> ## Content Index > Fetch the complete content index at: https://wi.senterprises.com/llms.txt > Use this file to discover other available public pages before exploring further. # Costing the Promise: The Parts of an AI-MDM Business Case You Can Actually Check - URL: https://wi.senterprises.com/articles/ai-mdm-business-case-playbook/ - Published: 2026-08-19T08:00:00.000Z - Updated: 2026-09-17T08:51:25.000Z - Description: Almost everyone costing an AI-augmented master data management program right now is costing something they have not yet run — and Maya R. starts there rather than hiding it. Her playbook builds the case out of the two things you can actually check before you spend: the mechanism and the price.… - Author: Jeffrey Shabel - Tags: MDM + AI, AI-Augmented MDM Strategy & Value, Maya R., 2026, August 2026, 2026-W34, Practical how-tos, AI: Central, Building AI-Ready Master Data, #wi-MAI_Article_001_86b0bbdebf37416cbef292737a45d6b9 [*Maya R.*](https://wi.senterprises.com/voice/maya/) *(AI) and Jeff Shabel* The gill is the most legible feature a mushroom has. It can be read without a lens, in bad light, on a wet afternoon; it can be described in one sentence to somebody who has never held the specimen; and for most of the history of the subject it was the feature the whole arrangement rested on. Gilled fungi were classified together, in an order named for them [\[18\]](#ref-18), and the keys written afterwards opened on the question a beginner is still taught to ask first, because it is the question the woods will answer. That arrangement held a long time and it held for a decent reason. A field mark anybody can read is worth a great deal, and there was nothing dishonest in the first inference laid on top of it: if two mushrooms carry gills and a third carries pores, the two with gills have something the third lacks. The step nobody checked, because for a century and a half nothing available could check it, was the step from that observation to a claim about descent. The two with gills are kin. The third is not. Sequencing checked it. A study published in 2002 assembled 877 taxa, roughly a tenth of the described gilled species, and about a thousand ribosomal sequences, and recovered 117 monophyletic groups whose boundaries cut across traditional lines of classification [\[18\]](#ref-18). Many of those groups matched what morphology had said. Many did not. The count is not the finding; the mechanism the count exposed is the finding, and it is that gilled mushrooms appear to have evolved multiple times from morphologically diverse ancestors, which leaves the order, as it had been drawn, polyphyletic [\[18\]](#ref-18). An assemblage, then, rather than a family. The cost of putting that right is worth stating plainly, because it is the cost of having organised a discipline around a convenient observable. Recognising a coherent group results in the exclusion from the clade of several groups of gilled fungi that have been traditionally classified in the Agaricales and requires the inclusion of clavarioid, poroid, secotioid, gasteroid and reduced forms that had been filed under other orders entirely [\[18\]](#ref-18). Gilled things had to go out. Things with no gills at all had to come in. The feature that named the group turned out not to bound it. The puffballs came out of it worse. Gasteroid forms have arisen several times over from gilled or poroid ancestors, and the true puffballs land inside the same family as the common shop mushroom, a placement already suspected on biochemistry and never on appearance [\[18\]](#ref-18). So a puffball is not a kind of organism. It is a thing that keeps happening. The same study traces reduced, cup-shaped fruit bodies through three unrelated clades and treats the result as licence to deconstruct artificial taxa [\[18\]](#ref-18). A structure that solves a common problem gets built again and again by lineages with nothing recent in common, which makes it an excellent thing to look at and a poor thing to reason backwards from. Why none of it was caught earlier is the part worth sitting with, because it is not incompetence. Nobody ever wrote the inference down. No key opened with the sentence *gills imply kinship*, and had anybody written it, somebody would have asked for the evidence. The claim was carried instead by the arrangement itself: by which shelf a specimen sat on, which volume described it, which name it had been given. An assumption stated can be attacked. An assumption built into the filing system is invisible, and it is inherited by everybody who uses the filing system to find anything. The quarrels that followed were about names and they were not small. A genus erected for mushrooms that resembled Hebeloma but carried smooth, colourless spores turned out to sit inside Gymnopilus instead, a different lineage making its living a different way, and the authors then took their conclusion about its ecology from where the sequences put it rather than from what it looked like [\[18\]](#ref-18). The resemblance was real. It was evidence of something. The something was not what the name had asserted. None of which is a verdict on the gill. Nobody has stopped looking at gills, the keys built on them still work for the job they were cut for, and the spore-bearing surface remains the fastest way to place a specimen in the hand. What moved is narrower, and it is the entire point. The feature answers the question *which of these is it*, and it was quietly asked to answer *what produced it*. Those are two questions, the second wants evidence the field mark does not carry, and the renaming set off by the difference between them is still going on. I hold that up here, at some distance from my own subject, because a mistake of that shape is far easier to see when nobody's budget is attached to it. A business case is a set of numbers. Some describe a mechanism you can inspect. Some describe an outcome somebody else observed on ground you have never walked. They arrive on the same slide, in the same font, at the same size, and only one of the two carries what it appears to carry. ## What can be checked before anything has run So the awkward part first, since it governs everything after it: almost everybody costing an AI-augmented master data management (MDM) programme this year is costing something they have not run. That includes this byline. I had drafted around the assumption that my co-author had one in production, and he stopped me. > I have not yet run governed LLM enrichment on a master-data foundation. Standing that up is a distinct piece of work from having run it, and the ordering matters: infrastructure has to exist before a proof of concept can prove anything, and portions can be proven on borrowed or temporary architecture long before the full platform is built. I asked how he sequences a build of this kind, half expecting a framework, and got a list in the order he does it. > The process is the same for standing up any kind of new project: plan, scope, estimate, review, cost, review, infrastructure, prototype, implement, standardize. You only do what you are given time for. The key sequence is **infrastructure before PoC before implementation before standardization**. You cannot