What is a customer intelligence layer, and who holds yours?

In the 12 months to 9 September 2026, one Australian store network replied to 1,389 of its five-star Google reviews. In the same 12 months, 460 of its one and two star reviews got no reply at all.

No single listing shows that split. A listing shows a rating, a count and whether each review has a reply. The split by star rating appears only when every review at every store is counted in one place, and someone has decided what counts as a complaint.

That counting and deciding is the subject of this note. A customer intelligence layer is the set of definitions and judgements that sits between customer records and the decisions made from them. A business with dozens or hundreds of sites already has much of the material for this layer in public, in the reviews and star ratings on every listing, and very little of it anywhere it can read as a whole.

What is a customer intelligence layer?

Records say what happened: a review was posted, a rating changed. Decisions are what the business does next: which store gets help first, which complaint gets answered first. The layer in between is mostly choices someone has to write down, such as treating a one or two star review as a complaint.

It is easy to mistake for the tools around it. The CDP Institute defines a customer data platform as software that “creates and maintains a persistent, unified customer record that is accessible to other systems”. Business intelligence reports on questions already asked, and a feedback tool collects what customers say. The layer decides what the records mean for this business, wherever it sits. Source: CDP Institute

Where does a business’s customer intelligence already live?

Six Australian store networks received about 35,000 Google reviews across 590 listings in the 12 months before each was captured, between 13 August and 10 September 2026 (NTWRK review censuses).

Think of each listing as a drawer. Open one and you get that store’s rating, its review count and its reviews. The questions a network asks run across the drawers: which stores are slipping, and which step in the service keeps going wrong. No drawer holds the answer, because no drawer can see the others.

Left, three stacked bands: records at the bottom, decisions at the top, and between them the layer, which holds what counts as a complaint and which gap is worth a look. Right, nine store listing cards, each holding a dashed fragment of the same layer and a bar that adds up only that listing's own reviews.

From records to decisions

Decisions

which store gets help first

The layer

what counts as a complaint

which gap is worth a look

Records

reviews and star ratings

Where it sits today

each listing adds up only its own reviews

What does the layer show that a listing does not?

None of the three patterns below appears on a listing or in its totals.

Last on reply rate. First on complaints answered.

Three networks, called A, B and C here, show that a reply rate does not say who was answered (censuses captured 9 and 10 September 2026).

Network A replied to 20.7% of its reviews in the 12 months before capture, the lowest of the three. Yet it answered 90.9% of its 1,224 complaints (one or two star reviews), against 11.9% of everything else. Almost half of its 2,270 replies went to complaints, which were about one review in nine.

Network B replied to 83.4% of all its reviews, and to 57.8% of its 441 complaints. That left 186 complaints without an answer in a year when it replied to 11,290 other reviews.

Network C is the network this note opened with. It answered 43.2% of its 810 complaints and 44.2% of its 3,141 five-star reviews, so a five-star review was slightly more likely to get a reply than a complaint.

A listing shows whether each review has a reply. The split by star rating appears only once every review at every store is counted.

Two ways to rank the same replies, 12 months before capture

Slope chart, one line per network, joining the share of all reviews replied to (left) with the share of complaints answered (right). Network A rises from 20.7 percent to 90.9 percent. Network B falls from 83.4 percent to 57.8 percent. Network C moves from 41.5 percent to 43.2 percent. Network A's line is highlighted.

Replied to all reviews

Complaints answered

0%50%100% A 20.7% B 83.4% C 41.5% 90.9% A 57.8% B 43.2% C

complaint = a one or two star review

The same figures as the chart above. Complaint means a one or two star review.

Network	Replied to all reviews	Complaints answered	Comparison group answered
A	20.7% (2,270 of 10,953)	90.9% (1,113 of 1,224)	Other reviews 11.9% (1,157 of 9,729)
B	83.4% (11,545 of 13,847)	57.8% (255 of 441)	Other reviews 84.2% (11,290 of 13,406)
C	41.5% (1,968 of 4,739)	43.2% (350 of 810)	Five-star reviews 44.2% (1,389 of 3,141)

The rating on the listing is running late.

A listing shows one rating, rounded to one decimal. Of 223 stores in three networks with 10 or more reviews in the last 12 months, 64 sit more than 0.3 stars from that rating (censuses captured 13 August and 9 September 2026).

The clearest case is a national bedding franchise. Its listing ratings, weighted by review count, come to 4.64. Its 3,296 reviews in the 12 months to 13 August 2026 averaged 4.85. Nineteen of its 73 stores sit more than 0.3 stars above their listing, and none sits that far below. In the other two networks the gap more often runs the other way: 31 stores below their listing, 14 above.

Picture the regional review. The improving store is judged on a number that has not caught up with it, and the slipping store keeps credit for a better past. The comparison here uses a plain 12-month average, not Google’s own calculation.

