Fourteen sources, none of them invented

Written as if Bidwell had commissioned it. This is the artefact I would put in front of a client before spending any of their money, and the part that matters is the method justification: why an interview and not a survey, and what I would refuse to report.

Primary question

When an ecommerce brand has already let one paid-media agency go, what evidence does the next one have to put in front of them before a call is worth taking?

Secondary questions

Which claim on an agency page is read as a lie first? What does “measurement” mean to a finance director as opposed to a growth lead? At what point does a flat retainer stop feeling like a saving and start feeling like a risk?

Method, and why this one

Semi-structured interviews, 45 minutes, remote. The question is about belief and prior experience, not about task success, so an interview beats a usability test: I need the story of the last agency, and that only comes out in conversation. I would pair it with a 20-minute unmoderated first-click test on the pricing and case-study sections, because “can they find the number” is a task question and interviews are bad at it.

What I would not run

A survey. With a population this small and this variable, a survey would give me percentages of a sample I could not defend, which is exactly the failure mode this brand is positioned against.

Recruit

Twelve participants: eight buyers, four sellers. Buyers screened on UK ecommerce, £1m–£25m turnover, £15k+ a month in paid media, and having ended an agency relationship in the last 24 months. Split four growth/ecommerce leads and four finance-side signatories, because the synthesis suggests they read the same page for different things. Four agency-side participants for the counter-view, recruited outside my own network.

Screener knock-outs

Anyone currently in a pitch process with an agency I have worked with. Anyone whose media is bought by a parent company. Anyone who has never seen an invoice for it.

Sample size reasoning

Eight buyers is where I would expect theme saturation for a question this focused; the four-and-four split means each sub-segment gets four, which is enough to notice a pattern and not enough to claim a rate. I would not report any finding as a percentage at n=12.

Timeline

Week 1 screener and recruit. Weeks 2–3 interviews, six a week, transcribed same day. Week 4 synthesis and first-click test. Week 5 readout, with the raw notes handed over rather than summarised away.

Ethics and consent

Recorded with consent, transcripts pseudonymised, commercial figures aggregated before anything is shown to anyone. Participants get the readout. No client names in the output without written sign-off.

Forty-five minutes, five parts. The craft being demonstrated here is that none of these questions tells the participant what I expect to hear. There is no “how frustrating was the reporting?” and no “would you prefer a flat fee?”, because both hand the answer over. Artefacts come last and unbranded, including mine, so the earlier answers stay clean.

1. 01

   Warm-up

   5 min

   Get them talking in their own vocabulary before I introduce any of mine.

   Questions

   • Tell me what you sell and who buys it.
   • Walk me through who touches the ad account in a normal week.
   • What was the last thing you changed about how you spend?

   Probes

   • What made you change it then rather than earlier?
   • Who else was in that decision?
2. 02

   Context

   10 min

   Reconstruct the last agency relationship end to end, without naming a cause for them.

   Questions

   • Take me back to the last time you brought an agency in. What was happening in the business that month?
   • How did you decide between the ones you spoke to?
   • What did the first ninety days look like?
   • How did it end?

   Probes

   • What did you expect to happen that didn't?
   • Who told you it was going wrong, and how did they know?
   • If you could rewind to the pitch, what would you ask that you didn't?

   Why it is worded like that. Deliberately no “what went wrong”. That question assumes the answer and gets a rehearsed grievance rather than a sequence.

3. 03

   Deep dive

   20 min

   Reporting, measurement and money, in that order, using their words for all three.

   Questions

   • Show me the last report you got, or describe it. What did you do with it?
   • Where do you go to find out how the month actually went?
   • When two systems disagree about a number, what happens next in your business?
   • How was the agency paid, and how did that arrangement come about?
   • Describe a time you were asked to justify the media budget to someone else.

   Probes

   • What did you look at first on that report?
   • Was there anything on it you skipped every month?
   • Who won that disagreement, and what settled it?
   • How did you feel about the fee at month one, and at month twelve?