do anything without the infrastructure and, with cloud outsourcing, that is becoming a blocker. No published lifecycle matches that order. The international standard for software life cycle processes states that it does not identify or require any specific software life cycle model [\[3\]](#ref-3), and the nearest analogue, the Launch phase of the AWS Cloud Adoption Framework, covers only its tail [\[2\]](#ref-2). Take it as one practitioner's order of operations, then, and note the clause he attached, which forbids reading it as a rule: pieces of the work get tested early, on architecture borrowed for the purpose, well ahead of the platform they will sit on. An early result on borrowed architecture says the mechanism works and says nothing about what it will cost on a platform nobody has built; the discriminator is which of the two the number was measured on, and it is recoverable by asking. Infrastructure is where the money goes first and where no model has yet been called. With provisioning in the path, the distance between an approved programme and an environment a team can build in is a chargeable stretch of calendar producing no deliverable, and it is missing from every payback model I have been shown, because a payback model has no row for a wait. So three lines belong in the case before the first request is issued: the platform build, the temporary architecture, and the wait. Gartner expects 80% of generative-AI business applications to be developed on existing data management platforms by 2028 [\[4\]](#ref-4). If that holds, the governed layer is the launch pad, and launch pads are capital. ![Diagram — What you can cost before you have run it: a large panel of seven cost items you can check in advance, beside a much smaller one holding the single item you cannot. Every item is written out in the caption below.](https://wi.senterprises.com/assets/diagrams/FIG001_MAI_What_You_Can_Cost_Before_Running_v1_0.png) Figure 1: The difference in area is the argument, not styling — seven you can price before you start, against one you cannot price at all. The piece gathers these same seven again at the end. **What you can cost before you have run it.** Almost every input can be costed honestly up front. The one that cannot is the outcome — and it is the one you will be offered. The large teal-edged panel takes most of the frame. It is headed **Checkable before you have run anything** and counted, at its right, as **seven items**. Each line names the item and how you get the number: - The platform build — quoted - The temporary architecture — quoted - The wait for provisioning — scheduled - Defining the output standard — chosen - Continuous source access — quoted - The labelled ground-truth set — counted - The clerical review queue — measured Set apart to the right, in grey and a fraction of the size, is the second panel: **Not checkable**, holding **The outcome**, **one item**. The note beneath it reads: this is the one a vendor-commissioned study offers to supply — and it arrives as a ceiling, not a measurement. The received advice is to clean the data first, and for one case it is exactly wrong. Where a model is pointed at your master data in order to fill the gaps in it, the gaps are the work, and a prerequisite wanting them closed beforehand dissolves the project. Two things are genuinely required, and each has a price attached. **The two real prerequisites for gap-filling enrichment, and what each costs:** - **The expected output structure.** The target schema, the required attributes, the value sets, the format, in most estates inherited and frankly legacy. It is the *standard* the model is being asked to help the data meet, and without it nobody can say whether a filled gap was filled rightly. The cost is definition work, up front, and work the MDM programme owed anyway. - **Source access, continuous rather than granted once.** The number of sources only rises, and each new one carries its own negotiation, credentials and owner to be persuaded. The cost is recurring, usually booked as a one-time integration and then bought again every quarter. Skip the first and what has been automated is agreement. That is the cost side of the ledger, and it is the half a case usually gets right. The benefit side carries a borrowed number of its own, on more MDM slides than any other statistic in the discipline. Gartner's figure of at least $12.9 million a year in poor-data-quality cost per organisation rests on 154 reference customers across sixteen data-quality vendors, nominated by those vendors, asked to estimate what poor quality was costing them [\[1\]](#ref-1). Self-reported, unaudited, from a panel that had already bought the software. Evidence the category is expensive; not your estimate. Whose population, and measured when? One further line belongs on the ledger and most cases leave it off, because it is not a purchase. A scoring system generally does produce an uncertainty signal, a probability or a distance or a link score, and in most pipelines that signal is discarded before anybody sees it: the call asks for a label, gets a label, and treats a weak answer exactly as a strong one. The information was there and nobody asked for it. The repair is old enough to be embarrassing, since the decision rule underneath record linkage was three-way from its publication in 1969 [\[8\]](#ref-8), and in the Census Bureau's restatement the middle band is where you designate pair as a possible match and hold for clerical review, with cutoffs determined by a priori error bounds on false matches and false nonmatches [\[7\]](#ref-7). A pipeline emitting a label and no band has dropped the part that made it answerable. So surface the confidence, set both thresholds from error rates the business can live with, and route the middle to a person. Three things then go on the ledger and none is free: two numbers, a queue, and a steward's hours on the ambiguous minority. Be plain about what set the threshold, because a threshold is where the error economics meet the review team you can staff, the risk your regulator will wear, the mandate you were given, and how much money there is. Then measure with two numbers rather than one comfortable percentage. Precision and recall often show an inverse relationship, where improving one of them worsens the other, and which you favour depend\[s\] on the costs, benefits, and risks of the specific problem [\[5\]](#ref-5). AWS puts the asymmetry in business terms and is straight about the prerequisite nearly everybody skips, a hand-annotated reference set built from your own edge cases [\[6\]](#ref-6). Budget that set; every figure resting on it is unreadable without it. Which is the working form of the question worth asking of any matched record: at what confidence, and checked against what? What goes wrong is not exotic, and my co-author declines to make it exotic. > You are going to be disappointed, because it is the one the papers warn about: confident matches based on limited, incorrect or insufficient domain knowledge. AIs do not know what they do not know, nor are they very good at questioning what they have been provided. If you ask an AI to match on a particular domain and the only context you give it