Ahikam Kaufman

Replacing them, they argue, is an enormous price to pay for technology that could instead sit on top of what is there.

• The alternative is an intelligent execution layer that understands how data relates, which system is authoritative, which accounting policy applies, which entity owns a transaction, what evidence supports an entry, and who may approve it.
• Agents then perform reconciliations, journal entries, transaction validation, investigations, and close activities while the ERP remains the system of record.
• Buyers should ask when replacing a system is the better choice.

The post names the systems finance teams have built their operating environments around — SAP, Oracle, NetSuite, Workday, Salesforce, billing platforms, banks, data warehouses, and internal applications — then contrasts replacing them with adding an execution layer across them.

Read original

The control layer, not the model, is the hard part.

• Permissions, approvals, segregation of duties, transaction-level traceability, audit trails, model governance, and limits on what an agent may do all need to be built, monitored, and maintained.
• Because AI now touches forecasts, budgets, reconciliations, and the close, governance becomes a financial-control question, often making the Office of the CFO, rather than IT, responsible.

Starting with a Deloitte CFO Signals finding that 19% of CFOs own AI governance — more than CEOs, boards, audit committees, or chief risk officers — Kaufman lays out the questions that raises: what data did the AI use, where did it come from, what assumptions did it make, what actions did it take, who approved them, and can the output be traced to the underlying transaction, contract, invoice, or source system?

Read original · Original 2

The costs of an AI deployment are not all visible at procurement: workflows start depending on one platform, financial logic gets embedded in its configuration, team processes adapt around its limitations, and audit history accumulates inside it.

• When pricing changes, business needs shift, or an acquisition brings in a different ERP, switching costs become real.
• Ask about dependencies before signing, not after the technology becomes part of daily operations.

Kaufman cites a Deloitte finding that 84% of finance leaders expect technology sovereignty to significantly affect investment decisions through FY2027, then offers four procurement questions: does deployment require replacing existing systems, where will financial rules and process knowledge live, can supporting evidence be accessed independently, and what happens when you add an entity, change an ERP, or want a different approach?

Read original

Decision quality rests on the work underneath it — transactions validated, accounts reconciled, exceptions investigated, and numbers people can trust.

• A faster analysis tool working from incomplete data still produces a weak answer.
• The investment case should explain what the business will do with the time a faster close creates, such as seeing a cash shortfall earlier or catching a billing problem before it grows.
• That is how they urge finance leaders to look beyond productivity statistics.

A contract amendment that never makes it into billing distorts the forecast a more capable analysis tool then reads. If finance identifies and resolves the mismatch during the month, with supporting details attached, the business has more time to act — an outcome worth measuring alongside hours saved.

Read original

Finance teams building their own agents face the same questions as any other AI buyer — which model runs each process, how much context to send, and how many calls a task requires — plus finance-specific requirements: approvals, segregation of duties, traceability, audit trails, and model governance. Their conclusion: the cost of running an agent includes models, token consumption, orchestration, controls, monitoring, testing, and maintenance, all of which belong in the build-versus-buy calculation.

Kaufman builds on Tony J., the CFO of Gigamon, who wrote that tokens are becoming a new unit of enterprise economics.

• They extend the idea to finance, asking how many calls a reconciliation requires and how much consumption occurs when an agent investigates thousands or millions of transactions.
• As a maker of finance automation software, Kaufman says Safebooks includes unlimited token usage so costs stay predictable as usage scales.

Across the posts, Kaufman draws on CFO arguments and analyst survey findings, credits each source by name, then translates the ideas for finance buyers.

Read original

Anyone reading just one piece should start here to understand what they argue finance AI should and should not change.

Kaufman argues that AI adoption is no reason to replace core systems.

• They list what finance teams have accumulated inside the systems they already run — customer history, accounting logic, controls, permissions, integrations, custom workflows, institutional knowledge — and make that accumulation the main reason replacement is expensive.
• Their alternative is an “intelligent execution layer” across the existing stack: something that knows which system is authoritative, which accounting policy applies, which entity owns a transaction, what evidence supports an entry and who may approve it.
• Agents can then handle reconciliations, journal entries, transaction validation, investigations and close activities while the ERP remains the system of record.
• The piece takes a position rather than offering a product tour, and names systems by category (SAP, Oracle, NetSuite, Workday, Salesforce, billing platforms, banks, data warehouses).