   Why it is worded like that. “When two systems disagree” is the pivotal question and it is written to be answerable with a story rather than an opinion. It does not mention attribution, Shopify or ROAS, so whichever of those they reach for first is data.

4. 04

   Reaction

   8 min

   Put artefacts in front of them last, so nothing earlier is contaminated by my framing.

   Questions

   • Here are three agency homepages with the names removed. Read them the way you would if you were shopping. Talk as you go.
   • Anything here you'd want to check before you believed it? How would you check it?
   • Here's a page that publishes its prices and a page that doesn't. What does each one tell you about the company?
   • Here's a case study. What's missing from it?

   Probes

   • You paused there. What were you thinking?
   • What would have to be true for that number to be real?
   • Is there anything on any of these you'd want removed?

   Why it is worded like that. Comparative, unbranded, and the artefacts include my own so it can lose. The last question invites criticism explicitly, because “what's missing” gets far more than “what do you think”.

5. 05

   Wrap-up

   2 min

   Catch what the script missed.

   Questions

   • If you were briefing me on this category, what would you tell me that I haven't asked about?
   • Who else in your business should I be talking to?

Seven themes. Prevalence is expressed as a count of sources, never as a count of people, because no people were asked. A theme carried by one source is labelled as carried by one source and has not been promoted on the strength of sounding right.

1. T1Surfaced in 5 of 14 sources

   The reported number and the bank account disagree

   Sources [4] [5] [6] [9] [10]

   This is the loudest thing in the category and it is not framed as a measurement problem, it is framed as a trust problem. Polar Analytics states it as a scene rather than a statistic: “Meta says your ROAS is 4.2x. Google says 6.1x. Shopify says 2.3x. Same week. Same store.” The defaults that cause it are published and boring. Meta counts a 7-day click plus a 1-day view, Google a 30-day click, TikTok a 28-day click, so three platforms can each honestly claim the same order. Two measurement vendors then show what happens when someone actually tests it: platform reporting on branded search overstated by “up to 5X” in one geo holdout, and “roughly 33%” overstatement on a named Performance Max test.

   “Your marketing team celebrates results that your finance team cannot reconcile.”

   ImplicationA performance agency that leads with a ROAS number is speaking the language the buyer has already stopped believing. Leading with the reconciliation method beats leading with the result.

2. T2Surfaced in 4 of 14 sources

   The pricing model pays the agency to do the wrong thing

   Sources [2] [3] [9] [13]

   Percentage-of-spend at 10–20% is still the default in the category, and the criticism of it is being made by agencies themselves. Addictive Digital, who sell PPC management, write that the model breaks because “management hours increase at a much lower rate than spending”. Social Surge put the incentive plainly: it “rewards the agency for spending more, not for performing better”. Long minimum terms compound it, because “an agency on a 12-month lock-in gets paid whether it performs or not”.

   “Hidden or percentage-based pricing rewards the agency for spending more, not for performing better.”

   ImplicationFlat fee and a short notice period are not a nice-to-have here. They are the answer to the objection the buyer is already holding.

3. T3Surfaced in 3 of 14 sources

   The person who pitched is not the person in the account

   Sources [3] [9] [10]

   Stackmatix name it as a first-30-days failure: “Senior strategist disappears after contract signed.” Whitehat turn it into a buyer instruction, telling readers to “check who actually manages your account” and to ask “how many accounts they handle”. That second question is the sharper one, because it is answerable with a number and most agencies will not give it.

   “Check who actually manages your account, and ask how many accounts they handle.”

   ImplicationPublish the ratio before anyone has to ask for it, and name the people.

4. T4Surfaced in 4 of 14 sources

   Nobody publishes a price

   Sources [1] [2] [13] [14]

   Both competitor homepages I fetched put a contact form where a number should be. Brainlabs promise “fully transparent pricing” and “fully loaded rate cards upfront”, but only after you get in touch. Impression Digital publish award badges and client revenue figures but no fee. Meanwhile the benchmark data exists and is easy to find: about £1,040 a month is the UK PPC management average, and full-service retainers run £3,500–£16,750. The information asymmetry is manufactured, and buyers know it.