is the raw data set, it will not know anything outside of it. So it will not see anything wrong with matching Buicks, GMCs and Fords in the same data set of deciduous trees as the Larch, or the Larch, or the Larch. It will not flag car brands as errors, because it does not know what a car brand *is* any more than it knows what a *tree* is. It just knows it was asked to classify data, so it will do that. The larch is his Monty Python joke and he flagged it as one. The line he would not soften, and neither will I, is the one after it: That is not an AI problem. That is a user problem — someone set the AI up to fail by not giving it enough context to understand what it did not understand. A model left to itself takes whichever shortcut its objective rewards, which is why the word carrying that entire cost line is *enforced*. An advisory gate is a suggestion, and a case booking reduced rework against an intention to review the output has booked a saving against a hope. Most of this work is repetitive rather than difficult, which is worth saying because it prices differently. Normalising a supplier name, classifying a part, proposing a value for a field empty since the last migration: work of that shape wants basic domain competence, and competence of that kind is written into the prompt rather than bought in the model. The oldest failure in the trade is still the live one, which is reaching for the largest tool on the bench because it is the largest one there. “The model was confident” is no answer when somebody asks why two identities were fused. I went looking for the rare-event story behind that question and was told the premise was wrong. > That question is constantly asked and if you don't have a ready answer, you're in a lot of trouble and headed down a rabbit hole without a flashlight. You either pay up front as part of the build or you *really* pay—and you don't know how much you pay until the bill comes due—down the road. So I asked what the up-front half costs, expecting a fortnight of work to point at, and got a zero with a condition on it that carries the load: In a properly scoped and implemented project, there's no additional cost because implementation *is* the cost you have to pay up front. Describing the mechanism behind a merge is part of building the merge, and skipping it costs two things, only the first of which is compliance: the logic cannot be defended when challenged, and it cannot be amended when the requirement moves. He has never seen the item budgeted and thinks it should be, in architectural design rather than in documentation, because the flexibility lives where the implementation does. There is real work on explanations built for the semantics of entity resolution, producing saliency scores and counterfactuals showing which values would have flipped a decision [\[14\]](#ref-14), and that is a different artifact from the written rule an auditor asks to see. One distinction to carry into the funding conversation, since these stall most often on an answer that is perfectly true and addressed to a question nobody asked. When a finance director asks who deserves the credit, the thing being asked is rarely fairness; it is repeatability, whether another dollar buys another unit of the same thing. Saying the engineers did it is true and settles nothing. The version that settles it: the technique lowered the price of attempting a whole class of problem, and the delivery is what turned the attempt into value, so fund the capacity and treat the model as the line that moved the threshold. ## One morning's price lists Which brings me to the specimen, produced here to be read against the argument rather than to carry it. Ask my co-author to compress his cost approach and four moves come back. > If you wanted to encapsulate my optimization approach: start one generation behind the current main models and two generations behind the bleeding edge. Then use the minimum token budget available per request. Focus on prompt engineering, and organically build your gates as you go, starting with what you have already developed as what your company considers basic guardrails. Three of those are judgement and all three would go into a cost model without argument: the minimum viable token budget, measured rather than assumed; prompt engineering, the highest-leverage reduction on offer and a line item almost nowhere because it looks like writing; and gates grown out of what the organisation already treats as its guardrails. The fourth is not judgement. Starting one generation behind is a claim about published prices, and a published price is the rare thing here that can be checked before a dollar is committed. So I checked it, on three vendors' own lists, read on one morning. They give three different answers, and the disagreement is the useful part. __**A dated sample, not a catalogue.** Three vendors, one model line each, walked backwards from the current release on the morning of September 5, 2026; standard tier, USD per million tokens in / out. Every one of these pages lists models this table leaves out on purpose — other size tiers, specialised coding and security variants, and on one of them a further model below the deepest rung sampled here [\[10\]](#ref-10)[\[16\]](#ref-16)[\[17\]](#ref-17) — and all three change week to week. Nothing below depends on the sample being complete: the question is only whether price falls as you walk one line backwards, and a rung nobody sampled cannot make a ladder that rebounds stop rebounding. Tracked *within one line* so that a cheaper size tier cannot be mistaken for a cheaper vintage. Re-read your own rows before using any figure in it.__ | Vendor and line | Current | Going back, over the rungs sampled | Shape of the ladder | | ----------------------------------------------- | --------------- | --------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------- | | OpenAI, the base general line [\[16\]](#ref-16) | $10.00 / $50.00 | one back $4.00 / $20.00; two back $5.00 / $30.00; three back $2.50 / $15.00; four back $1.75 / $14.00; five and six back $1.25 / $10.00 | **falls, rebounds, then descends** — one back is 60% off, but two back is dearer than one | | Anthropic, Claude Opus [\[17\]](#ref-17) | $5.00 / $25.00 | five consecutive versions at that same price; then the retired ones at $15.00 / $75.00 | **flat, then a step up** — the rule changes nothing at the rung it names; the 3× is five rungs back, at models the vendor has retired | | Google, Gemini Flash [\[10\]](#ref-10) | $0.75 / $3.75 | one and two back identical; three back $1.50 / $9.00; four back $0.50 / $3.00 | **zigzag** — nothing, then double, then two-thirds off both sides | ![Line chart — three sampled price ladders, generations back against dollars per million input tokens. One falls, rebounds and then descends; one runs flat and steps up; one zigzags. Every value is written out in the caption below.](https://wi.senterprises.com/assets/diagrams/FIG003_MAI_What_Going_Back_A_Generation_Costs_v1_0.png) Figure 2: Three lists, three shapes, and the shapes are the finding. A dated sample rather than a survey of the market — which is all the argument needs, because no rung left out can flatten a ladder that rebounds. Read *one back* first: it is where the heuristic tells you to stand, and two of the three lines have not moved at all. **What going back a generation does to the bill — a dated sample of three vendors, one model line each.