A decision to change how a public body gets something done, whether by contracting a service out, bringing it back in, making a part at home or moving a building, is often put to its decision maker as one number: what the change would save. This paper argues that, in one designed set of decision records, that number does exactly one job. It sets the direction of the recommendation. It does not weigh the options the record itself offers, it does not hold the quality the record itself worries about, and in a few places it does not even agree with the record's own words about time and about what its figures are called.
The question and the contribution
The records come from AEGIS, the safety and protection studio of the RISCON family. AEGIS publishes authored decision cases: synthetic settings, written by an author, that pose a question to a named role, offer options, mark the author's recommended option, list the inputs still missing, and say when the question would be opened again. A programme of 196 of these cases goes further. Each of its cases executes one small calculation, keeps a hash of its input, and records that a second run gave the same output. One of its 12 rules subtracts two costs from a third: the cost a change would avoid, less the cost that would remain after the change and the cost of leaving the present arrangement. The result is called the net saving, and when it is negative, a net cost. Twelve cases in the programme run this rule. They are this paper's material. They are synthetic, and nothing in them is a fact about any real budget, contract or facility.
The question is put to the record set, not to the world. When a decision record weighs a change of arrangement by an executed calculation of net saving, what does the calculation settle, what does the record leave outside its single line, and where does the record's own wording part from its own figures?
The paper's contribution is a close reading of where one designed record draws the edge of its arithmetic. It shows that the net figure carries the direction of all twelve recommendations; that the matters the records themselves call unknown fall outside the figure; that the middle option every record offers is argued in words and never priced; and that the records' units, labels and thresholds open small gaps between what is said and what is counted. None of this says what any option is worth.
Four debates the records touch
Four bodies of work bear on records of this shape, and each gives this paper a question to put to them.
The first is the normative rule for costs that lie in the past. Hal R. Arkes and Peter Ayton, The Sunk Cost and Concorde Effects: Are Humans Less Rational Than Lower Animals? (1999), describe the sunk cost effect as a greater tendency to continue an endeavour once money, effort or time has been invested in it, and treat it as an error against the normative cost-benefit rules of choice. Their article attempts to show that the adult form of the error is due to an overgeneralization of the rule "Don't waste". This paper reads the net-saving rule of the AEGIS records as written in the spirit of that normative standard: its three inputs are all costs still to come or still avoidable, and nothing already spent enters the subtraction. That reading is this study's own, not the article's claim about any record. The question it raises for the records is whether a rule that keeps the past out also keeps out what lies ahead but is hard to count.
The second is commensuration. Wendy Nelson Espeland and Mitchell L. Stevens, Commensuration as a Social Process (1998), define commensuration as the comparison of different entities according to a common metric. They note that it is evident in routine decision-making and a crucial vehicle of rationalization, analyse it as a mode of power, and discuss the cognitive and political stakes of calling something incommensurable. A net saving is a common metric in exactly this sense: it turns a medical post, a training course and a contract for logistics into amounts of the same unit. The question for the records is what they do with the things they themselves describe that the common unit does not take in.
The third is the make-or-buy choice of governments, which most of the twelve records are in fact about. Oliver E. Williamson, The Economics of Organization: The Transaction Cost Approach (1981), regards the transaction as the basic unit of analysis and holds that economizing is accomplished by assigning transactions to governance structures in a discriminating way, at the boundary between firms and markets and inside organizations. Trevor L. Brown, Transaction Costs and Institutional Explanations for Government Service Production Decisions (2003), applies a transaction cost framework, with institutional and market theories, to the way governments choose between producing services themselves and producing them through contracts, and shows that governments choose mechanisms to manage the transaction costs of different types of service. Oliver Hart, Andrei Shleifer and Robert W. Vishny, The Proper Scope of Government: Theory and an Application to Prisons (1997), ask when a government should provide a service in-house and when it should contract out, and argue that where contracts are incomplete a private contractor's incentive to reduce cost is typically too strong, because the contractor ignores the adverse effect on quality that cannot be written into the contract. Simon Domberger, Contracting Out by the Public Sector: Theory, Evidence, Prospects (1997), reviews contracting out as a reform that introduces competition for the market through tendering, and considers conditions, such as contractual incompleteness and the ownership of physical assets, that may impede efficient contracting. Taken together, this body of work puts the attributes of the transaction and the quality that resists contract at the centre of the choice. The question for the records is where they put those attributes, if a single subtraction carries the recommendation.
The same literature also says the movement runs both ways. A. Hefetz, Privatization and Its Reverse: Explaining the Dynamics of the Government Contracting Process (2004), finds that local government contracting includes movements from public delivery to markets and from market contracts back to in-house delivery, and that government management, monitoring and principal-agent problems are most important in explaining both. Among the twelve records, three ask about turning work over to a contractor and three ask about bringing work in-house or making at home. The question for the records is whether the same arithmetic serves both directions, and what it leaves to the monitoring that the record promises afterwards.
The fourth is the value of waiting. Avinash Dixit, Investment and Hysteresis (1992), observes that firms do not invest as soon as a price rises above long-run average cost but wait until it rises well above it, and explains this through the theory of investment under uncertainty: by analogy with financial options, waiting has positive value because time brings more information about a project's prospects. The article draws its observation from firms, and this paper does not carry its theory over to the records; it borrows only a question. Every one of the twelve records offers a middle course, such as a pilot, a partial switch or a shorter renewal, which in this study's reading is a way of buying time and information. How does a record whose only figure is a net saving treat an option whose point is to learn before committing?
The debate between the debates
These works do not agree on what a figure should carry. The normative rule in Arkes and Ayton asks that only incremental costs and benefits count, which favours a clean subtraction of forward costs. The transaction cost tradition in Williamson and in Brown asks that the choice of arrangement turn on the attributes of the transaction, which a single amount does not show. Hart, Shleifer and Vishny add that where quality cannot be contracted, cost and quality pull against each other, so that a cost reduction can come with an adverse effect on quality that the contract does not hold. Espeland and Stevens step back from all three and ask what is done when unlike things are made comparable in one unit, and what is at stake in refusing to do so. Dixit adds a reason why a figure that is positive today need not mean acting today.
This paper does not settle the debate. It uses it to read a record set that says plainly what it counts and plainly what it does not know, so that the distance between the two can be read sentence by sentence.
How the paper proceeds
Chapter 1 gives a short account of the data and method and then two case sections: what the net figure settles, and what stays outside it. An appendix gives the records read, word for word in this study's translation. The conclusion returns to the four debates, states the contribution again, and names the limits of a reading built on twelve authored synthetic cases.