   “Fully transparent pricing, fully loaded rate cards upfront.”

   ImplicationPublishing four real numbers is the cheapest differentiator available in this category.

5. T5Surfaced in 2 of 14 sources

   Case studies are outcome-only, with no counterfactual

   Sources [13] [14]

   Impression Digital’s homepage carries “24% increase in off-peak revenue”, “£1m in monthly revenue”, “€6.5m revenue”, “600% increase in sales”. Real clients, real numbers, and not one of them says what would have happened anyway. Brainlabs go further the other way and put no figures on the homepage at all. A buyer who has just been told by two measurement vendors that platform numbers overstate by 33% to 5× has no way to read either page.

   “600% increase in sales.”

   ImplicationShow the before state at length, name the test that produced the number, and admit the losses. The write-up is the proof, not the percentage.

6. T6Surfaced in 3 of 14 sources

   Creative is where the upside is, and its size is routinely overstated

   Sources [7] [8] [12]

   The best-evidenced figure I could find is NCSolutions’ analysis of roughly 450 CPG campaigns: creative contributes 49% of incremental sales, against 21% for brand, 14% reach, 11% targeting and 5% recency. The number that circulates in paid-social sales decks is “about 70%”, and chasing it back leads to a 2006 Project Apollo finding of 65% for FMCG sales lift, reported by Nielsen in 2017. Separately, IAB UK’s 2025 figures show social up 21% and video up 20% year on year against search at 6%, so the money is genuinely moving to the formats where creative decides the outcome.

   “In 2006, Project Apollo found that 65% of a brand’s sales lift from advertising came from the creative.”

   ImplicationSell the creative line hard, but never with a borrowed platform statistic. Use your own hit-rate data.

7. T7Surfaced in 1 of 14 sources

   Media relationships are short, and that is normal

   Sources [11]

   ANA and 4As put average client-agency tenure at about seven years, but media-only agencies at 3.7 years against 7.3 for integrated full-service. Single source, US, trade body, no published sample size, so I have not promoted this beyond context. It does explain why lock-ins exist and why a short notice period reads as confidence rather than as weakness.

   ImplicationTreat a short contract as a product feature and say why. Do not pretend churn is unusual.

Insight to opportunity

Three readers, not one

Segments inferred from the sources, not from interviews. They are hypotheses that the plan above is designed to test, and I would expect at least one of them to be wrong.

The useful part. Six things on this site are different because of what I read, and two of them are things I had to take out.

WCAG 2.1 AA, audited against the built page rather than against the class names. Nine findings: two Critical, four Major, three Minor. All six Critical and Major are fixed in the code. Of the three Minors, one is fixed outright, one is improved but knowingly left below the 3:1 line with the reason stated, and one is not fixed at all and has a conforming alternative instead. An audit that closes everything is usually an audit that looked at nothing. Full working is in A11Y.md.

9

Findings raised

6

Critical and Major, every one fixed

7

Fixed outright. 1 partly, 1 answered with an alternative

1. Critical

   1.4.10 Reflow

   Hero number overflowed the viewport at all three breakpoints and scrolled the document sideways. Sized at min(23vw, 320px) against Archivo at width axis 125%, where “£41.6m” measures 4.43em, or 1.02× the content box at any width.

   Sized in container-query units instead: min(21.5cqw, 300px) against a container-type wrapper, so the glyph run is 95% of the content box at every width. html carries overflow-x: clip as a structural backstop.

2. Critical

   1.4.4 Resize text / 1.4.10 Reflow

   The calculator's payoff chart was a fixed viewBox="0 0 600 104" SVG with no minimum width. At 390px it scaled to 0.51, rendering currency figures at about 6.7 CSS px and row labels at about 5.1 CSS px.

