Do Sri Lankan Hotels Name the Fix? A 7,551-Reply Regional Index

Every review-reading guide tells you to look at how a hotel replies to its worst review. Nobody tells you where that habit is worth the effort. We built an index for it: of the 7,551 owner replies in the byc.lk corpus written under a review scoring below 7/10, we counted how many name an actual repair, replacement, renewal or upgrade — a specific artefact that was dealt with, rather than an apology or a policy restatement.

Across Sri Lanka the answer is 1,338 replies, 17.7%. That number is useless on its own. Split by region and it ranges from 3.6% in Pasikuda to 32.7% in Weligama — a nine-fold spread inside one small country. Where the volume of bad reviews sits is a different map: see the bad-review risk of every destination, which runs from 5.7% in Habarana to 20.9% in Wadduwa. In Pasikuda, roughly one complaint reply in 28 names the thing that broke. In Weligama it is one in three.

1. The short answer

2. What we counted, and what we deliberately excluded

A reply counts as naming a fix only if its lowercased text contains one of a fixed set of stems: fix, repaired, replaced, replacement, maintenance, renovated, renewed, upgraded. Compliments, thanks and apologies are ignored. So is every phrase that merely promises future action without an object (“we will look into it” scores zero, correctly).

Three exclusions that matter:

• Replies under 60 characters are dropped. A three-word courtesy reply cannot contain a fix, and counting it as a failure would drag every big chain's number down for no good reason.
• Replies that only quote the review are not scored. A number of the corpus's reply fields begin with a truncated copy of the guest's own text; we require the reply body to be the thing that matched, not the quotation.
• We do not score quality, only mention. A reply that says “we replaced the mattress” under a complaint about the mattress scores the same as one that says “we have replaced the towels, renewed the garden and upgraded the Wi-Fi” under a complaint about staff. This is a crude instrument. It counts whether the property reached for a concrete object, not whether it fixed the right thing.

3. The full index

Complaint replies by region, nationwide rate 17.7%:

• Weligama 32.7% (n=168 complaint replies) · Unawatuna 28.3% (219) · Polonnaruwa 23.6% (110) · Tangalle 23.3% (232)
• Arugam Bay 21.8% (238) · Hikkaduwa 21.3% (188) · Dambulla 21.2% (146)
• Yala 19.5% (487) · Trincomalee 19.2% (229) · Kandy 19.0% (522) · Ella 18.2% (483)
• Negombo 16.9% (649) · Nuwara Eliya 15.7% (363) · Sigiriya 15.5% (317) · Nilaveli 15.0% (100) · Galle 15.0% (267)
• Anuradhapura 12.6% (317) · Colombo 11.4% (809) · Bentota 11.1% (298) · Pasikuda 3.6% (168)

Sample sizes are honest here, and they are not all large: Nilaveli and Polonnaruwa rest on 100–110 complaint replies, which is enough to see a pattern and not enough to rank two places apart. The ends of the table are wide enough to be real: Pasikuda at 3.6% against Weligama at 32.7% would require a ninefold miscount to explain away.

4. What the top and bottom have in common

The regional ranking is not a quality ranking. Sigiriya (15.5%) has the highest property-weighted score of any major region and sits below the national rate on naming fixes. Colombo (11.4%) is a city where nearly every property is a managed hotel with a marketing department and a standard paragraph; its replies are long, warm and objectless.

The clearest single example of the bottom end is a Pasikuda resort with 36 complaints and 36 replies, none of which names a repair. Read a month of them and you learn that the property answers promptly, apologises properly and never once mentions an object. Whatever its maintenance process is, the reply is not evidence of it.

The top is the opposite shape. Culture Resort answering a complaint about a night without air conditioning names the technician who could not be found, the room upgrade, the meal offered in compensation and the mosquito net that was replaced. That is not better writing. It is a diff against a real work order, and it is only possible if a work order exists.

5. The honest caveat: this measures ownership as much as care

A family guesthouse replying from the owner's phone can say exactly what was repaired, because the owner knows. A 200-room chain answers through a template for reasons of legal caution, brand voice and scale — and its maintenance department may be excellent while its replies say nothing. So the index measures whether the person writing the reply has access to the operational facts, which is usually a question of size and ownership type.