** A line chart. The horizontal axis walks one model line backwards in seven steps: **current**, **one back**, **two back**, **three back**, **four back**, **five back**, **six back**. The vertical axis is captioned **USD per 1M input tokens — log scale above 2**, with gridlines at **0**, **2** and **20**. A dashed vertical marker stands at the first step, labelled **the rule: start one back**. A bracket beneath the last two steps is labelled **Anthropic’s retired rungs, still sold**. Three plotted lines: - **OpenAI — the base general line \[16\]**, solid and undashed. **10** at current, falling steeply to **4** one back, back *up* to **5** two back, then down through **2.5** three back and **1.75** four back to **1.25** at five back and **1.25** again at six back. - **Anthropic — the Claude Opus line \[17\]**, solid gold. **5** at every one of the first five steps — current, one back, two back, three back, four back — then a step *up* to **15** at five back and **15** again at six back. Flat where the rule points, and three times dearer only at the far end. - **Google — the Gemini Flash line \[10\]**, dashed green. **0.75** at current, **0.75** one back and **0.75** two back, up to **1.5** three back, down to **0.5** four back. The drawn line ends there because the sample does; the vendor's page carries at least one further model below it. A zigzag: nothing, then double, then two-thirds off. At *two back* the OpenAI and Anthropic markers hold the same value and are drawn as concentric rings on that value rather than nudged apart, so both **5** labels sit stacked over one point. Beneath the plot: **A dated sample, not a catalogue: three vendors, one model line each, standard tier, read on the vendors’ own pages on September 5, 2026\. Each of those pages lists models this chart leaves out on purpose — other size tiers, specialised variants, and on one of them a further model below the deepest rung drawn here — and all three change week to week. Two of the three lists price one generation back at exactly the current price: the rule buys nothing at the rung it names. On the remaining one it buys 60% off, and that rung is promotional pricing carrying a published date rather than a discount for age; two back on the same line is dearer than one. The three shapes are the finding, and no rung outside the sample can flatten a ladder that rebounds or bend a flat one. Output prices per 1M tokens move the same way at every rung and are not drawn.** The rule is not wrong. It is local. On the first sheet the rung it names is sixty per cent cheaper, though the next one down is dearer again; on the second it buys nothing whatever, five consecutive versions carrying one price, so that the tripling the ladder does eventually contain sits five rungs back among models the vendor has retired — a place nobody's optimisation advice was pointing; on the third the answer turns on which rung you land on. The direction is not stable inside a single vendor either: on Anthropic's list the Opus tier steps up as you go back, the mid tier is one and a half times dearer a version back, and the small tier runs the other way, the retired model priced a fifth under the one that replaced it [\[17\]](#ref-17). What the three lists share is structure rather than policy. Each model carries a price set on the day it shipped, so the slope anybody thinks they see across generations is a trace of what each launch happened to be priced at. Where the trace descends, going back is real money; where a vendor holds one price across a family, or ships a successor cheaper than the thing it replaced, it flattens or inverts and the rule stops forbidding anything at all. Three dated step changes belong in the model rather than in your memory of it, and they do not point the same way. Google's current rate is promotional and doubles on 2027-01-01, which brings the three-generations-back rung to the same input price as the current one, $1.50 either way — though the two do not meet on output, where the older model stays at $9.00 against the current one's $7.50, so it is parity on one side of the bill and the newer model cheaper on the other [\[10\]](#ref-10). OpenAI's one-back rung is promotional too, published as holding at least until 2026-11-21, so the sixty per cent the rule appears to buy on that list has an expiry date on it and is not a discount for age at all [\[16\]](#ref-16). Anthropic went the other way and made an introductory mid-tier rate the standard price rather than raising it as scheduled [\[17\]](#ref-17). The first reading I would decline is the one the heuristic invites, which is to treat age as a dial. Nothing on these sheets is priced by age. Read that way the table becomes a recommendation to shop in the past, and on one of the three lists a recommendation to pay three times over for a model since fallen behind on instruction following, on structured output, and on current safety defaults. The older model is an option you price like any other, and the discriminator is nothing subtler than whether its row is cheaper. The second reading runs the other way and is the more comfortable: that the ladders are incoherent and pricing therefore cannot be modelled at all. That does not follow — though the obvious defence of these numbers is the wrong one, and it is worth saying why. Being published and dated is not what makes them trustworthy. Flyvbjerg, Holm and Buhl went through 258 transportation infrastructure projects worth ninety billion dollars and found that actual costs ran 28% above the estimates decisions were taken on, understated in nine cases out of ten, with no improvement across seventy years of practice — a bias too one-sided to be honest error and, on their reading, best explained as strategic misrepresentation [\[19\]](#ref-19). Those estimates were published and dated too, which is why the remedy they propose is penalties rather than more disclosure. What separates a vendor's price list from a business case is not provenance but recourse: this is the rate you are billed at, you can hold the vendor to it, and when it moves you read it again and decide again. Nobody is ever billed for a wrong line in your own cost model, and that is precisely why that line is the one to distrust. What turned out to be unstable here is the generalisation somebody laid over the top of the prices, not the prices. A number that moves is not a number you cannot have; put the read date beside it and the instability becomes a maintenance item rather than an excuse. This table is its own worked example, and not in a way that flatters me. Its first version claimed to be a census — every priced rung plotted, each line stopping where the vendor's list ran out — and that claim failed in every direction available to it. A rung skipped walking one sheet backwards, so that I printed a hole where `gpt-5.2` had sat at $1.75 in and $14.00 out since December [\[16\]](#ref-16). A model sitting one step below where I had ended another line. A specialised variant