   Rebuilt as real DOM: percentage-width bars on a shared track with the figures as live text at 17–19px, so they scale with the user's font settings and never fall below their set size.

3. Major

   4.1.3 Status Messages

   aria-live="polite" with aria-atomic="true" was bound to the calculator's whole output block, which is wired to three range sliders. The budget slider alone has 95 stops, so a single drag queued dozens of ~40-word announcements.

   The output block no longer announces. One compact sentence in a visually-hidden live region is published 600ms after the last change.

4. Major

   1.4.3 Contrast (Minimum)

   Footer legal and column-heading text used white at 50% and 55% opacity on the slate surface #1C1C1A. Measured 4.32:1 and 5.08:1 against the 4.5:1 requirement for 12–13px text.

   Raised to 65% and 70%. Now 7.01:1 and 8.14:1.

5. Major

   2.1.1 Keyboard

   Horizontally scrollable regions, meaning every chart and every before-and-after table, were not reachable by keyboard, so their overflowed content could not be scrolled without a pointer.

   Each scroll container is now tabIndex 0 with role="region" and a name that says what it is and that it scrolls.

6. Major

   1.3.1 Info and Relationships

   The mobile menu numbered its links 01–05 by array position, so three of five numerals disagreed with the numeral printed on the section the link landed on, and “06 Notes” collided with the Pricing section's own 06.

   Each link now carries the numeral of the section it targets. Notes and Research sit outside the numbered run and take the brand mark instead.

7. Minor

   2.5.5 Target Size

   Case-study filter buttons were 40px tall, under the 44px target the rest of the site holds to.

   Raised to 44px with slightly wider padding.

8. Minor

   1.4.11 Non-text Contrast

   Chart gridlines were #D8D8D4 on white, which is 1.43:1.

   Darkened to #C2C2BD (1.85:1). Still below 3:1 and accepted: gridlines are supporting decoration, the axis values are labelled at 7.50:1, and every chart carries a full data table as the conforming alternative.

9. Minor

   1.4.1 Use of Colour

   Chart hover readouts are pointer-only; keyboard and touch users get no per-period figure from the SVG.

   Not fixed in the SVG. The “Show the numbers” disclosure renders the identical series as a real table with row and column headers, which is the equivalent alternative. Recorded rather than hidden.

Contrast, measured

Every pair computed from the sRGB relative-luminance formula, not eyeballed. Large text is 24px, or 18.66px at 700 weight and above.

The whole system, and one component documented properly underneath it.

Component: Button

Fourteen, all fetched and read between 1 and 3 August 2026. Seven are marked because the publisher has a commercial interest in the finding they report; that does not make them wrong, it makes them worth labelling.

1. 01

   UK Marketing Agency Retainer Costs 2026: Real Data

   Whito. Read 2 Aug 2026.Commercial interest in the finding

   Single-channel retainers £1,250–£3,500 a month; full service £3,500–£16,750. PPC management alone typically £500–£1,500 (about £1,040 average) on top of ad spend. Retainer prices up more than 30% since 2023.

2. 02

   How much do PPC agencies charge?

   Addictive Digital. Read 2 Aug 2026.Commercial interest in the finding

   Percentage-of-spend runs at “10–20% of your total ad budget”. Their own objection to it: “management hours increase at a much lower rate than spending”. Cites WebFX 2025 that 83% of businesses spending £1,000–£100,000 a month paid 5–10% of budget in fees.

3. 03

   How to Choose a PPC Agency UK: Pricing, Red Flags & What to Ask

   Whitehat SEO. Read 2 Aug 2026.Commercial interest in the finding

   Ten red flags. Four are about verification rather than price: “Ask for brand vs non-brand reporting”, “Check who actually manages your account”, “ask how many accounts they handle”, “Verify you own the ad accounts”. Also flags invoices that merge ad spend with fees.

