A single pound figure for what a business is losing is usually a sum of things that cannot be added

Five kinds of thing and one line of arithmetic

The demo workspace on HighRegard's site shows an opportunity feed running on synthetic data for an imaginary clinic. Across a 28-day window it lists £352.07 of recoverable paid spend, 24 potential conversions, 269 potential organic clicks, one data or tracking risk and eight quality improvements ^[18].

An earlier version of that feed added those figures together and printed the sum at the top of the page. It looked decisive. It was also meaningless, because a pound, a conversion, a click and a warning are not the same kind of thing, and a total of them has no unit. The headline was removed and the page now says so plainly: "Kinds are not added together: pounds, conversions and clicks are different things" ^[18].

The same arithmetic runs every day in audit emails and agency proposals: a tool crawls a website or an account, attaches a value to each thing it finds, sums the values and announces what the owner is losing each year. The pitch depends on the sum being read as a measurement.

I paid for marketing before I sold any. I built my clinic's website, Google Business Profile, analytics and tracking myself, with no agency, while working as a pharmacist, and I ran Google Ads for it on a small daily budget and read the click and conversion data myself. I have been on the receiving end of the one-number pitch, and I have seen from the inside how little of what sits behind such a number is actually money.

What each of the four kinds actually is

A saving is money, and it can be measured

Spend on search terms that never converted is a bounded, observed amount in the account's own reports. Turning it into a saving needs one assumption, which is how much of that spend can be stopped without losing conversions from nearby terms. That assumption can be written down and argued about, but the underlying figure is a measurement.

A conversion estimate is a probability multiplied by an assumed value

An extra conversion is a forecast, and a pound figure for it is a forecast multiplied by a number somebody chose. Google's documentation is candid about where the number comes from: conversion values let advertisers "learn the total business value that Google Ads generated, rather than simply the number of conversions that have happened", and using one value for every conversion "may be less representative of different values of each transaction" ^[9]. For the bidding strategy that maximises conversion value, the advertiser must first "set up conversion tracking with transaction-specific values" ^[10]. The value is an input supplied by the business, not an output discovered by the platform, and any pound figure for extra conversions inherits it.

A click is a click

Search Console defines a click as "how often someone clicked a link from Google to your site" ^[13]. It is a count of visits, not of customers. Converting clicks to money needs a conversion rate the page may not yet have and an assumed value per conversion. Attribution adds another layer: Google's models "let you choose how much credit each ad interaction gets for your conversions" ^[12]. A click count turned into pounds through a chosen attribution rule and an assumed value is two assumptions deep before it reaches the total.

A risk is a possible loss, not a gain

A broken conversion tag, a suspended profile or a tracking gap is not an opportunity to earn anything. It is a possible loss, and often a distortion of every other figure on the list. Google records a conversion when its tag recognises the defined action after an ad click ^[11]. Break the tag and the conversions do not stop; the record of them does.

I have seen how a broken conversion tag makes a working campaign look like a failing one. Adding a risk to a total as if it were a gain is a category error, and giving it a nominal pound value so that a ranking formula will work is worse, because the total then includes a number invented for a different purpose.

Why the sum fails even when every line is honest

A single figure could still be defended if the items were independent and commensurable. They are usually neither. Two recommendations about the same query cannot both deliver their full estimate, because the same searches cannot be won twice. And if the tracking is broken, every paid estimate on the list rests on a corrupted record, so the honest state of those items is blocked until the tracking is fixed, not summed alongside it.

Government appraisal guidance has worked through this problem for public projects. HM Treasury's Green Book tells practitioners to state their assumptions "to make sure that these are transparent and that they are agreed by decision makers", to run sensitivity analysis on how changes in key assumptions move the summary metrics, and to calculate switching values, "the values that a key assumption would need to change to, in order to make an option no longer value for money" ^[14].

It also names the bias that single-figure totals feed on. Appraisers "should explicitly adjust their appraisals for optimism bias. This is the proven tendency for appraisals to be over-optimistic about key assumptions", and the adjustment runs one way: costs and durations up, benefits down ^[14]. An opportunity total built from best-case conversion rates and full recovery of every wasted pound is an appraisal with the optimism left in.

What the advertising rules say about a figure in a pitch

The UK's non-broadcast advertising code is explicit about figures. Rule 3.1 says marketing communications "must not materially mislead or be likely to do so"; rule 3.3 says they "must not omit material information"; rule 3.9 says they "must not mislead by omitting significant limitations and qualifications" ^[1].