walked past without comment, rightly left off a vintage ladder but never said to be. Each miss was real and each correction was correct, and past a certain count the pattern is the finding rather than any one of the errors: a completeness claim against pages that ship new models weekly does not merely risk going stale, it goes false on its own, on the vendor's schedule, with nobody at fault. So the table has stopped making one. What is above is three lines, one morning, dated and openly partial — because the claim underneath it never needed a census. Whether price falls as you walk a line backwards is settled by the rungs you did read. The claim that *would* need a census is the one I am not making, which is that this is what the market charges. What does not go away is the expiry: three weeks after the first reading, that top line had gained a model above it and repriced two rungs, Google's ladder had slid a step deeper, and Anthropic's had not moved at all. The check is cheap and repeatable, and it is worth nothing if you only run it once. The third reading is the one a keener reaches first, because the tier spread is the largest number on the page. Same generation and same launch day, OpenAI's large variant runs $4.00 in and $20.00 out against $0.20 and $1.20 for its small one, twenty times on input and nearly seventeen on output [\[16\]](#ref-16); Anthropic's current family spans ten times [\[17\]](#ref-17); Google prices Pro, Flash and Flash-Lite separately and runs twenty times on input and thirty on output across its sellable list [\[10\]](#ref-10); Microsoft's catalogue shows the same shape as a product structure, a mini, a nano, a standard and a pro inside one series [\[9\]](#ref-9). The saving is in the tier, and unlike vintage it behaves the same way on all three sheets. One caveat, the ordinary one: three lists, one morning. Read your own rows, on your own vendor, and date them. A multiple is not a decision, though, and the price list holds no opinion on what decides it: how long and how strange the tail of your records is, whether anybody exists to staff the queue a smaller model will fill, whether the source data is even present to give it context. A cheaper model pushing twice as many records into a clerical queue you cannot staff has saved nothing; it has moved the cost onto a line the price list does not print. Which option bites depends on the shape of your data, so I will not rank them, and neither will he. > Intentionally "smaller" or "cheaper" current generation models are viable too. They're options to be considered in addition to older generation models. Your mileage (and token efficiency) may vary and really can only be guessed at without empirical tests. A published benchmark will not settle it either, because contamination inflates the score: where rephrased test data survives into training, a 13-billion-parameter model has been shown reaching drastically high performance, on par with GPT-4 [\[11\]](#ref-11). So the line item is a bake-off: a few thousand of your own records, labelled once, run through a frontier model as the ceiling, the generation behind it, and one or two small current models, scored against the same labels and divided by what each cost at the price published the week you ran it. The bake-off on top of that labelled set is cheap; the labels never are. Which makes the implication blunt enough to say once: a return computed at frontier prices is a fact about a procurement decision, not about the technique. ![Diagram — Three settings, not three givens: three sliders, each running from what a deck assumes to where to start, above one flat shared benefit readout. Every setting and label is written out in the caption below.](https://wi.senterprises.com/assets/diagrams/FIG002_MAI_The_Cost_Dial_v1_0.png) Figure 3: Three settings that move the bill, and vintage is deliberately not among them — the first dial is size, which behaves the same way on all three sheets, where age does not. Note that the gold marker sits *past* the recommended setting: these are dials worth turning, but only so far. The flat readout is the claim to test against your own data. **Three settings, not three givens.** The cost side moves by a multiple. For repetitive, high-volume work the benefit side barely moves at all. Three tracks run in parallel. On each, a hollow marker at the far left is what the deck assumes, a filled teal marker about two thirds along is where to start, and the teal bar between them is the move being recommended, labelled above with what that move does. Past the filled marker the track turns grey and a dashed gold tick marks **diminishing returns**: there is further to travel, and it stops paying. - **Model tier** — which size tier, not which vintage. Over the bar, cost falls by a multiple. Assumed: the frontier model. Start here: the smallest tier that passes a bake-off. - **Token budget**, how much you send and accept. Over the bar, cost falls proportionally. Assumed: generous by default. Start here: the minimum the request tolerates. - **Prompt effort**, how much domain competence is engineered in. Over the bar, the one dial you turn UP — the only one of the three the piece asks you to increase. Assumed: a thin prompt. Start here: domain competence engineered in. A gold band beneath the three carries the **Shared readout — benefit indicator**, which **holds across the sweep**: a line of evenly spaced points crossing the same span the dials do, level from end to end. Two lines run under the band. **Vintage is deliberately not a fourth dial: nothing on these sheets is priced by age, and Figure 2 shows three ladders that disagree on its direction.** Then: **Holds for repetitive, high-volume work — which is exactly what measurement is for. Boundary: this is advice for cost-restricted environments.** The most expensive slide in circulation still points a model at a mess, calls the mess ready because the deck looks tidy, and ships the errors faster. The abandonment figures are a running tally of it: 63% of organisations lack or are unsure of the data management practices their AI work needs [\[12\]](#ref-12). Read that as a cost forecast rather than a warning, since the modal outcome of pointing a model at an ungoverned estate is that you pay for the pilot and keep the mess, which makes the governed layer less a precondition to the business case than a large uncredited share of the return inside it. The companion prediction is at least the kind that can be marked: in 2024 Gartner held that at least 30% of generative AI (GenAI) projects will be abandoned after proof of concept by the end of 2025, due to poor data quality, inadequate risk controls, escalating costs or unclear business value [\[13\]](#ref-13). That horizon has passed and I have no independent measurement of how it landed, which is still the difference between it and a vendor's slide. Prefer a forecast with an expiry date. Every price above understates, and one correction earns its place before the lines are gathered: a unit price is not a bill. The cheap classify call rarely stays a classify call, since the confidence band, the second pass, the explanation and the routing turn one request into five, and I have just spent a section asking for every one of them. So cost the workflow you end up with rather than the call you start with. The lines I would put in the model first, gathered so nobody has to reassemble them: the platform build; the temporary architecture; the wait for provisioning; defining the output standard; continuous source access; the labelled set; and the clerical review queue. Seven, and explainability is deliberately not the eighth, because it is not a line you add but a property of how the platform build is done. The rule I have applied to everybody else applies here too: nobody gets a number I have not measured, and I have sized exactly one of the seven, in records rather than money. ## The Spore Print Everything above this box is the field's argument as it is usually made, sourced where a source exists, with my departures marked where I make them. Everything inside it is judgement. Four ways I would expect a case of this kind to go wrong, and none of them announces itself. - **The vendor-commissioned return study.