For brands planning a campaign

Useful as the opening frame for thought leadership aimed at finance and accounting leaders who already have an ERP and are wary of rip-and-replace pitches. A fintech vendor can build a campaign around the build-on-top argument and the concrete question "which system is authoritative?" rather than around a migration story, which is a harder sell to this audience.

Kaufman builds an argument around one finding from Deloitte's CFO Signals survey: 19% of CFOs own AI governance, more than CEOs, boards, audit committees or chief risk officers.

• They add other survey figures: 93% of surveyed CFOs say their organizations use AI across multiple key functions; 44% use it for financial planning and budgeting; 41% use it to analyze financial data; 59% name balancing speed of deployment against risk management as a major governance challenge.
• They then list the questions this raises: what data did the AI use, where did it come from, what assumptions did it make, what actions did it take, who approved them, and can the output be traced back to the transaction, contract, invoice or source system?
• As agents move from generating answers to executing financial work, they argue, controls should meet the standard finance already demands of people: traceable, auditable, permissioned and grounded in the company's financial context.

For brands planning a campaign

The strongest angle for a governance-and-controls message. A fintech company could turn the traceability questions into a checklist or diagnostic for prospects to apply to their own AI deployments, crediting the survey rather than restating it, and pair each question with the control a finance AI product must provide.

It is the best starting point for anyone evaluating finance AI vendors and concerned about lock-in.

Kaufman asks CFOs to evaluate an AI deployment by how hard it would be to reverse.

• They identify dependencies that are easy to miss during procurement: workflows that start relying on one platform, financial logic embedded in its configuration, team processes built around its limitations and audit history accumulating inside it.
• They then describe why a company might need to change: pricing changes, evolving business needs or an acquisition that brings in a different ERP.
• They cite a Deloitte finding that 84% of finance leaders expect technology sovereignty to significantly affect investment decisions through FY2027.
• Their four procurement questions: Does deployment require replacing existing systems?
• Where will financial rules and process knowledge live?
• Can supporting evidence be accessed independently?
• What happens when you add an entity, change an ERP or want a different approach?

For brands planning a campaign

A vendor-neutral evaluation framework that a finance AI company could publish as a buyer's guide, mapping each question to a concrete commitment: portability of data and rules, access to evidence, and support for multiple ERPs. It reads as advice to the buyer, not a demo, making it better suited to thought-leadership placement than a feature post.

Kaufman starts with a Gartner figure: 45% of CFOs weight their AI investment toward productivity and only 20% toward decision quality.

• They accept that boards want more than hours saved, then argue that the two outcomes are linked in finance: decision quality rests on validated transactions, reconciled accounts, investigated exceptions and numbers people can trust.
• Their example is a contract amendment that never reaches billing, distorting the forecast that a more capable analysis tool then reads.
• If finance catches that mismatch during the month, with the evidence attached, it sees the exposure sooner and gives the business more time to act.
• They end with questions a business case should answer: Will the extra speed identify a cash shortfall earlier, catch a billing problem before it grows or support a spending decision with validated numbers?

For brands planning a campaign

Use this to shift an ROI conversation from hours saved to decision quality, using the contract-amendment example as an illustration. A fintech marketer can turn the closing questions into a short framework for customers building their own internal business case — more useful than another productivity statistic.

Kaufman responds to a piece by Tony J., CFO of Gigamon, noting that every prompt, agent action and context window costs money, and that usage can scale quickly when autonomous agents run thousands of actions.

• They apply the point to finance: which model should run each process, how much context to send, how many calls a reconciliation takes and how much consumption results when an agent investigates thousands or millions of transactions.
• They add finance-specific requirements: permissions, approvals, segregation of duties, transaction-level traceability, audit trails, model governance and controls over what an agent may do.
• They frame the full set as part of the build-versus-buy decision and describe Safebooks' choice to include unlimited token usage with those controls built in.

