Original research · 8 September 2026

What actually goes wrong with inventory and ERP software

We collected 2,324 public reviews of 19 systems that UK product businesses actually shortlist, then counted what the unhappy ones say. Three findings held up. One did not, and we have written that down too.

By Martin Gjini · Method, dataset and limitations are on this page.

  • 2,324reviews analysed
  • 19systems covered
  • 840complaints coded
  • 1,098from the UK
Finding 1

Praise is solicited. Complaints are volunteered.

Trustpilot records whether each review was posted unprompted or collected through an invitation the company sent out. Split every review by that flag and a straight line appears.

Share of reviews the vendor asked for, by rating. n = 2,324.
RatingReviewsRequested by the vendor
1 star 727
8.5%
2 star 113
15.4%
3 star 81
35.5%
4 star 331
45.2%
5 star 1,072
47.2%

A five star review is roughly six times more likely to have been requested by the vendor than a one star review.

The obvious objection is timing. Invitation programmes are a recent habit, and one star reviews go back years, so the gradient could be an artefact rather than a pattern. So we ran it again, per vendor, restricted to a single shared time window.

Same window, same vendor. Share of reviews the vendor asked for.
SystemWindow from5 star1 to 2 star
Orderwise Sep 2024 93% 0%
Odoo Nov 2025 82% 1.1%
Linnworks Jun 2022 72% 25.5%
Unleashed Mar 2026 42% 0%
Cin7 Core Mar 2025 38% 0%
Selro Sep 2017 35.9% 4.5%
Cin7 Omni Mar 2025 35% 0%
Veeqo Jul 2021 35% 15.5%

Every vendor. Same window. No exceptions.

What that means if you are buying

The score on a vendor's website is substantially a measure of how hard that vendor asks. The part nobody prompted, the signal that was volunteered, sits almost entirely in the complaints. Read those first, and read them before the demo rather than after.

Finding 2

People leave over the relationship, not the software

Every one and two star review was coded into complaint categories using a rule set published in full below. Grouping those categories into complaints about the product and complaints about the company gives the result that surprised us most.

75.0%

mention the relationship: support, commercial terms, account management, or the gap between what sales promised and what arrived

53.9%

mention the product: bugs, missing capability, integrations, or implementation

31.7%

never criticise the software at all. Their complaint is entirely about the company selling it

Share of the 840 negative reviews mentioning each theme. Categories overlap, so they do not sum to 100.
ComplaintAllUK only
Support quality or responsiveness 50.1% 50.0%
Functionality gaps 28.8% 33.8%
Reliability and bugs 21.2% 23.1%
What sales promised vs what arrived 18.8% 20.5%
Integrations 18.7% 23.1%
Price increases 16.7% 22.3%
Contract and cancellation terms 16.0% 13.0%
Implementation and onboarding 13.7% 12.4%
Value for money 9.9% 11.8%
Account management 7.1% 5.2%

relationship    product

Combined, the commercial complaints (price rises, value, contract terms) reach 37.1% of all negative reviews and 41.0% of the UK ones. British reviewers raise price increases noticeably more than the global average: 22.3% against 16.7%.

In their own words

Verbatim, attributed and dated. All from UK reviewers.

Our price went up from £12000 a year to £79000. The service and features have not improved.
Trustpilot review of Linnworks · Oct 2022
Increased the price from £2160.00 p/y to £18,000 p/y with no actual improvement of the service.
Trustpilot review of Linnworks · Nov 2022
Linnworks tried to increase prices by 400% within 30 days notice. The way they treat old customers is appalling.
Trustpilot review of Linnworks · Jun 2024
Cannot raise a support ticket anymore without paying for a certain plan. This software used to be brilliant but after a flurry of recent changes they have broadly ruined it.
Trustpilot review of Unleashed · Jun 2026
System lacks lots of basic functionality. Anything that does exist, is an add-on module at extra cost.
Trustpilot review of Mintsoft · May 2026
I was sold this as a one stop shop but it is far from it.
Trustpilot review of Unleashed · Mar 2021
Finding 3

The loyalty penalty

Negative reviews that mention using the system for several years are far more likely to be about money and contract terms than about the product.

Reviews citing multi year use (n = 133)
57.9%
Everyone else (n = 707)
33.2%

Share whose complaint is about price, value or contract terms. Chi square 29.1, 1 degree of freedom, p < 0.001.

Long term customers are about 1.7 times more likely to be complaining about pricing or contract terms than newer ones. The pattern in the text is consistent and it is not subtle: the system worked, the business grew, and the price followed the growth.