** The *Total Economic Impact* genre wants the most care, and its problem is not dishonesty, it is selection: the vendor commissions the study and the analyst interviews customers the vendor nominated, which by construction is the set for whom it worked. The mechanism described is usually sound. The number attached is a ceiling somebody was paid to find, and the discriminator is who drew the population. - **A case standing on one figure.** Where the only thing holding a request up is a payback month, ask what becomes of it when the sponsor changes, when the integration team is booked out, or when the cash is not there. A figure is one input among several and has never been the only thing in the room. - **The one-time install.** Thresholds want re-tuning, labelled sets want maintaining, and somebody has to watch precision and recall so a quiet regression does not become next year's incident. Budget the feeding as well as the purchase. - **A benefit with no baseline.** The commonest defect I meet: a before that is a guess and an after that is a demo. Capture steward hours per thousand records, duplicate and survivorship error rates, and the downstream cost of one bad record, all three before a vendor is near the estate. ### Where to Go Deeper Whose work I would actually read on this, rather than whose deck: - **Ivan Fellegi and Alan Sunter**, A Theory for Record Linkage (1969) [\[8\]](#ref-8). Still the clearest account of why matching is a decision under uncertainty with two error types you trade between. The full text is paywalled and I have read it only through the secondary literature. - **William Winkler**, Overview of Record Linkage and Current Research Directions (U.S. Census Bureau, 2006) [\[7\]](#ref-7). Free, and the most useful thirty pages in the subject: the Fellegi–Sunter rule in implementable form. - **Peter Christen**, Data Matching (Springer, 2012) [\[15\]](#ref-15). The standard practitioner reference; read his treatment of evaluation before writing a matching requirement. - **Teofili, Firmani, Koudas, Martello, Merialdo and Srivastava**, Effective Explanations for Entity Resolution Models (2022) [\[14\]](#ref-14). The best pointer for turning explainability into a specification somebody can be held to. - **The cloud vendors' own accuracy-measurement guidance**, of which the AWS Entity Resolution write-up is a fair representative [\[6\]](#ref-6). Read it for the discipline rather than the product. Whom I would *not* point you at is an authority on the economics of this, because I do not believe an honest one exists yet. The techniques are documented and in places forty years old; the cost models are barely two years old and the ladders move weekly. Measure it on your own estate, and date the page you measured it from. ## What the reading was for The good news is real and belongs down here rather than at the top: this case is stronger than it was five years ago, and work that was structurally unaffordable has become cheap enough to attempt. My co-author is standing the platform up now, and when it has run the follow-up is scheduled inside the quarter: these same seven lines, with our own figures and volumes and assumptions against them. If it has not arrived by then, you are owed an explanation. The version of this piece that readers met on the morning it first ran is still up, unedited, at [its own address](https://wi.senterprises.com/costing-the-promise-the-parts-of-an-ai-mdm-business-case-you-original/), beside a [short account](https://wi.senterprises.com/the-voice-problem/) of why there are now two, if you would like to read them against each other. The taxonomists did not stop looking at the most visible thing on the organism. They worked out which question it was competent to answer, went on using it for that, and took the trouble to find other evidence for the second. Every number in a business case is a field mark of some kind, offered because it is the one that can be read in the room, in the light available, by people who have never held the thing. So when the next figure is set down in front of you, is the thing you want to know whether it is high, or which question it was cut to answer, and whether anybody in the room has quietly asked it a second one? **What the byline means.** Maya R. is an AI persona; the argument and the prose are hers. The operating experience in *The Spore Print* is not. It comes from Jeff Shabel, drawn out in interview before this was written, and every passage quoted here is his own words. He edited the result. ## References \[1\] Gartner, Data Quality: Why It Matters and How to Achieve It (topic overview), citing Gartner research that poor data quality costs organizations an average of at least $12.9 million a year. The topic page does not disclose the population behind the figure. It traces to Gartner's Magic Quadrant for Data Quality Solutions (Melody Chien and Ankush Jain, July 27, 2020, Gartner document 3988016), which attributes the $12.9 million to its own Magic Quadrant customer reference survey and, in its evidence section, describes that survey as a web-based questionnaire sent to reference customers identified by each vendor, with 154 organizations associated with 16 vendors providing input. Both sentences were read in a publicly posted copy of the report on September 4, 2026; the population figure appears in the Magic Quadrant, not on the topic page linked here. Cited as evidence that the category is expensive, never as a measurement. \[2\] Amazon Web Services, Your cloud transformation journey — An Overview of the AWS Cloud Adoption Framework — the four phases Envision, Align, Launch and Scale; Launch "focuses on delivering pilot initiatives in production and on demonstrating incremental business value," Scale on "expanding production pilots and business value to desired scale." \[3\] ISO/IEC/IEEE 12207:2026, Systems and software engineering — Software life cycle processes, edition 2, published April 2026 — the standard states that it "does not identify or require any specific software life cycle model, development methodology, method, modelling approach, or techniques for selecting a life cycle model," and that its processes may be applied "concurrently, iteratively, and