bol is not a platform in trouble. It is a good platform with four specific, measurable holes, and every complaint I could find is a symptom of one of them.

If you read one thing, read the yellow boxes. That is the failure as a person on the platform actually meets it.

Ordered by damage done: 01 is the one to fix first

1. 01

   Built for retrieval, not for discovery

   What happens

   bol’s on-site search is genuinely world class: a Baymard UX award and a top-ten placing, scoring 91.8 on desktop and 92.0 on mobile. Homepage and category navigation, the surface people use when they do not yet know the words, scores 0.4 on desktop and 30.8 on mobile. That is the widest internal gap in the entire benchmark.

   What the user actually experiences

   You can find a named product in seconds. Arrive wanting “something for a six-year-old’s birthday” and the site has almost nothing to offer you, so you go back to the search box and start guessing words.

   EVIDENCE   Search 91.8 desktop / 92.0 mobile  ·  Category navigation 0.4 desktop / 30.8 mobile  ·  26 guidelines per surface FIX → Treat category navigation as a first-class surface with its own quality bar, not as the fallback for a failed search.
2. 02

   Mobile checkout is the weakest thing bol operates

   What happens

   Mobile cart and checkout scores 6.5 out of 100 across 88 guidelines. Desktop scores 21.9 across another 88. Together that is 176 of the 629 guidelines in the benchmark, roughly 28% of everything measured, sitting on the revenue path and failing.

   What the user actually experiences

   Everything up to the decision to buy feels competent. The moment you commit is the moment the experience is at its worst, on the device most people are holding.

   EVIDENCE   Mobile checkout 6.5  ·  desktop 21.9  ·  176 of 629 guidelines live in cart and checkout FIX → Mobile checkout first. Largest failing pool in the benchmark, and the shortest line between a fix and a number that moves.
3. 03

   Forty-eight thousand partners, one brand, no visible seam

   What happens

   bol carries the reputational risk for fulfilment decisions made by 48,000 independent partners, and the interface does not make that boundary legible before purchase. Reviewers report wrong items, counterfeit goods, unresponsive sellers, and dropshipped stock at several times its origin price.

   What the user actually experiences

   You bought from bol. What arrived is not what you ordered, and the answer is that it came from a partner. You did not choose a partner. You chose bol.

   EVIDENCE   1,387 returns and refunds complaints  ·  1,180 customer-service complaints (klacht.nl, 5,755 all time)  ·  Play Store reviews citing AliExpress stock at “8x the price” FIX → Make seller identity, dispatch origin and the dispute route legible at the point of choice, not after the parcel arrives.
4. 04

   An accessibility plan aimed below the legal bar

   What happens

   bol publishes an accessibility statement with a named external auditor, which is more than most Dutch retailers manage. It states partial conformance and discloses the failures honestly. The plan targets roughly 70% of WCAG Level A issues by 2026, with video in 2027. Level A is the floor; the European Accessibility Act and EN 301 549 expect AA.

   What the user actually experiences

   bol’s own list of what is broken reads like a description of the transaction: unclear page titles, headings and links, input fields without names, error messages that display poorly, and text that truncates when you enlarge it.

   EVIDENCE   bol accessibility statement  ·  audited by Cardan, site Sept 2024, apps Nov 2024, WCAG-EM  ·  target ~70% of Level A by 2026 FIX → Re-baseline against AA rather than A, and land the fixes in the design system so they ship once and apply everywhere.

Three scoreboards, three different answers

bol is rated very differently depending on when you ask. That spread is the most useful diagnostic available, because each measure is capturing a different moment in the relationship.

The Trustpilot split is the tell: 45% five-star and 33% one-star, with almost nobody in between. bol is not mediocre for everyone. It is excellent for most people and badly broken for a third of them, and the difference is which failure path you happen to land on.

The complaints

Nineteen, bucketed by severity and collapsed by default. Open a bucket, or filter to one.

What it translates to

Eight themes. The perception column is the one that produces the one-star third of the Trustpilot distribution.

How that compares

bol is second in its own market, and the gap in front of it is instructive. Coolblue is rated on the same axis, at national scale, and beats bol by 1.3 stars while carrying three times the review volume. What Coolblue is praised for is precisely what bol is criticised for: service reachability and filtering.

The read: bol’s competitive problem is not Amazon. It is Coolblue, and the axis is accountability rather than assortment. A marketplace with 48,000 partners cannot copy Coolblue’s model, but it can copy the thing customers actually respond to: knowing who is answerable, before they buy.

Business model innovation

bol is two businesses wearing one interface: a first-party retailer and a marketplace of 48,000 partners, serving 14 million customers. Almost every complaint in path 01 sits on the seam between them, and the interface currently hides that seam rather than designing it.

