and execution costs, there are likely to be far more practical opportunities to apply tokenization to improve performance.
Tokenization also facilitates deriving graph-native signals from systems. These are features derived from the often-complex structure of relationships—such as nodes and edges—rather than the standalone attributes of a record used in predictive AI. These signals capture how an entity sits in a network and how it connects and relates to others over time. Potential uses are models that, for example, are faster and better able to find risks, including fraud, as well as identifying customer-related opportunities and threats, such as upsell and churn. The technology also makes more operational data accessible, since training can use tokens instead of records with sensitive information (such as personal identification information or bank data) to remain compliant with local data privacy requirements.
In finance department operations, tokenization can provide additional security against fraud in at least these respects:
- Requiring fewer manual steps and therefore faster execution in reconciliation processes because tokens bind the full invoice-to-pay process, streamlining any multi-way match control or anomaly detection across systems.
- Providing greater security for bank and personal identity information because approvers, outsourcers and other parties only handle tokens instead of raw data.
- Instilling a control that provides greater document integrity because the signed token encodes the details into a hash value. A change to any of the details breaks the validity of the item.
- Ensuring a unique invoice token so the same invoice can’t be paid twice, even across multiple systems. It also prevents factoring fraud because the same invoice cannot be pledged twice.
In retrieval-augmented generation (RAG), tokenization enables prompts to reference a specific document or section token such as an invoice or clause in a contract. So, RAG reliably references the specific source, resulting in fewer hallucinations and mis-joins. Since tokens carry signatures and provenance, an assistant can cite and preview the correct artifact while also enabling detokenization only for those with adequate permissions. Using tokens to compose emails, summaries, change orders and credit memos allows stitching token-linked facts (such as amounts, terms or SKUs) instead of scraping PDFs, making the process faster, more efficient and potentially more reliable. This can be especially useful in cases where specific contract or version tokens are used to prevent mixing old terms into new quotes, correspondence or renewals.
Tokens can also be used for agentic systems. Agents can be constructed with scope-limited tokens to more reliably control their actions. For example, an agent used by a specific human may be limited in the size or scope of a purchase order that can automatically be approved (cannot exceed $500 and only for vendors in schedule A), enabling automation with high-level controls. And “schedule A” can be a pointer to a dynamic list of approved providers.
In AI operations, idempotency allows certain operations to be performed multiple times without changing the result beyond the initial application. Using tokens makes it safe for agents to source or query the same data multiple times, ensuring that a duplicate request will produce the same outcome as the first and not cause unintended side effects. For example, using document tokens as idempotency keys avoids duplicate payments, orders or credits. Agents can also verify signed tokens before acting, and then, on completion, register token-based logs for end-to-end auditability and rollback. Where tasks or processes orchestrate multiple agents (for example, billing, collections or support), these pass tokens, not raw files, reducing latency, ensuring accuracy and promoting efficiency.
I was initially inspired to think about tokenization in business processes because of a recent spectacular bankruptcy of a major corporation that seemed like it involved fraudulent duplicate pledging of its receivables. A potential knock-on benefit of tokenizing a mass of relatively small-value invoices would be improving the efficiency of the factoring process to lower total processing and control costs, not just defaults. Although preventing fraud should by itself be a reason to apply the technology, decades of experience should tell us that this isn’t enough. There are multiple use cases for tokenization as a foundational technology used in business systems, even beyond those listed here. It may be early, but I strongly recommend that business software providers examine how to utilize tokens to enhance the applications’ usefulness and performance.
Regards,
Robert Kugel
Fill out the form to continue reading.