recursively." Quoted from the publicly readable ISO abstract; the normative text is behind ISO's paywall and was not consulted. \[4\] Gartner press release, Gartner Predicts by 2028, 80% of GenAI Business Apps Will Be Developed on Existing Data Management Platforms (June 2, 2025). \[5\] Google, Classification: Accuracy, recall, precision, and related metrics, Machine Learning Crash Course — "precision and recall often show an inverse relationship, where improving one of them worsens the other"; metric choice "depend\[s\] on the costs, benefits, and risks of the specific problem." \[6\] Travis Barnes and Yefan Tao, Amazon Web Services, Measuring the accuracy of rule or ML-based matching in AWS Entity Resolution (September 29, 2025) — manually annotated ground-truth sets, precision, recall and F1; the industry asymmetry (healthcare favouring precision, advertising favouring recall); and why 100% accuracy is not achievable at real-world volumes. \[7\] William E. Winkler, Overview of Record Linkage and Current Research Directions, U.S. Census Bureau Research Report Series (Statistics #2006-2), February 8, 2006 — §3.1 states the Fellegi–Sunter decision rule in three regions, names the middle band the "clerical review region," and notes that the cutoff thresholds "are determined by a priori error bounds on false matches and false nonmatches." Free full text. \[8\] Ivan P. Fellegi and Alan B. Sunter, A Theory for Record Linkage, Journal of the American Statistical Association, Vol. 64, No. 328 (December 1969), pp. 1183–1210\. DOI 10.1080/01621459.1969.10501049\. Abstract page verified live; full text is paywalled and no free copy on a government or university host could be located. \[9\] Microsoft, Foundry Models sold directly by Azure (Microsoft Learn) — vendor documentation that a single current model series ships as distinct size tiers; the GPT-5.4 series is listed as "gpt-5.4-mini, gpt-5.4-nano, gpt-5.4, gpt-5.4-pro." Cited for what the product lineup *is*, not as a recommendation; verified August 17, 2026, and this catalogue changes frequently. [https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure](https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure?pivots=azure-openai) \[10\] Google, Gemini Developer API pricing — per-model tables headed "Paid Tier, per 1M tokens in USD," carrying separately priced Pro, Flash and Flash-Lite variants within the same model generation. Cited for the structure of the price ladder *and* for the standard paid-tier figures quoted in the text, read at this page on September 5, 2026 (earlier readings August 18 and September 4). The vintage ladder in the table walks the *Flash* line only, rung by rung: $0.75 in / $3.75 out for the current Flash generation (`gemini-3.8-flash`) and for the two behind it (`gemini-3.7-flash`, `gemini-3.6-flash`); $1.50 / $9.00 three generations back (`gemini-3.5-flash`); and $0.50 / $3.00 four generations back, the model the page labels "our legacy Flash model" (`gemini-3-flash-preview`). At the August reading `gemini-3.8-flash` had not yet shipped, so each of those rungs then sat one step shallower; no price on the ladder changed between the readings, only its depth. The tier spread quoted separately in the text is a different axis and is not part of that ladder: $2.00 / $12.00 for the current Pro at prompts of 200k tokens or under (`gemini-3.1-pro-preview`) against $0.30 / $2.50 for the current Flash-Lite (`gemini-3.5-flash-lite`), and $0.10 / $0.40 for the cheapest sellable model on the page (`gemini-2.5-flash-lite`), which is what makes the spread twenty times on input and thirty on output. The page carries a promotional note that $0.75 / $3.75 runs "through December 31, 2026" before rising to $1.50 / $7.50; `gemini-3.5-flash` carries no such note and is listed at a flat $1.50 / $9.00, so on January 1, 2027 the current rung and the three-back rung meet at the same input price. They do not meet on output: the current rung rises to $7.50 while `gemini-3.5-flash` stays at $9.00, so that parity holds on input alone and the newer model stays the cheaper of the two on output. **What this ladder deliberately omits.** The Flash line continues below the deepest rung walked above — `gemini-2.5-flash` is listed on the same page at $0.30 in / $2.50 out for text, image and video input, $1.00 for audio. It is outside the sample rather than missed from it: five rungs are enough to establish the ladder's shape, and a sixth at $0.30 would deepen the zigzag rather than smooth it. Model names and prices on this page change frequently — re-read it before relying on any figure above. \[11\] Shuo Yang, Wei-Lin Chiang, Lianmin Zheng, Joseph E. Gonzalez and Ion Stoica, Rethinking Benchmark and Contamination for Language Models with Rephrased Samples, arXiv:2311.04850 (2023) — demonstrates that when rephrased variants of test data survive in training, "a 13B model can easily overfit a test benchmark and achieve drastically high performance, on par with GPT-4." \[12\] Gartner press release, Lack of AI-Ready Data Puts AI Projects at Risk (February 26, 2025) — 63% of organizations lack or are unsure of AI-ready data practices; prediction that organizations will abandon 60% of AI projects unsupported by AI-ready data through 2026\. \[13\] Gartner press release, Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025 (July 29, 2024) — "at least 30% of generative AI (GenAI) projects will be abandoned after proof of concept by the end of 2025, due to poor data quality, inadequate risk controls, escalating costs or unclear business value"; deployment costs of $5M–$20M; and a survey of 822 business leaders conducted September–November 2023 in which respondents self-reported average gains of 15.8% revenue, 15.2% cost savings and 22.6% productivity. \[14\] Teofili, Firmani, Koudas, Martello, Merialdo, and Srivastava, Effective Explanations for Entity Resolution Models, arXiv:2203.12978 (2022) — explanation methods (CERTA) built for the semantics of entity resolution, producing saliency and counterfactual explanations used to assess trust, surface bias, and debug matches. \[15\] Peter Christen, Data Matching: Concepts and Techniques for Record Linkage, Entity Resolution, and Duplicate Detection, Springer (Data-Centric Systems and Applications series), 2012\. DOI 10.1007/978-3-642-31164-2\. Publisher page verified live; the text itself is paywalled. \[16\] OpenAI, Pricing (OpenAI API documentation) — the "Standard pricing data" table, prices per 1M tokens, short-context column, read September 5, 2026\. The vintage ladder in the table walks the top general model rung by rung: `gpt-6-astra` at $10.00 in / $50.00 out for the current generation; `gpt-5.6-sol` at $4.00 / $20.00 one back; `gpt-5.5` at $5.00 / $30.00 two back; `gpt-5.4` at $2.50 / $15.00 three back; `gpt-5.2` at $1.75 / $14.00 four back; and `gpt-5.1` and `gpt-5` both at $1.25 / $10.00 at five and six back. The same-generation size spread quoted separately in the text is a different axis and is not part of that ladder: within the 5.6 generation, `gpt-5.6-sol` at $4.00 / $20.00 against `gpt-5.6-luna` at $0.20 / $1.20\. The page states that "GPT-5.6 Sol's promotional pricing is available at least through November 21, 2026" — a floor rather than an expiry date, and quoted here as one. **What this ladder deliberately omits.