The ten types of innovation

Doblin’s framework argues that winning innovations combine three to five reinforcing types, and that most organisations over-invest in product performance alone. Mapped to bol, the neglected types are exactly where the complaints live.

The read: bol has innovated hard on Configuration and thinly on Experience. All four Experience types are underperforming or failing, and that cluster is where every complaint in path 01 lives. A combination play across Service + Brand + Customer engagement is three reinforcing types, which is exactly what the framework says wins.

Integrative thinking

Roger Martin describes this as facing the tension between opposing ideas constructively and, rather than choosing one, generating a resolution that contains elements of both and is superior to either. This role is built on that skill: the mandate is real, the authority is not, and the two have to be held together. Three tensions worth resolving rather than picking sides on.

Design thinking, and where the squiggle actually goes

The squiggle running through this page is the argument: uncertainty is front-loaded, and clarity is earned by narrowing. The temptation at bol is to skip the divergent half, because the Baymard report already tells you what to fix. For two of the four root causes that is correct. For the other two it would waste the year.

The gaps worth attacking

Every large retailer draws the same generic complaints. The strategic question is which are worse at bol than at the alternatives, or uniquely available to it. Five hold up.

1. 01

   A 92 and a 0.4 in the same product

   Almost no competitor has that spread. It is not a capability problem, because the search team has proved the capability exists in the building. It is a question of where attention has been pointed.

2. 02

   The checkout that carries the revenue is the worst surface measured

   6.5 on mobile, across the largest guideline pool in the benchmark. Nothing else on this list has a shorter path from fix to measurable return.

3. 03

   Nobody has designed the marketplace seam

   Amazon hides it, Zalando hides it, bol hides it. The first Dutch marketplace to make accountability visible before purchase takes Coolblue’s strongest argument away from it, at a scale Coolblue cannot reach.

4. 04

   An accessibility programme with real foundations and a short-aimed target

   The awareness work, the design system focus, the external audit and the published statement are all in place. The plan simply aims at Level A. Re-pointing an existing programme is far cheaper than starting one.

5. 05

   The craft question nobody has answered yet

   What product design becomes when AI writes the artefacts is unanswered across the whole industry. bol has an explicit mandate to answer it. That is a rare thing to be handed, and it is the reason this role is interesting.

Where the points are lost

Accessibility

This is the part of the analysis I can speak to with the most direct experience, and it is also where bol has done more than most of its peers. The statement below is bol’s own, published, with a named external auditor. I am reading it as a practitioner, not catching anyone out.

The fix list, in the order the data supports

1. 01

   Rebuild mobile checkout

   6.5 across 88 guidelines, on the highest-traffic surface. Rebuild against the guideline set, with accessible form labelling and error handling designed in rather than retrofitted.

   Closes the largest failing pool in the benchmark and a disclosed accessibility failure in one pass.
2. 02

   Give category navigation a quality bar of its own

   0.4 on desktop. A real taxonomy with visible hierarchy, persistent filters, and headings and links named clearly enough to be usable by screen reader and by browser alike.

   Also closes bol’s disclosed “unclear page titles, headings and links” finding.
3. 03

   Re-baseline accessibility to AA

   Re-scope the existing programme from ~70% of Level A to full Level AA, prioritised by transaction impact rather than by issue count, and fixed in design-system components so each fix ships once.

   The programme, the auditor and the awareness work already exist. This is re-pointing, not starting.
4. 04

   Bring customer accounts up to the site average

   20.5 mobile, 37.3 desktop. Order and dispute history with visible status and deadlines, a clear route to a human, and returns that can be tracked without contacting anyone.

   The surface through which every other fix reaches a customer who has a problem.
5. 05

   Make the product page carry the seam

   163 guidelines at middling. Seller identity, dispatch origin, fulfilment rating and dispute route shown before purchase, alongside the structured attributes that make comparison possible.

   The cheapest place to fix root cause 03, because the decision is made here.
6. 06

   Port desktop’s filtering lessons, and mobile’s

   Mobile filtering scores 68.5 against desktop’s 45.6, while desktop search scores 91.8. Each surface has solved something the other has not, inside the same company.

   Highest points-per-effort on the list: the work already exists, on the wrong platform.

The benchmark and bol’s own accessibility audit point at the same rooms from completely different directions: checkout, forms, navigation labelling. When a commercial measure and a legal obligation agree on the priority, the sequencing argument makes itself.

The constraint is the design problem. Diagnose before you standardise, because a bar imposed on a craft you have not read gets complied with rather than adopted, and compliance is worthless in a discipline that runs on judgement.

Reading the shapes

Skill shape is a useful shorthand, but only mapped honestly at three levels: craft (interaction, visual, systems, prototyping, research, accessibility, writing), collaboration (how they work with engineering and product), and influence (mentoring, direction-setting, stakeholder work). Someone can be deep in craft and flat in influence. That is a placement question, not a deficiency. AI is changing the shape of the bar, not the stem: it widens what one person can credibly cover, and it makes depth rarer and more valuable.

