Framing: the sentence that decides before the arguments begin
Direct continuation of epistemic marking: the epistemic operator sits at the level of the sentence; the frame sits at the level of the dossier. Framing is the most upstream human act in a tender response -- prior even to the identification of requirements. From another angle, it rejoins what was already at work in the executive summary, in what the CCTP does not say, and in the analysis of anaphora on the Transformer's vocation.
In 1981, Amos Tversky and Daniel Kahneman published in Science an article that would reconfigure the theory of rational decision. "The Framing of Decisions and the Psychology of Choice" sets out an experiment that seems, at first glance, to have nothing extraordinary about it.
Subjects are presented with the following scenario: an Asian epidemic threatens to kill six hundred people, and two response programs are under consideration. First formulation: "Program A will save 200 people; program B has a 1/3 probability of saving all 600, and a 2/3 probability of saving none." Seventy-two percent of subjects choose A.
The same question is then put -- exactly the same -- to another group, but reformulated: "With program A, 400 people will die; with program B, there is a 1/3 probability that no one will die, and a 2/3 probability that all 600 will die." Seventy-eight percent of subjects now choose program B.
Same options. Same probabilities. Same expected outcomes. Preferences reversed by nothing but a reformulation. The result is so robust that it will be replicated hundreds of times and will enter every cognitive psychology textbook as the canonical case of the framing effect.
This is the mechanism that silently drives the reading of a technical proposal in a tender response.
What a frame is
Framing has less to do with the argument than with what the argument takes for granted before it begins.
Charles Fillmore, in Frame Semantics (1982), lays down the operational definition: a frame is a background cognitive structure activated by a word or a phrase, which determines which elements will be perceived as relevant, which will be ignored, and which relations will be held to be self-evident. When someone utters the word "buy," your brain automatically activates a commercial frame that contains a buyer, a seller, a good, a price, a transfer of ownership. You do not need the sentence to mention these elements -- they are there, through activation of the frame.
When a set of specifications writes "migration to a new ERP," the frame activated in the reader is one of technical continuity: there is a current system, there will be a new system, the passage from one to the other is the stake. Everything is taken for granted: that an ERP is needed, that the successor must resemble the predecessor, that the business processes will survive the operation without structural modification. The frame itself is not argued -- it is imposed by the formulation.
A response that merely describes how the migration will be conducted remains a prisoner of the CCTP's frame. It may be excellent on execution, yet it creates no differentiating value. A response that proposes an alternative frame -- "refoundation of the business value chain on the occasion of a technical switchover," or conversely "stabilization of the existing base to preserve operational control" -- immediately displaces the question. The buyer is no longer reading a comparison of executors, but a strategic proposal.
Lakoff and the politics of framing
George Lakoff generalized the observation to all political communication. In Don't Think of an Elephant! (2004), he documents the systematic framing work undertaken by the American right since the 1970s, driven by Frank Luntz and the Heritage Foundation.
Emblematic example: "tax relief." The phrase becomes dominant in the 1990s and imposes itself on the entire American fiscal debate. Yet the expression contains a complete frame. Relief presupposes that there exists an affliction from which one must be delivered; the tax is therefore, through activation of the frame, an affliction. Whoever opposes "tax relief" is, by construction of the frame, the agent who maintains the affliction. The Democrats, by replying "we oppose Bush's tax relief," spent twenty years arguing within the Republican frame -- and reinforcing it with every reply.
Lakoff sums up the rule: "when you argue against a frame, you activate it." And an activated frame, even when contested, gains ground against itself.
Ronald Reagan, in his inaugural address of January 20, 1981, executes perhaps the most profitable reframing operation in American political history: "Government is not the solution to our problem; government is the problem." The sentence contains no data, no demonstration, no argument. It lays down a frame -- the state as problem -- that will structure thirty years of debate and configure the public reading of any dossier involving public action. When Bill Clinton declares, in 1996, "the era of big government is over," he will be operating within the frame Reagan had laid down fifteen years earlier.
Why framing operates more deeply than the argument
The cognitive reason was formalized by Kahneman in Thinking, Fast and Slow (2011). The human brain has two processing systems: a fast, automatic, associative System 1, which activates frames; a slow, deliberate, analytical System 2, which produces arguments. The temporal hierarchy is asymmetric: System 1 activates the frame before System 2 mobilizes its arguments, and System 2, when it intervenes, almost always works within the frame already activated.
Practical consequence. An excellent argument within an unfavorable frame loses against a mediocre argument within a favorable frame. The bid manager who truly wants to win does not attend only to the quality of their arguments -- they attend to the frame that will be activated in the buyer before the arguments begin to be read. And that frame is played out in the opening sentences of the chapter on understanding the need.