For brands planning a campaign

A concrete way to differentiate on predictable economics rather than model quality. A campaign could ask finance teams to price the hidden variable costs of an agent running a reconciliation, then position fixed or unlimited consumption as a control and budgeting feature, not just a pricing convenience.

They list the systems finance already runs on — SAP, Oracle, NetSuite, Workday, CRM, billing platforms, banks, data warehouses — and the accounting logic, controls, permissions, integrations, and institutional knowledge built into them, calling replacement an enormous price to pay.

• The alternative is an intelligent execution layer that understands how the data relates, which system is authoritative, which accounting policy applies, which entity owns a transaction, and who can approve it.
• Agents then run reconciliations, journal entries, transaction validation, investigations, and close activities while the ERP stays the system of record.

Opens with a blunt claim that enterprises do not need a new ERP, then lists the systems and knowledge embedded in them before introducing the layer approach.

No post in the corpus covers when replacing a system is the right call, so the argument reads one-sided.

For brands planning a campaign

A vendor selling finance automation can position Kaufman on migration pain and preserving a customer's existing stack. Their framing gives a partner a concrete counter to rip-and-replace competitors, but co-marketing should expect the argument to center on control and traceability, not speed alone.

Examples: Source 1 · Source 2

They cite Deloitte CFO Signals figures — 19% of CFOs now own AI governance, 93% say their organizations use AI across key functions, 44% use it for planning and budgeting, and 41% for financial data analysis — and note that 59% name balancing speed of deployment against risk as a top governance challenge. Approvals, segregation of duties, transaction-level traceability, audit trails, and model governance are controls finance teams must build, monitor, and maintain.

For brands planning a campaign

For a fintech selling governance, audit, or controls features, Kaufman makes the case that the CFO owns AI governance directly. The thought-leadership angle is control infrastructure for agents, not AI capability; expect Kaufman to address the Office of the CFO rather than IT.

Examples: Source 1 · Source 2

Kaufman extends the point to finance teams building their own agents: which model should run each process, how much context to send, how many calls a reconciliation requires, and how to optimize consumption as models and pricing change.

• They list the full cost of running an agent as models, token consumption, orchestration, controls, monitoring, testing, and ongoing maintenance, and argue these belong in the build-versus-buy calculation.
• Their stated answer is that Safebooks includes unlimited token usage, keeping economics predictable as usage scales.

Builds on a named CFO's argument, credits the CFO, then extends it to Kaufman's product category — a borrowed-authority format they also use with analyst research.

For brands planning a campaign

Useful for positioning against metered AI pricing. A partner can use this frame for cost-predictability messaging to finance buyers; expect Kaufman to argue in total-cost-of-ownership terms rather than per-token or per-seat rates.

Examples: Source 1

Their argument is that decision quality rests on the work underneath it — transactions validated, accounts reconciled, exceptions investigated, numbers people can trust.

• The example of a contract amendment that never reaches billing, then follows the data into the forecast, shows how a faster analysis tool can still work from an incomplete picture.
• Kaufman argues the investment case should say what the business does with the time a faster close creates: identify a cash shortfall earlier, catch a billing problem before it grows, or make a spending decision on already-validated numbers.

For brands planning a campaign

A vendor pitching automation should pair time savings with named business outcomes. Kaufman offers a ready argument for outcome-based return-on-investment framing; any co-created content should name a specific business decision the faster process enables.

Examples: Source 1

They cite Deloitte's finding that 84% of finance leaders expect technology sovereignty to significantly affect investment decisions through FY2027, and pose procurement questions: Does deployment require replacing existing systems?

• Where do financial rules and process knowledge live?
• Is supporting evidence accessible independently?
• What happens when an entity or ERP changes?
• Their answer for Safebooks is that it works across existing systems, with the ERP as system of record and a Financial Data Graph connecting financial context across them.

For brands planning a campaign

This is the corpus's strongest procurement-pushback frame. A partner can use it to reach buyers worried about lock-in or argue for interoperability; expect Kaufman to frame the product as the lower-commitment option, not the more capable one.

