Elasticsearch analyzers in index settings has no affect

I'm trying to set an analyzer to an index in its settings.

// PUT /customers 
{
  "mappings": {
    "properties": {
      "name": {
        "type": "text"
      },
      "title": {
        "type": "text"
      }
    },
    "dynamic": false
  },
  "settings": {
    "analysis": {
      "analyzer": "ik_smart"
    }
  }
}

Then I indexed some data.

// POST /customers/_doc
{
  "name": "张三",
  "title": "工程师"
}

// POST /customers/_doc
{
  "name": "李四",
  "title": "测试员"
}

Analyze with the ik_smart analyzer

//GET /customers/_analyze
{
  "text": "李四工程师",
  "analyzer": "ik_smart"
}

// gets tokens ['李四', '工程师']

And with the default analyzer

// GET /customers/_analyze
{
  "text": "李四工程师"
}

// gets tokens ['李', '四', '工', '程', '师']

Finally search for '李四工程师'

// GET /customers/_search
{
  "query": {
    "bool": {
      "must": {
        "multi_match": {
          "query": "李四工程师",
          "operator": "or",
          "type": "cross_fields",
          "fields": [
            "name^10",
            "title^7"
          ],
          "analyzer": "ik_smart"
        }
      }
    }
  }
}

// gets empty hits

If I put the ik_smart analyzer into each field

// PUT /customers 
{
  "mappings": {
    "properties": {
      "name": {
        "type": "text",
        "analyzer": "ik_smart"
      },
      "title": {
        "type": "text",
         "analyzer": "ik_smart"
      }
    },
    "dynamic": false
  }
}

Then the request works fine.

I suspect that the first settings does not apply the analyzer to any fields within the index.

I use the ik_smart plugin here, which provides an ik_smart analyzer, because Chinese language does not have spaces to separate characters, ik_smart is a dictionary-based analyzer. Without the ik_smart analyzer, any data containing Chinese words or sentences would be indexed as single characters. So it works when I use the default analyzer in searching as well, because the phrase 李四工程师 breaks into ['李', '四', '工', '程', '师'], and matches the indexed data. But that does not provide a very accurate relevancy.

If I use the ik_smart analyzer in searching, I get tokens ['李四', '工程师'], which do not match the indexed data.

So, why isn't the settings.analysis.analyzer works as expected?

What is the use of this setting anyway if it does not have any affects?