This understanding is not a contemporary innovation. Classical rhetoric had formalized it under the concept of the exordium -- the opening sequence of a speech, whose function is threefold according to Cicero (De Oratore) and Quintilian (Institutio Oratoria): to capture the goodwill of the judge (captatio benevolentiae), to awaken attention (attentum parare), and to prepare understanding (docilem facere). All three operations play out before the first proof, because the battle is decided before the arguing begins.
The executive summary and the chapter on understanding the need fulfill, in a tender response, exactly these three functions. They capture the evaluator's goodwill by demonstrating mastery of the context. They awaken their attention by proposing a fresh reading. They prepare their understanding by laying down the frame within which the rest of the dossier will be read. A technical proposal with no exec sum, or with a standard exec sum, has already lost part of the evaluation before the evaluator opens chapter 1 -- the defeat owes less to deficient content than to the missed exordium.
Framing in bid management: three decisive moments
Three passages of a technical proposal concentrate the framing stakes.
The executive summary and the title of the chapter on understanding the need. These are the two doors through which the evaluator enters the dossier -- the exordia in the classical sense. The first sentence of the exec sum, the title of chapter 1, and the two or three opening lines that follow them activate a frame that conditions the reading of the pages to come. "Understanding your ERP migration project" accepts the CCTP's frame and signs the executor's position. "Understanding your transformation: what a change of ERP makes possible" displaces the question, without frontally contesting it. "Understanding your context: a change of ERP as a lever for stabilizing the IS base" displaces it again, in another direction. The senior bid manager chooses their title by cognitive configuration of the reader, more than by aesthetics.
The anchoring sentence of the solution. Once the need has been reformulated, the solution must inscribe itself within the adopted frame. "Our approach is built around three pillars" is a standard, expected, neutral frame -- which produces a standard, expected, neutral reading. "Our organization treats your transformation as a change of organization equipped with a system, rather than as a change of system requiring a change of organization" installs a thesis in eighteen words. The entire rest of the dossier will be read within that thesis.
The comparative frame of references. A track record presented as "we have run forty-three ERP migration projects" activates a quantitative frame (the mass of experience). Presented as "we have intervened on three transformations comparable to yours, from which we draw the lessons set out here," it activates an analytical frame (the transferability of experience). The second frame is massively more powerful for the senior evaluator, because it signals a meta-cognitive capacity -- the ability to extract from an engagement what transfers and what does not.
The rule of legitimate reframing
Not every reframing is legitimate. A reframing that ignores the client's explicit request is read as a non-compliance; a reframing that frontally contradicts the contracting authority's culture is read as expert arrogance. The practical rule is precise: the reframing must open rather than replace.
Concretely, the technical proposal keeps the literal wording of the CCTP in the sections where it is expected (contractual commitments, deliverables, schedule) and proposes its alternative frame in the positioning sections (understanding of the need, vision, added value). The evaluator finds what they are looking for in the places where they look for it, and discovers a fresh reading in the places where they expect one. The reframing operates without assaulting the client's frame.
Why LLMs reproduce the dominant frame
The architecture of large language models is, by construction, hostile to strategic reframing.
The training distribution encodes the dominant frames. The pre-training corpus contains millions of responses to problems formulated within the majority frame of their era and their sector. When you ask an LLM to draft a chapter on understanding the need for an ERP project, the most probable completion reproduces the majority frame of the CCTPs it saw during training -- hence the frame the client has already adopted in their CCTP. The AI accepts the brief's frame, because that is statistically the most frequent completion.
RLHF rewards conformity. Human annotators prefer responses that "answer the question asked." A response that reformulates the question before answering it is massively downvoted as "off-topic" or "deviant." The gradient therefore pushes the model toward the response that obediently inscribes itself within the question's frame -- exactly the opposite of the move made by the senior partner who opens their answer with "before I answer, I would like to put the question differently."
The work of Bender et al. (FAccT 2021) on "stochastic parrots" made the phenomenon explicit more broadly: LLMs reproduce, with high fluency, the frames present in their data, without creating new ones. Santurkar et al. (2023), in "Whose Opinions Do Language Models Reflect?", measured the alignment of contemporary models with the dominant political framings of the training sources -- and confirmed that AI production reflects, without distinguishing itself from, the majority framing of the corpus.
Consequence in bid management. An AI-generated technical proposal almost always accepts the CCTP's frame. The understanding of the need reformulates the brief's statement without displacing it. The solution inscribes itself within the reading grid imposed by the buyer. The track record is presented within the standard quantitative frame. For a senior evaluator, the signature of the AI dossier owes less to over-confidence or to the saturation of tricolons than to the flat conformity to the contracting authority's frame, without the slightest attempt at a cognitive proposal.