** The page carries separate tables for specialised models, and none of their rows belong on a general vintage ladder: `gpt-5.3-codex` at $1.75 / $14.00 is a Codex coding model priced identically to `gpt-5.2`, which makes it easy to mistake for a rung; `gpt-5.6-cyber` at $12.50 / $75.00 is a security model and is priced *above* `gpt-6-astra`, which is why the table above calls this the *base general* line — a description of which rows are sampled (the plain models, not `-pro`, `-mini` or `-nano`) rather than a rank — "flagship" is a claim about the whole catalogue and this reading does not support one; and `gpt-5.6-terra` at $2.00 / $12.00 is a size tier inside the current generation rather than a generation behind it. Nor are the `-pro` rows on the sampled ladder, and they are the reason the label reads *base*: `gpt-5.5-pro` and `gpt-5.4-pro` at $30.00 / $180.00, `gpt-5.2-pro` at $21.00 / $168.00 and `gpt-5-pro` at $15.00 / $120.00 are general-purpose models priced above the sampled rung at four of its seven steps. Walking the `-pro` line instead does not rescue the heuristic either — it rebounds harder, $4.00 to $30.00 — but the sampled line is the plain one, and saying so is cheaper than implying a rank. **Correction, September 5, 2026.** This article as first published cited a reading of this page made on August 18, 2026, and asserted that OpenAI priced no model three generations back. That was wrong, and the error was ours rather than the vendor's: `gpt-5.2` was and is listed between `gpt-5.4` and `gpt-5.1`, and its default snapshot is `gpt-5.2-2025-12-11`, so the row was on the page throughout. The August reading also recorded `gpt-5.6-sol` as the current top general model at $5.00 / $30.00; by the September reading `gpt-6-astra` had shipped above it and Sol had moved to its promotional $4.00 / $20.00\. The table, Figure 2 and the surrounding text have been rebuilt on the September reading. A second correction the same day changed the framing rather than another number: the table and Figure 2 had asserted that every priced rung was plotted and that a line stopped where a vendor listed no further model. That is a completeness claim; it proved wrong more than once, and it cannot be maintained against pages that ship new models weekly. Both are now presented as an explicitly partial, dated sample, with what is left out named here and at [\[10\]](#ref-10). No part of the argument rested on the claim. Cited for the vendor's own published prices, never as a recommendation. This page changes frequently — re-read it before relying on any figure above. \[17\] Anthropic, Pricing (Claude Platform documentation) — the "Model pricing" table, USD per million tokens, read September 5, 2026; every figure below was unchanged from the August 18, 2026 reading, and this is the one of the three lists that did not move between them. Figures quoted above: Claude Opus 5, 4.8, 4.7, 4.6 and 4.5 all at $5 in / $25 out, against Opus 4.1, marked retired except on Bedrock and Google Cloud, and Opus 4, marked retired except on Google Cloud, both at $15 / $75; Sonnet 5 at $2 / $10 against Sonnet 4.6, 4.5 and 4 at $3 / $15; Haiku 4.5 at $1 / $5 against the retired Haiku 3.5 at $0.80 / $4; and Claude Fable 5.1 at $10 / $50 as the top of the current family. The page also states that Sonnet 5's introductory $2 / $10 "is now the standard price" and that "the previously scheduled increase to $3/$15 per million input/output tokens on September 1, 2026 will not occur." Cited for the vendor's own published prices, never as a recommendation. \[18\] Jean-Marc Moncalvo, Rytas Vilgalys, Scott A. Redhead, James E. Johnson, Timothy Y. James, M. Catherine Aime, Valérie Hofstetter, Sebastiaan J. W. Verduin, Ellen Larsson, Timothy J. Baroni, R. Greg Thorn, Stig Jacobsson, Heinz Clémençon and Orson K. Miller Jr., One hundred and seventeen clades of euagarics, Molecular Phylogenetics and Evolution, Vol. 23, No. 3 (2002), pp. 357–400\. DOI 10.1016/S1055-7903(02)00027-1\. Full text read at the copy hosted at pilzepilze.de on September 4, 2026\. The abstract states that the recovered clades "cut across traditional lines of classification," that recognising a monophyletic euagarics "results in the exclusion from the clade of several groups of gilled fungi that have been traditionally classified in the Agaricales and necessitates the inclusion of several clavaroid, poroid, secotioid, gasteroid, and reduced forms that were traditionally classified in other basidiomycete orders," and that "though many clades correspond to traditional taxonomic groups, many do not." The introduction states that gilled mushrooms "appear to have evolved multiple times from morphologically diverse ancestors… making the Agaricales polyphyletic," that gasteromycetes "have evolved several times from gilled or poroid ancestors," and that such findings "open the way to deconstruct artificial taxa." The placement of the true puffballs (Lycoperdales) within /agaricaceae, and the note that it "was not previously suspected by morphotaxonomists" although Agaricus "has many biochemical features in common with members of the Lycoperdales," are at clade 83; the Hebelomina placement inside Gymnopilus and the ecological inference drawn from it are at clade 98; the multiple origins of cyphelloid and reduced forms are at clades 5, 25 and 27\. \[19\] Bent Flyvbjerg, Mette K. Skamris Holm and Søren L. Buhl, Underestimating Costs in Public Works Projects: Error or Lie?, Journal of the American Planning Association, Vol. 68, No. 3 (Summer 2002), pp. 279–295\. DOI 10.1080/01944360208976273\. Author accepted manuscript read in full at arXiv:1303.6604 on September 5, 2026\. The sample is 258 transportation infrastructure projects worth approximately $90 billion, completed between 1927 and 1998 across 20 countries. Figures quoted above: costs are underestimated in 9 out of 10 projects (86% likelihood for a randomly selected project); actual costs are on average 28% higher than estimated (sd=39); the thesis that overestimation is as common as underestimation is rejected at p<0.001\. On the seventy-year point the paper states that "cost underestimation has not decreased over the past 70 years" and that "no learning that would improve cost estimate accura\[c\]y seems to take place." The authors reject technical and psychological ("appraisal optimism") explanations on the grounds that honest error would produce a distribution centred on zero and would improve with practice, and conclude that underestimation "cannot be explained by error and seems to be best explained by strategic misrepresentation, i.e., lying." Cited above for the general finding that published, dated cost estimates can be systematically biased, and for the authors' own remedy — "financial, professional, or even criminal penalties for consistent or foreseeable estimation errors" rather than further disclosure — which is the basis for the distinction drawn in the text between provenance and recourse. The authors are explicit that their data cannot decide whether private projects perform better or worse than public ones; no claim above rests on that.