4. 04

   Incrementality Testing Examples: Two Surprising Results

   Measured. Read 2 Aug 2026.Commercial interest in the finding

   A premium fashion retailer’s geo holdout on branded search found platform reporting overstated impact “up to 5X”. With brand search off the brand lost “seven orders”, then cut brand search spend “over 84%” while holding “99%+ of orders”; total media spend fell “46% in six months”.

5. 05

   Understanding Google Ads incrementality testing

   Haus. Read 3 Aug 2026.Commercial interest in the finding

   One named, measured result: Caraway’s Performance Max test found “Google’s platform reporting overstated Performance Max impact by roughly 33%”. The article’s other ratios are explicitly illustrative and I did not use them.

6. 06

   Why Your Shopify ROAS Never Matches Across Tools

   Polar Analytics. Read 2 Aug 2026.Commercial interest in the finding

   “Meta says your ROAS is 4.2x. Google says 6.1x. Shopify says 2.3x. Same week. Same store.” Default windows: Meta 7-day click plus 1-day view, Google 30-day click, TikTok 28-day click. A 10–20% variance against Shopify is normal; over 50% means something is broken.

7. 07

   Creative’s Still the Biggest Driver of Sales, Analysis Finds

   MarketingCharts, reporting NCSolutions. Read 3 Aug 2026.

   Across roughly 450 CPG campaigns: creative 49% of contribution to incremental sales, brand 21%, reach 14%, targeting 11%, recency 5%. Media’s combined share has fallen from 36% to 30%.

8. 08

   When it Comes to Advertising Effectiveness, What is Key?

   Nielsen. Read 3 Aug 2026.

   The origin of the recycled creative statistic: “In 2006, Project Apollo found that 65% of a brand’s sales lift from advertising came from the creative.” Base is nearly 500 FMCG campaigns from 2016 to Q1 2017.

9. 09

   Marketing Agency Red Flags: How to Spot a Bad Agency

   Social Surge. Read 2 Aug 2026.Commercial interest in the finding

   Vanity metrics lead the report because “the revenue story isn’t worth telling”. Deliverables written as “Ongoing optimisation” and “Strategic oversight” that “commit the agency to precisely nothing”. On lock-ins: “An agency on a 12-month lock-in gets paid whether it performs or not.”

10. 10

    Warning Signs You Hired the Wrong PPC Agency

    Stackmatix. Read 3 Aug 2026.Commercial interest in the finding

    The post-signature failure mode: “Senior strategist disappears after contract signed”; “Campaigns never restructured from your previous setup”; “Account changes happening without your knowledge”; no evidence of creative testing over three months or more.

11. 11

    Client-Agency Relationship Tenure Has Doubled Since 2016

    ANA and 4As, April 2025. Read 3 Aug 2026.

    Average client-agency tenure is about seven years, but media-only agencies average 3.7 years (44 months), against 7.3 years for integrated full-service and ten years for experiential. No sample size published; US trade body.

12. 12

    Digital Adspend 2025: UK’s digital ad market reaches £40.5bn

    IAB UK with Oliver Wyman. Read 3 Aug 2026.

    £40.5bn, up 10% against UK GDP growth of 1.4%. Search 44% of spend (£17.9bn, up 6%); social 28% (£11.5bn, up 21%); video 23% (£9.3bn, up 20%); retail media £3.8bn, up 18%.

13. 13

    Homepage (competitor observation)

    Brainlabs. Read 2 Aug 2026.

    “What’s your next best move to maximize revenue through media?” Positions on profit rather than platform metrics. No pricing published, only a promise of “fully loaded rate cards upfront” after you make contact. Enterprise logo wall, three role-titled testimonials, no homepage figures.

14. 14

    Homepage (competitor observation)

    Impression Digital. Read 3 Aug 2026.

    “Performance Marketing Agency of the Year”. No pricing published. Case studies do carry numbers: “24% increase in off-peak revenue”, “£1m in monthly revenue”, “€6.5m revenue”, “600% increase in sales”. Every one is outcome-only, with no counterfactual stated.