Rule 3.7 is the one a one-number pitch most obviously strains. Before publishing, marketers "must hold documentary evidence to prove claims that consumers are likely to regard as objective and that are capable of objective substantiation" ^[1]. CAP's guidance adds that "objective claims must be supported by objective evidence", lists claims "about the benefits or effects of using a product or service" among those needing it, and applies the test of what consumers take the claim to mean, not what the advertiser intended ^[2].

On savings claims, CAP says "the basis of any savings claims should be clear and accurate" ^[3].

More recent ASA guidance is closer to the marketing case. Reviewing rulings on household energy savings claims in April 2026, the ASA noted that the claims "were based on specific sets of data, some of which were modelled, which would be considered material information that needs to be made clear in the ad", and that an "up to" figure needed evidence "that a significant proportion of customers can actually achieve the maximum saving" ^[4]. Read across, a modelled figure for what a business is losing is material information about its own basis, and an "up to" gain needs evidence that the maximum is commonly reached.

Whether any of this applies when the buyer is a business

Partly, and the limits matter. The CAP Code covers advertisements in emails and paid online space and marketing on a company's own website, and defines a consumer as "anyone who is likely to see a given marketing communication, whether in the course of business or not" ^[5]. But the Code does not apply to "live oral communications, including telephone calls" or to "private correspondence" ^[5]. A number spoken in a sales call, or typed into a bespoke proposal, is largely outside the ASA's reach.

Business-to-business advertising has its own law. The Business Protection from Misleading Marketing Regulations 2008 prohibit advertising that "deceives or is likely to deceive the traders to whom it is addressed" and affects their economic behaviour, and among the matters to be weighed are the "results to be expected from use" of the product ^[6]. A projected return is a result to be expected from use.

The consumer law that has drawn most attention lately is narrower than often assumed. The Competition and Markets Authority's guidance on the unfair commercial practices provisions of the Digital Markets, Competition and Consumers Act 2024 says a trader selling "to businesses only" does not need to consider them ^[7]. It still describes the fault well: a misleading action includes information that "is not actually false, but is arranged, displayed or otherwise presented in a way that is likely to deceive" ^[7]. A total of true lines that reads as a measurement is exactly that shape of problem.

What Google says about promises made on its behalf

Google's hiring advice for search optimisation firms is short and blunt. "No one can guarantee a #1 ranking on Google", it says, and it warns readers to "be careful if a company is secretive or won't clearly explain what they intend to do" ^[8]. Its suggested questions for a supplier include "What kind of results do you expect to see, and in what timeframe? How do you measure your success?" ^[8].

Neither question is answered by a total. A figure for what you are losing says nothing about what will be done, over what period, or how the result will be measured afterwards.

What the research says about measuring returns at all

Even where the intention is honest, the confidence implied by a single figure is hard to earn. Randall Lewis and Justin Rao analysed 25 large field experiments with major US retailers and brokerages, collectively representing $2.8 million in advertising expenditure, and found that "the median confidence interval on ROI is over 100% wide, the smallest exceeds 50%" ^[15]. Reliably measuring an effect "can easily require more than ten million person-weeks", and they call selection bias "a crippling concern for widely-employed observational methods" ^[15].

An experiment at eBay reached a similar conclusion from the other direction. Tom Blake, Chris Nosko and Steven Tadelis found that "brand-keyword ads have no measurable short-term benefits" and that "returns from paid search are a fraction of conventional non-experimental estimates" ^[16].

Both studies concern very large advertisers and neither is about a clinic in Bradford. If firms with millions of customers and randomised tests cannot pin a return to within a factor of two, a tool inferring one for a small account from a few weeks of observational data cannot honestly print a point estimate without a range. Sarah Warby, a senior marketer who had worked at Heineken and Sainsbury's, told Marketing Week in 2019: "you often can't pull the thread all the way through to an ROI and you shouldn't try because that's the point at which you look like an idiot when you are talking to anyone who is a chartered accountant" ^[17].

The case for one number

The critics of the single figure should concede what it gets right. Business decisions are frequently binary. An owner will sign the proposal or not, and a list of five kinds of thing with a note on each does not by itself say which. Asking for a total is a request for a decision rule, not a failure of numeracy.

Nor is there anything dishonest about summarising a well-labelled range. The Green Book, for all its insistence on stated assumptions, still produces summary metrics; its instruction is to test how they move and report the switching values, not to refuse to summarise ^[14]. A report that says the recoverable spend is measured, that the conversion figure rests on a stated value per lead, and that the range across plausible values runs from a low case to a high case, can end with a sentence saying what the author would do.

The distinction is between a total that hides its parts and a summary that shows them. And the reader remains free to add what they choose. If a clinic owner decides what a lead is worth to them and adds the saving to the valued conversions, that is their arithmetic on their assumptions, which is the only kind that should ever produce one number.