Examples: Source 1

Kaufman opens with a question or blunt claim, brings in third-party research from Deloitte or Gartner and credits analysts and CFOs by name, breaks the argument into short lines and question lists, and closes by describing how Safebooks addresses the problem, often with a reframing line such as token economics becoming operating economics.

• They also pick up other people's arguments — a CFO's post or an analyst survey — and extend them to finance.
• Taken together, the posts move from where AI runs, to who governs it, to what it costs, to what the business gets from it.

For brands planning a campaign

Co-marketing with Kaufman would most plausibly take the form of research-anchored commentary: a survey finding or partner data point, a practical translation for finance leaders, and a specific product or service tie-in. A brand should bring original data or a named customer story; generic promotional copy does not fit the format.

Examples: Source 1 · Source 2 · Source 3 · Source 4

The posts present Safebooks as a finance-AI platform that runs agents across existing systems with controls, traceability, and governance built in.

View example

These are references to published research, not a stated commercial relationship.

View example

This is a research citation, not a stated commercial relationship.

View example

Gigamon is named as the CFO's employer; the post does not describe any relationship between Kaufman and Gigamon.

View example

Epstein writes from the perspective of a former CFO and investor, often for an event or investing audience. Kaufman writes from a finance-automation product angle, focusing more on execution architecture, controls, and vendor reversibility.

Read original · Original 2

Johar frames governance as a domain-specific policy layer alongside probabilistic models, often writing for a technologist or platform-selection audience. Kaufman frames the case around an execution layer over the existing finance stack and a specific vendor's controls.

Read original · Original 2

Gurfinkel's posts center on FinanceOS and a finance-native ticketing product, using a multi-ERP customer example. Kaufman's posts center on Safebooks and a Financial Data Graph, with more emphasis on agent execution, token economics, and procurement reversibility.

Read original · Original 2

Epstein's posts speak to finance leaders from the perspective of a former CFO who now invests.

• They ask what actually changed in margins, pricing, retention, collections, or analysis—and whether any of it can be measured.
• That suits a fintech vendor whose thought-leadership angle is proving AI's value to a CFO audience, and whose buyers already accept that AI belongs in finance and need the economics argument instead.

You give up granular control detail.

• Epstein's posts frame the value and measurement question but do not explain permissions, approvals, segregation of duties, transaction-level traceability, or how to govern a specific agent action inside the ledger.
• A campaign built around “how to keep an AI process auditable” would need another source for the mechanics.
• Epstein's posts also promote a finance conference, making them less reusable as evergreen teaching content than a control checklist.

Examples: Source 1 · Source 2

Johar's posts take finance-specific cases—journal entries, invoice disbursements, three-way matching, reconciliations—and argue that probabilistic models must sit behind deterministic policy enforcement, authorization gates, dual sign-off, and audit links to the underlying transaction.

• One post also lays out a pre-funding ROI test covering outcome, fully loaded cost, and break-even threshold.
• That is a stronger match for controllers and finance technologists who want the mechanics, not a CFO-level narrative.

You give up the broad CFO framing around adapting systems, avoiding an ERP replacement, and keeping a route out of a platform once finance has committed to it.

• Johar's voice reads more as a technology and data leader than a CFO peer.
• One post comments on a CFO.com article and carries a vendor hashtag, so more of its substance rests on others' research than on original analysis.
• The campaign would need additional material to carry a CFO-facing storyline.

Examples: Source 1 · Source 2

The posts describe launching a finance-native ticketing system inside a broader finance platform and follow a named multi-ERP customer—Essence Group, with subsidiaries in the US, UK, and Spain—where outside firms keep local books in their own systems.

• They make the case that AI outputs can be trusted only when data is consolidated, mapped, and given financial context.
• That fits a vendor that wants its own platform story told through finance-operations examples rather than a neutral checklist.

You give up neutrality and control-specific depth.

• The content is heavily product- and company-framed, so it reads as vendor promotion rather than independent thought leadership.
• A fintech buyer assessing governance and controls gets less checklist-style teaching than Kaufman provides.
• The specific mechanics of permissions, audit trails, and traceability across an AI agent are also thinner, and the positions are tied to the author's own platform, limiting their value as third-party validation.

Examples: Source 1 · Source 2