The Centaur Finance Function
By David Shebay
October 2026
7 min read
Chess kept evolving into a pure machine game because its boundaries are fixed. Finance and accounting do not have completely fixed boundaries, and that difference should shape how we build the finance function of the next decade.
The match that changed the question
In May 1997, Garry Kasparov lost a match to IBM’s Deep Blue, and many observers concluded that chess had ceased to be a meaningful human contest. That conclusion proved wrong. By Kasparov’s own account, interest in the game grew after machines surpassed the best human players.
Finance leaders now face the same question in a different arena. If a digital worker can reconcile, classify, compute and draft, what remains for the professional? Chess offers a useful answer, and also a precise point at which the analogy stops.
Advanced Chess and the lesson of 2005
The winning formula in human-machine chess was process, not talent and not computing power. Kasparov responded to his defeat by promoting partnerships in which a human player works alongside an engine, a format that became known as Advanced Chess, and its practitioners as centaurs.
The decisive evidence came in 2005, when Playchess.com hosted a freestyle tournament open to any combination of people and computers. The winners were two amateur players running three ordinary computers, who defeated grandmasters and the chess supercomputer Hydra. Their advantage was skill at coaching the machines to look deeply into the right positions.
Kasparov’s conclusion, which I paraphrase here, was that a weak human with a machine and a better process outperformed a strong computer alone, and, more remarkably, a strong human with a machine and an inferior process. In a world where everyone has the machine, the process is the asset.
In a world where everyone has the machine, the process is the asset.
Why chess endured: a bounded world
Chess endured because it is a closed system: sixty-four squares, thirty-two pieces, fixed rules, one objective and an unambiguous result. A move is either legal or it is not, and the rulebook has been effectively stable for generations. That stability let the game absorb the engine, which changed how players prepare and train without changing what the game is.
Finite does not mean small. Claude Shannon’s well-known estimate puts the number of possible chess games at roughly 10120, far beyond exhaustive search. Machines prevailed by searching and evaluating inside boundaries that never moved. Chess is therefore the strongest possible case for the machine: a bounded game, perfectly specified, where the human edge narrowed over time as engines improved.
Finance and accounting: a game whose rules keep changing
Finance has no fixed board, because the rulebook is rewritten while the game is in play. Each of the changes below altered more than a calculation: it altered the boundary of what counts as a correct answer.
| Change | Authority | What moved |
|---|---|---|
| Revenue recognition | ASC 606 (ASU 2014-09); IFRS 15 | Industry-specific rules replaced by a five-step model that depends on judgment about performance obligations and variable consideration |
| Leases | ASC 842 (ASU 2016-02); IFRS 16 | Lease obligations brought onto the balance sheet, with new estimates for discount rates and lease terms |
| Credit losses | ASC 326 (ASU 2016-13), CECL | Move from incurred-loss to expected-loss estimation using forward-looking information |
| Insurance contracts | IFRS 17 | A new measurement model, effective for annual periods beginning on or after January 1, 2023 |
| Global minimum tax | OECD Pillar Two | A 15 percent minimum rate that requires new cross-jurisdiction computations |
| State sales tax nexus | South Dakota v. Wayfair (2018) | Removal of the physical-presence test, multiplying state collection obligations |
| US federal business tax | Federal legislation enacted July 2025 | Mid-year changes to provisions that required forecasts and tax provisions to be re-run |
The boundaries also move at three different speeds. Standard setters move in years, legislatures and tax authorities in months, and the client’s own business in weeks, through a new contract term, a new product or an acquisition. On top of that, the inputs themselves are ambiguous: no one hands the system a list of legal moves.
The parallel, drawn carefully
The machine should lead wherever the boundary is fixed, and the human should lead wherever the boundary is being set. The comparison below shows why the two domains divide the work differently.
| Dimension | Chess | Finance and accounting |
|---|---|---|
| Rules | Fixed and fully specified | Revised continuously by standard setters, legislatures and the client’s own business |
| Objective | One: checkmate | Several, often in tension: compliance, accuracy, speed, cost, decision usefulness |
| A correct answer | Unambiguous | Frequently a judgment within a range |
| Machine advantage | Total, once engines matured | Strong on bounded tasks such as matching, reconciliation, classification and variance flagging |
| Human advantage | Narrowed over time | Interpreting new guidance, setting estimates, deciding materiality, owning the sign-off |
Two consequences follow. First, the process premium that Kasparov observed should be larger in finance than in chess, because the boundaries shift and someone must keep retraining the process. Second, accountability does not transfer: management certifies the financial statements and auditors sign the opinion, so a person must always own the judgment.
The centaur operating model: training digital workers like new hires
The centaur model treats a digital worker as a new member of the team that must be trained, supervised and promoted on evidence. The 2005 winners did not trust the machine blindly; they coached it, checked it and decided. The same discipline applies to a finance team:
- Define the lane. Give each digital worker a bounded task with stated inputs, outputs and tolerances, so that its work can be judged as correct or incorrect.
- Train on the firm’s own standards. Use the client’s accounting policies, prior-period examples and review comments, not generic knowledge alone.
- Supervise in layers. One digital worker produces, a second reviews, and a person handles exceptions and approves the result.
- Measure before extending autonomy. Track accuracy, exception rates and cycle time, and widen a worker’s scope only as the evidence supports it.
- Retrain when the boundary moves. A new standard, a changed tax rule or a revised client policy triggers an update to the training and a retest against a prior period before the next close.
The fifth step is the one that chess never required, and it is what separates a finance centaur from a chess centaur.
I have written separately about the leader who manages this blended team of people and digital workers; that thought leadership on Digital Worker Leaders is available on LinkedIn.
What finance leaders should do now
Leaders should sort their work by how stable its boundary is, and assign each category to the right partner.
- Inventory the work by boundary. Separate tasks with fixed rules from tasks whose rules depend on interpretation or are changing.
- Move bounded work to digital workers. Start where inputs, outputs and tolerances can be written down.
- Keep people on boundary-setting. Reserve interpretation of new guidance, estimates, materiality and policy decisions for qualified professionals.
- Build a change-control loop. Link every standard, tax and policy change to a retraining and retesting step.
- Invest in process and in coaches. The 2005 winners were skilled at directing machines, and that skill is learnable.
- Name the accountable owner. For every digital worker output, a named person signs.
Chess could welcome the machine because its board never changes. Finance cannot count on that, so the winning finance function will be the one with the best process for combining digital speed with professional judgment, and the discipline to keep that process current as the rules move. The word cybernetics comes from the Greek for helmsman, and steering through changing conditions is precisely the skill this model asks of its leaders.
David Shebay
Partner | Automated Finance, CFGI
David Shebay leads CFGI’s Automated Finance practice, pairing the firm’s finance practitioners with AI and supervision technology to execute the Office of the CFO’s most critical work, with the accountability of a CPA and the speed of AI.
Get in touch →Build the finance function that wins the next decade.
See how CFGI’s Automated Finance pairs professional judgment with supervised digital labor, with the accountability of a CPA and the speed of AI.