The honest alternative, and how the HighRegard feed applies it

The alternative is to keep the kinds separate, print the assumption behind each figure, label how each estimate was reached and state how confident it is. HighRegard's opportunity feed is one implementation.

Every recommendation carries one of five kinds: a saving, conversions, organic clicks, a risk or a quality improvement. Each kind is totalled on its own and never across kinds, so the top of the page reads as recoverable paid spend, potential conversions, potential organic clicks, a count of risks and a count of quality improvements ^[18]. Items are ranked by estimated impact multiplied by confidence and divided by effort, and the page states that "impacts are estimates for the 28-day window and are labelled as such" ^[18].

Each estimate carries a class describing how it was reached: measured, meaning observed in the account's own data; empirical, meaning calculated from the account's history; heuristic, meaning taken from a stated rule; or scenario, meaning conditional on something that has not yet happened ^[18]. Where a conversion estimate is expressed in pounds, the assumption is printed beside it: the account's actual cost per conversion where one exists, or a stated £45 per lead assumption where it does not, labelled as an assumption until a booking system supplies a real value ^[18]. Inside each recommendation, what was measured is labelled observed, what was projected is labelled estimated, and what is advised is labelled judgement, so a reader can reject the judgement without disputing the observation.

Two further rules follow. Estimates below a materiality threshold are dropped before they reach the list, because a line that promises almost nothing is noise dressed as advice. And where the tracking is broken, the paid items that depend on it are shown as blocked rather than counted.

The written assessment on the overview page is generated by a language model that is never allowed to type a number. An earlier version that formatted figures itself wrote a four-figure spend as if it were millions; the model now refers to figures by name and the application inserts them afterwards, rejecting any output with a number not in its table.

The site's composite score follows the same logic. Its methodology page states that a module with no data "is excluded from the mean and shown as 'not measured'", that the weights "are opinions, held consistently and published so they can be disagreed with", and that the score "does not count lost customers, forecast revenue, or promise that a higher score produces" a better outcome ^[19]. It is labelled a draft ^[19].

What this means for a local business

A clinic, a dental practice, a solicitor or a plumber reading a report with one number at the top can test it in a few minutes without technical knowledge.

1. Ask what kinds of thing were added together. If savings, projected conversions, clicks and risks sit in one sum, the total has no unit and should be set aside in favour of its parts.
2. Ask which assumption turned each item into pounds, and whether it is written down. Google's own documentation treats a conversion value as something the advertiser supplies ^[9].
3. Ask which items depend on another being true first. A projection of conversions on the same page that flags the conversion tag as broken rests on a record the report itself says is wrong ^[11].
4. Ask how each figure was reached: measured, calculated from history, taken from a rule, or conditional on a change.
5. Ask the supplier Google's questions: what results they expect, in what timeframe, and how success will be measured ^[8].
6. Ask whether the claim would survive the advertising code. An objective claim in an email or on a website must be backed by documentary evidence held before publication ^[1].

A good agency will have the parts of its total to hand and will not mind being asked for them. The one that cannot produce them has told you something useful about how the total was built.

What the evidence does not settle

It is not established that separated, labelled estimates produce better decisions than a single total. That is a reasonable inference from the appraisal guidance, but no study fetched for this piece tested it on small-business buyers of marketing, and owners faced with five labelled figures may simply add them up in their heads and lose the labels.

The advertising rules are clear on substantiation and on the basis of savings claims, but the ASA guidance on modelled savings was written about household energy and the CAP guidance on savings claims about retail promotions ^{[3] [4]}. Applying them to a projected marketing return is an argument from principle, not a ruling. And much of the market for these numbers happens in sales calls and bespoke proposals, which the Code does not cover ^[5].

The research on measuring returns comes from very large US advertisers ^{[15] [16]}. It shows how wide the uncertainty is even under ideal conditions; it does not tell a small business what its own uncertainty is. HighRegard's feed is described here as it runs on synthetic data ^[18], and the composite score is a draft not yet shown on any live customer workspace ^[19]. Whether the method holds up against real accounts is something the site will have to show rather than assert.

What remains uncertain

The argument against adding kinds together is a matter of arithmetic and does not depend on any study. What remains open is whether the alternative is used well. A labelled range is only as honest as its labels, and a heuristic printed as a heuristic can still be a poor heuristic.

The safest position for a reader is the one the regulators and the appraisal guidance converge on without quite saying it: treat every pound figure as a claim with a basis, ask to see the basis, and do the adding yourself. Whether the industry moves that way, or whether the single number survives because it closes deals, is not something the evidence can yet say.