Where the tangles are, and how to iron them out

Seven tangles account for most of what slows a design craft down at scale. Each has a visible signal, a matched intervention, and something measurable. The common failure is applying one remedy, usually process, to all seven.

Four ongoing pipelines

Craft development fails when it is an event. These are loops with a cadence, a defined input and output, and a stated return, so each strengthens an individual and pays the organisation back. Deliberately small, because a pipeline costing a day a week gets cancelled the first time delivery gets tight.

How I would know it is working

Five measures, chosen because they are hard to game and visible to the craft as well as to the management team. In a role without direct authority, published measures are the mechanism: they let people align with evidence rather than with a person.

1. 01

   Adoption without instruction

   Whether domains take up the direction because it is useful, or because they were asked. The only honest test of influence in a role with no reporting line.

2. 02

   Coverage

   No capability sits with fewer than two confident owners. The best early indicator of a resilient craft, and the easiest to lose quietly.

3. 03

   Measured quality, not asserted quality

   A baseline anyone can run, moving in the right direction. Baymard’s 629 guidelines and bol’s accessibility audit are two ready-made instruments for exactly this.

4. 04

   Specific opinions about AI

   Whether designers can say precisely where a tool helps their work and where it does not. Vague enthusiasm and vague resistance are the same failure.

5. 05

   Where the craft looks for answers

   Whether people go to the system, the documentation and each other, or to one person. A healthy craft distributes its own expertise, including away from me.

Sources

1. Trustpilot: bol.com
2. Google Play: bol app
3. klacht.nl: complaints register for bol.com
4. bol: accessibility statement (toegankelijkheidsverklaring)
5. Cardan: digital accessibility and the EAA, in conversation with bol
6. Baymard Institute: UX benchmark
7. Trustpilot: Coolblue
8. Trustpilot: Amazon.nl
9. Trustpilot: Zalando
10. Test-Aankoop: bol.com complaints
11. Storing Vandaag: bol.com outage tracker
12. Doblin: Ten Types of Innovation
13. Roger Martin: The Opposable Mind (integrative thinking)
14. Google Design: Material 3 Expressive research

How I would allocate the capacity

Commitments only mean something if they add up. This is the whole hundred per cent for a craft-leadership role, and what happens to it when the organisation is mid-transition, which bol says it is.

1. 01

   Define what design at bol needs to become, and own the roadmap

   I would start by reading the craft rather than declaring a direction: sit with Staff Designers and PDMs across domains, observe real delivery, review artefacts rather than opinions, and publish the picture back before proposing anything. A direction people recognise themselves in gets adopted; one that arrives as a framework gets tolerated. Path 04 sets out the full three-phase plan.

   COMMITMENT → 40% of capacity on craft direction and the quality bar that makes it measurable.
2. 02

   Answer the AI question with evidence from our own product

   The role names the conundrum precisely: AI-generated artefacts are fast and generic, and generic is the opposite of design. I would run the Frontier pipeline monthly, timeboxed experiments on real bol problems with a written verdict on what held, what produced mush, what it cost, and whether it becomes normal practice. My working position is that AI should own the artefact and the craft should reclaim what it was never given time for: framing the problem, judging quality, and knowing what should not be built.

   COMMITMENT → 25% of capacity, and a published verdict every month rather than a point of view.
3. 03

   Keep design and research coupled, whatever the org chart says

   I have spent six years producing research that had to survive contact with product owners, developers and business analysts who did not commission it. What makes research stick is not proximity in the org chart, it is whether the finding arrives with a design that makes it actionable. Every Baymard report I write ends in mockups for that reason, and I would hold the same standard here.

4. 04

   Represent design in cross-craft governance, and mean it

   Operating alongside the Heads of Craft for Engineering, Product and Product Analytics on tooling, team composition and ways of working requires being able to move between altitudes in a single conversation. I can have a substantive discussion with a Staff Designer about interaction quality and walk into the next meeting to argue about builder profiles, because consulting gave me no other option: the deep dive and the executive readout were the same two hours.

5. 05

   Turn direction into standards, tooling and enablement without bureaucracy

   The quality bar I would bring is a measurement, not a policy: a baseline anyone can run against their own domain. bol already has two ready-made instruments for it, the Baymard guideline set and its own accessibility audit. And accessibility is where I would prove the model first, because it is the one area where a legal obligation and a commercial metric point at the same components. I have audited eCommerce against WCAG 2.1 and 2.2 for three years across EU and US markets, and I hold the Web Accessibility Specialist certification.

   SEE → Path 03 is this thinking applied: 629 guidelines and bol’s own audit, ranked into six fixes.

Some things I like