Human reframing as an act deeper than prompting
In the age of reasoning models, there is a recurring debate over the levers for improving AI production. Prompting -- the art of formulating the request -- and fine-tuning -- the supervised adjustment of the model on a specific corpus -- are the two canonical technical approaches. Both operate within the model's generation frame.
The strategic reframing performed by the human is of another nature. Without instructing the model or retraining it, it redefines the space of possibilities within which the model will produce. Since the advent of reasoning models, this operation is one of the most fertile levers of human-AI collaboration, and probably the most profitable of them all.
The reason is mechanical. A Transformer produces a probability distribution over the next token, conditioned on the context. The context includes the brief, the examples provided, the explicit instructions -- but above all, silently, the cognitive frame activated by the first words. A model asked to "draft the chapter on understanding the need for an ERP migration project" activates the completion distribution conditioned by the thousands of ERP migration chapters seen during training -- a distribution whose mode is, by construction, the sector's majority frame. The most probable completion is also the most banal.
The same model, asked to "draft the chapter on understanding the need on the premise that this ERP migration project is, in reality, an organizational transformation equipped with a change of system," operates on a different distribution, conditioned on an alternative frame. The most probable completion is no longer the same. The model accesses inferences, analogies, anchorings that did not appear in the first distribution. Chained reasoning, since the appearance of o1-type models and their successors, multiplies this effect: an alternative frame laid down upstream orients not only the final completion, but the intermediate reasoning tree that produces it.
This shift of distribution is more powerful than a style instruction, deeper than a fine-tuning on a few thousand examples, sharper than an explicit reasoning chain added after the fact. The frame genuinely embodies a redistributed space of possibilities, which lets the AI extract itself significantly and reliably from the statistical consensus.
The consequence for bid management is precise. The division of labor between human and AI is inverted relative to spontaneous intuition. Strategic framing moves up to the human; the drafting production moves down to the machine; the coherence between the two passes through the quality of the frame laid down upstream. One hour spent reformulating the dossier's frame is worth, in leverage effect, ten hours of iterative prompting on a standard frame. It is the most profitable human operation in the chain -- far more than direct drafting or final proofreading.
Three operational disciplines
Identify the CCTP's frame before drafting. The senior bid manager devotes the first fifteen minutes of reading the CCTP to identifying the contracting authority's implicit frame, not to listing the functional requirements. Which verb recurs most often? Which lexical field structures the brief? Which questions does the CCTP not ask that could be asked? The identification of the frame conditions the strategic choice between acceptance and reframing.
Choose explicitly between adoption and displacement. On each major chapter, the bid manager makes a documented decision: adopt the client's frame (reliable-executor positioning) or propose an alternative frame (strategic-partner positioning). This decision is not entrusted to the AI. It is made upstream, and the AI is then mobilized to produce the drafting within the chosen frame.
Test the reversibility of the proposed frame. An alternative frame is solid only if it survives inversion. If one can imagine a credible competitor who would adopt the opposite frame with an equally defensible argument, then the proposed frame is merely a legitimate stance -- and the arguments defending it against inversion must then be prepared. If the inversion is indefensible, the frame is probably too obvious and brings no differentiating value. The right frame is the one that is defensible against a plausible inversion.
What remains to the human author
Strategic framing is one of the most profitable acts of a tender response, and one of the least reproducible by a generative AI. The reason has less to do with linguistic capacity -- models produce perfectly formed sentences -- than with architectural vocation: a Transformer trained by next-token prediction and refined by RLHF is optimized to produce the most probable completion of the context. Yet a reframing is, by definition, an improbable completion -- it gains in effectiveness the more it departs from the expected completion.
For a bid manager, a consultant, a communicator, framing stands as the most upstream decision, the one that configures everything that follows, rather than as a cosmetic layer added at the end of the dossier. As long as models are optimized for fluency within the dominant frame, it falls to the human to lay down the frame.
The machine can draft the content within the chosen frame.
The frame, you still have to lay it down yourself.
Principal sources: Tversky & Kahneman, "The Framing of Decisions and the Psychology of Choice," Science, 1981. Kahneman & Tversky, "Prospect Theory: An Analysis of Decision under Risk," Econometrica, 1979. Kahneman, Thinking, Fast and Slow, FSG, 2011. Lakoff, Don't Think of an Elephant!, Chelsea Green, 2004. Lakoff & Johnson, Metaphors We Live By, University of Chicago Press, 1980. Fillmore, "Frame Semantics," Linguistics in the Morning Calm, 1982. Entman, "Framing: Toward Clarification of a Fractured Paradigm," Journal of Communication, 1993. Bender et al., "On the Dangers of Stochastic Parrots," FAccT, 2021. Santurkar et al., "Whose Opinions Do Language Models Reflect?", arXiv 2303.17548, 2023. Cicero, De Oratore, 55 BC. Quintilian, Institutio Oratoria, c. AD 95.