That still makes it useful, for two reasons. First, if you are choosing between a guesthouse and a resort and you want to know what happens when something breaks, the reply field is a genuine signal on the guesthouse side, where it carries information. Second, an objectless reply on a small property where the owner clearly writes it themselves is a real warning — not about maintenance, but about attention.

What this cannot tell you: whether the fix happened, whether it happened before or after your complaint, or whether the property got better afterwards. There are no reply timestamps and no stay dates in this corpus, so every number here is cross-sectional. See the method section.

6. How to use it before you book

• Read the reply under the second-worst review, not the worst. The worst one gets the drafted paragraph from a manager. The second-worst usually gets the person who actually runs the place.
• Search that page for nouns. Air conditioning. Mattress. Shower. Mosquito net. Road. Boat. If a reply under a complaint about air conditioning does not contain the words air conditioning, the reply was written without reference to the complaint.
• Treat a promise as a non-answer. “We will ensure this does not happen again” is the phrase that scores zero here, and it appears by the thousand.
• Do not hold a Colombo hotel to a Weligama standard. Comparing the index by region tells you what reply behaviour to expect, not how to judge a single property.
• Cross-read with the review text. Our word study found that complaints about cleanliness almost never come from happy guests; if you see “dirty” in two recent reviews and no reply names cleaning, that is a booking to skip.
• Then read the verb tense. A reply can name a fix and still only promise one. Our reply-language audit finds 18.7% of complaint replies name a fix but just 2.9% say in the past tense that it was done — and why a 9.0 score can still be a misleading stay covers the score the reply is defending.

7. Where this comes from, and what to read next

The index is a re-cut of the dataset behind what hotels say back, which measured all 49,433 owner replies in the corpus, found that only 9.1% of them name a fix, that 39.9% dispute the guest and that 31.4% are exact duplicates. This page takes the same population and asks the narrower question — where does that 9% concentrate — because the national figure turns out to hide a nine-fold regional spread.

The instrument is deliberately crude and reproducible. Anyone with the same fields can recompute it; the stems are listed above and the exclusions are named. If you run it on a different slice and get a different number, the difference will be a denominator, not an opinion — the same trap that produced the 10.0% versus 13.6% reply-rate discrepancy documented on the original reply-rate page.

8. Method and limits

Sample. The 291,887-review slice of the byc.lk corpus (10-point scale, Booking.com source). Replies to reviews scoring strictly below 7 with a reply body of 60+ characters: 7,551. Of these, 1,338 (17.7%) contain at least one fix stem. Regional statements cover the 20 regions with 100+ qualifying complaint replies.

Instrument. Case-insensitive regex over the reply body: \b(fix|fixed|fixing|repair|repaired|replace|replaced|replacing|replacement|maintenance|renovat|renew|upgrad). A reply matching several stems counts once. No stemming beyond the listed forms, so “restored”, “serviced” and “sorted” are not counted — the true rate is therefore a floor, not a ceiling.

The largest limits. (1) No reply timestamps and no stay dates: nothing here is a before/after. (2) Mention is not action: a reply can claim a fix that never happened, and this instrument cannot tell. (3) Property type and ownership are not in the reply fields, so the ownership explanation in section 5 is an interpretation consistent with the pattern, not a tested effect. (4) Small regions: Nilaveli, Polonnaruwa and Dambulla rest on 100–146 complaint replies.

Frequently asked questions

10. The bottom line

Across 7,551 replies to bad Sri Lankan hotel reviews, 17.7% name something that was actually repaired, replaced or changed. That share is not evenly spread. It is 3.6% in Pasikuda and 32.7% in Weligama, and the ordering runs from small owner-run properties at the top to managed hotels and resort strips at the bottom.

The index does not tell you how good a hotel is. It tells you whether the person writing its replies has access to the operational truth — and on a property where the owner writes them personally, that is the single most informative paragraph on the page. If the reply to “the AC broke” never mentions the AC, you have learned something the star rating will never show you.

Method: 7,551 owner replies to reviews scoring under 7/10 with a reply body over 60 characters, drawn from the 291,887-review byc.lk corpus. Fix mention detected by regex over 14 stems; national rate 17.7% (1,338 replies). Regional table restricted to regions with 100+ qualifying replies, n shown for each. No reply or stay timestamps exist in the data, so all comparisons are cross-sectional and no claim is made about whether a stated fix was carried out. Last updated: 25 October 2026.

Plan your Sri Lanka trip with BYC at byc.lk — Free and honest. Some links are affiliate links and never change our advice.