Each dot a store: last 12 months minus its listing rating

Dot strip of 223 stores in three networks. Each dot sits at the store's average rating over the last 12 months minus the rating on its Google listing, on a scale from minus 1.1 to plus 1.0 stars, with a shaded band from minus 0.3 to plus 0.3. 64 stores fall outside the band. At the bedding franchise 19 fall above it and none below. In the second network 9 fall below and 11 above; in the third, 22 below and 3 above.

0.3 stars either side

Bedding franchise · 73 stores

0 below · 19 above

Second network · 65 stores

9 below · 11 above

Third network · 85 stores

22 below · 3 above

-1.0-0.50+0.5+1.0

listing flatters the store

store is ahead of its listing

The same 223 stores as the strip above. Stores with 10 or more reviews in the 12 months before capture.

Network	Stores with 10+ reviews in 12 months	More than 0.3 below listing	More than 0.3 above listing
Bedding franchise	73	0	19
Second network	65	9	11
Third network	85	22	3
Total	223	31	33

Customers praise the people. Complaints gather at the steps.

Five of the networks had praise and complaint mentions counted theme by theme, and staff is the top praise theme in all five, at one of them by a single mention. Complaints outnumber praise in four of the 31 theme rows, and three of those four are steps in the service. Checkout at one network drew 30 complaint mentions against 19 praise, and booking at another 64 against 42. At a third, the order itself drew 321 against 206. Each of those steps runs at roughly three complaint mentions for every two of praise. The fourth row is price, close to level at 254 against 244.

Themes were labelled by an automated step against a different list per network, so compare ranks and ratios, never shares. A store score says where customers were unhappy. The text says at which step.

Theme labels differ by network. Compare ratios, never shares.

Theme	Complaint mentions	Praise mentions	Complaints per praise mention
Checkout (one network) steps in the service	30	19	1.6
Booking (another network)	64	42	1.5
The order (a third network)	321	206	1.6
Price	254	244	1.0

Who is writing the definition?

The definition a buyer meets is mostly written by sellers. Asked “What is first-party customer intelligence and how is it different from a traditional customer data platform?”, two AI answer engines cited cdp.com in 81 of their 130 answers, liveramp.com in 77, teradata.com in 66, commonroom.io in 65 and qualtrics.com in 49 (NTWRK answer-engine panel, prompt cat-01, 2 June to 9 September 2026). A third engine’s 66 answers to the same prompt cited no source. All five sell software or data platforms, and cdp.com is managed by Treasure AI, which calls itself a Customer Data Platform company. Source: cdp.com

What does it cost when the layer is rented or scattered?

Reply effort is the easiest cost to count. Every one of network C’s 1,389 replies to five-star reviews took someone’s time while 460 complaints went unanswered. A complaint count by store, with no count by step, cannot show which step is failing.

The harder cost arrives when a supplier changes. If the definition of a complaint and the rule for which reviews get answered first live only inside a supplier’s tool, the next supplier starts without them. An export carries the records, and the learning stays behind. The removability test asks what still works once the supplier has gone.

What can public reviews not tell you?

Reviewers choose to write, so reviews are not a sample of customers. A review is dated when posted, which may not be when the visit happened.

Nor do these figures show that replying changes a rating. A 36-store specialty retailer replied to 93.1% of its 144 reviews in 2025, when 17.4% of them were one or two stars. In 2026 to 9 September it replied to 13.3% of 128 reviews, and 13.3% of those were one or two stars (census captured 9 September 2026). One retailer with nothing to compare against cannot show that replies make a difference, or that they do not.

Every figure here is observational: it shows where to look, never why.

How can you tell whether you hold your customer intelligence layer?

Four questions for the next network meeting:

1. Which stores have a listing rating more than 0.3 stars from their last 12 months, and in which direction?
2. Last quarter, what share of one and two star reviews got an answer, store by store?
3. Which step in the service draws more complaint mentions than praise, and at which stores?
4. If the review supplier left next month, which of these answers could your own team still produce?

If the honest answer to the last one is to ask the supplier, the supplier holds the layer.

Frequently asked questions

Is a customer intelligence layer the same as a CDP?

No. A CDP keeps a unified customer record. The layer is the definitions and judgements applied to records, including ones a CDP may never hold, such as public reviews.

Is it the same as business intelligence?

No. Business intelligence reports on questions already asked. The layer holds the definitions those reports rely on and can surface patterns nobody asked about.

Do Google reviews count as customer intelligence?

They are customer records. They become intelligence once a business decides what to look for across them, such as replies split by star rating.

How many reviews does a store comparison need?

No single number. In a simulation on four Australian store networks (10 September 2026), 900 reviews spread as 15 stores of 60 detected real differences between stores more often than 60 stores of 15 in three networks, and less often in the fourth.