What that means if you are buying

The price you are quoted is the price for the business you are today. The reviews suggest the thing worth asking about before you sign is not this year's number but what happens to it when you add users, orders, channels or warehouses, and what leaving looks like if the answer stops working for you.

What did not hold up

Recorded because a teardown that only confirms its author is not worth citing. Three things we expected to find, and did not.

  1. Price increase complaints are not measurably rising

    As a share of each year's negative reviews they move between 7% and 48% with no trend, and the earliest years have small samples. Absolute counts are broadly flat. Any claim that this is getting worse would not survive the data.

  2. Vendors do not ignore complaints

    Vendors replied to 52.5% of negative reviews and 50.6% of positive ones. There is no evidence here that complaints get less attention than praise.

  3. Score alone tells you very little

    Review counts across these 19 systems differ by three orders of magnitude. A 1.5 average across 50 reviews and a 4.7 average across 1,276 are not comparable numbers. Small samples are reported here but carry no weight, and nothing on this page ranks these products.

Method

Written out so the result can be checked and repeated rather than taken on trust.

Source and collection

Public Trustpilot reviews, collected on 8 September 2026. Each review page carries a structured data payload with the verbatim text, rating, date, reviewer country, language, vendor reply and the review's origin label, so nothing was retyped or summarised by hand.

Two constraints we hit, and how they were handled

Trustpilot caps public paging at 200 reviews per filtered view, so the collection was sliced by star rating to keep every slice under the cap. Separately, the first page of any filtered view silently ignores the filter and returns the unfiltered list. An early run that missed this contaminated every slice and was discarded. The collector now checks that the filter was applied and throws away any page where it was not.

Coverage

A complete census of every 1, 2 and 3 star review for all 19 systems. The 4 and 5 star pulls are capped at 100 per system, because they exist only to support the solicitation comparison. Exact rating distributions are taken from Trustpilot's own published statistics, so the percentages stay true even where our own pull was capped. Every capped slice is recorded in the dataset.

Solicitation

Taken from Trustpilot's own verification label rather than inferred. Where that field is empty the review's origin is used instead, because several Trustpilot invitation methods leave it unset. Reviews whose origin is ambiguous are left unclassified rather than guessed. 98.1% of reviews are classified.

Complaint coding

Rule based and published in full. Each review is split into clauses, and a category is only applied when its topic and a complaint signal appear within 60 characters of each other in the same clause. That constraint exists because a later grumble should not attach itself to something the reviewer was praising: "great onboarding, but no support" is a support complaint, not an onboarding one.

How accurate the coding is

38 assignments were sampled across all ten categories and judged by reading the exact clause that triggered each one: 34 correct, 3 wrong, 1 borderline, so roughly 89% precision. Read the category percentages with that error in mind. The two headline findings do not rely on this coding at all, since they come from Trustpilot's own labels and counts.

Limitations

  • One source. Trustpilot only. G2 and Capterra block automated access and Reddit was unreachable. Trustpilot attracts people motivated to post, in both directions.
  • English only. Odoo in particular has substantial non English review volume: 540 one star reviews across all languages against 263 in English. Those are excluded because they cannot be quoted faithfully without translation.
  • One slice is truncated. Odoo's one star reviews were capped at 200 of 263 by Trustpilot's paging limit.
  • Coding is approximate. Roughly 89% precision, and 14.4% of negative reviews match no category at all.
  • Country is the reviewer's as recorded by Trustpilot, which is not necessarily where their business trades.
  • This is not a product ranking. Nothing here says which of these systems is best, and the sample sizes could not support that claim if it tried.

Systems covered: Unleashed, Linnworks, Cin7 Omni, Cin7 Core, Mintsoft, Veeqo, Orderwise, Brightpearl, StoreFeeder, Selro, Khaos Control, NetSuite, inFlow, Katana, Epicor, MRPeasy, Fishbowl, Megaventory, Odoo.

Why we ran this

We build operations systems for UK product businesses, so we had an obvious interest in what the answer would be. That is exactly why the method, the failed hypotheses and the error rate are all on this page: so you can discount for it.

If the data had shown that buyers mostly complain about missing features, the honest conclusion would have been that off the shelf tools are fine and the problem is picking the right one. It does not show that. It shows that the thing most likely to go wrong is the relationship you are entering into, and that the bill for it tends to arrive years after the decision.

Talk to us about your operation

Findings and figures on this page may be reused with attribution to OpsMavix and a link to this page.