diff --git a/NAMESPACE b/NAMESPACE
index 85dd9fe84..5b8705d41 100644
--- a/NAMESPACE
+++ b/NAMESPACE
@@ -563,6 +563,7 @@ export(vec_chop)
 export(vec_compare)
 export(vec_count)
 export(vec_data)
+export(vec_deduplicated)
 export(vec_default_cast)
 export(vec_default_ptype2)
 export(vec_detect_complete)
diff --git a/R/deduplicated.R b/R/deduplicated.R
new file mode 100644
index 000000000..b69cbaa1e
--- /dev/null
+++ b/R/deduplicated.R
@@ -0,0 +1,40 @@
+#' Modify a function to act on a deduplicated vector input
+#'
+#' @description
+#'
+#' `r lifecycle::badge("experimental")`
+#'
+#' The deduplicated function acts on the unique values in the first input `x`
+#' and expands the output back to return. The return value is equivalent to `f(x)`
+#' but is significantly faster for inputs with significant duplication.
+#'
+#' @param f Function whose first argument will be deduplicated.
+#'
+#' @return A deduplicated function
+#' @export
+#'
+#' @examples
+#' x <- sample(LETTERS, 10)
+#' x
+#'
+#' large_x <- sample(rep(x, 10))
+#' length(large_x)
+#'
+#' long_func <- function(x) for(i in x) {Sys.sleep(0.001)}
+#'
+#' system.time(y <- long_func(large_x))
+#' system.time(y2 <- vec_deduplicate(long_func)(large_x))
+#' all(y == y2)
+vec_deduplicated <- function(f) {
+  function(x, ...) {
+    res <- vec_group_id_and_loc(x)
+    group_id <- unclass(res)
+    unique_loc <- attr(res, "unique_loc")
+    unique_x <- vec_slice(x, unique_loc)
+    f(unique_x, ...)[group_id]
+  }
+}
+
+vec_group_id_and_loc <- function(x) {
+  .Call(vctrs_group_id_and_loc, x)
+}
diff --git a/R/group.R b/R/group.R
index 2c91df28a..579cc20d0 100644
--- a/R/group.R
+++ b/R/group.R
@@ -30,6 +30,7 @@
 #'   Note that when using `vec_group_loc()` for complex types, the default
 #'   `data.frame` print method will be suboptimal, and you will want to coerce
 #'   into a tibble to better understand the output.
+#'
 #' @name vec_group
 #'
 #' @section Dependencies:
diff --git a/_pkgdown.yml b/_pkgdown.yml
index 812117b9f..76873bf97 100644
--- a/_pkgdown.yml
+++ b/_pkgdown.yml
@@ -131,3 +131,4 @@ reference:
   - vec_names
   - vec_as_location
   - vec_as_subscript
+  - vec_deduplicated
diff --git a/man/vec_deduplicated.Rd b/man/vec_deduplicated.Rd
new file mode 100644
index 000000000..5962a817a
--- /dev/null
+++ b/man/vec_deduplicated.Rd
@@ -0,0 +1,34 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/deduplicated.R
+\name{vec_deduplicated}
+\alias{vec_deduplicated}
+\title{Modify a function to act on a deduplicated vector input}
+\usage{
+vec_deduplicated(f)
+}
+\arguments{
+\item{f}{Function whose first argument will be deduplicated.}
+}
+\value{
+A deduplicated function
+}
+\description{
+\ifelse{html}{\href{https://lifecycle.r-lib.org/articles/stages.html#experimental}{\figure{lifecycle-experimental.svg}{options: alt='[Experimental]'}}}{\strong{[Experimental]}}
+
+The deduplicated function acts on the unique values in the first input \code{x}
+and expands the output back to return. The return value is equivalent to \code{f(x)}
+but is significantly faster for inputs with significant duplication.
+}
+\examples{
+x <- sample(LETTERS, 10)
+x
+
+large_x <- sample(rep(x, 10))
+length(large_x)
+
+long_func <- function(x) for(i in x) {Sys.sleep(0.001)}
+
+system.time(y <- long_func(large_x))
+system.time(y2 <- vec_deduplicate(long_func)(large_x))
+all(y == y2)
+}
diff --git a/src/group.c b/src/group.c
index 81c321783..ebc09f7cf 100644
--- a/src/group.c
+++ b/src/group.c
@@ -233,3 +233,54 @@ SEXP vec_group_loc(SEXP x) {
   UNPROTECT(nprot);
   return out;
 }
+
+
+// -----------------------------------------------------------------------------
+
+// [[ register() ]]
+SEXP vctrs_group_id_and_loc(SEXP x) {
+  int nprot = 0;
+
+  R_len_t n = vec_size(x);
+
+  x = PROTECT_N(vec_proxy_equal(x), &nprot);
+  x = PROTECT_N(vec_normalize_encoding(x), &nprot);
+
+  struct dictionary* d = new_dictionary(x);
+  PROTECT_DICT(d, &nprot);
+
+  SEXP out = PROTECT_N(Rf_allocVector(INTSXP, n), &nprot);
+  int* p_out = INTEGER(out);
+
+  R_len_t g_id = 1;
+
+  struct growable g_unq = new_growable(INTSXP, 256);
+  PROTECT_GROWABLE(&g_unq, &nprot);
+
+  for (int i = 0; i < n; ++i) {
+    uint32_t hash = dict_hash_scalar(d, i);
+    R_len_t key = d->key[hash];
+
+    if (key == DICT_EMPTY) {
+      dict_put(d, hash, i);
+      // Record group id
+      p_out[i] = g_id;
+      ++g_id;
+
+      // Record unique locs
+      growable_push_int(&g_unq, i + 1);
+
+    } else {
+      p_out[i] = p_out[key];
+    }
+  }
+
+  SEXP n_groups = PROTECT_N(Rf_ScalarInteger(d->used), &nprot);
+  Rf_setAttrib(out, syms_n, n_groups);
+
+  SEXP unq_vals = growable_values(&g_unq);
+  Rf_setAttrib(out, Rf_install("unique_loc"), unq_vals);
+
+  UNPROTECT(nprot);
+  return out;
+}
diff --git a/src/init.c b/src/init.c
index 9b338e3df..90495be30 100644
--- a/src/init.c
+++ b/src/init.c
@@ -27,6 +27,7 @@ extern SEXP vec_split(SEXP, SEXP);
 extern SEXP vctrs_group_id(SEXP);
 extern SEXP vctrs_group_rle(SEXP);
 extern SEXP vec_group_loc(SEXP);
+extern SEXP vctrs_group_id_and_loc(SEXP);
 extern SEXP vctrs_equal(SEXP, SEXP, SEXP);
 extern r_obj* ffi_vec_detect_missing(r_obj*);
 extern r_obj* ffi_vec_any_missing(r_obj* x);
@@ -211,6 +212,7 @@ static const R_CallMethodDef CallEntries[] = {
   {"vctrs_group_id",                            (DL_FUNC) &vctrs_group_id, 1},
   {"vctrs_group_rle",                           (DL_FUNC) &vctrs_group_rle, 1},
   {"vctrs_group_loc",                           (DL_FUNC) &vec_group_loc, 1},
+  {"vctrs_group_id_and_loc",                    (DL_FUNC) &vctrs_group_id_and_loc, 1},
   {"ffi_size",                                  (DL_FUNC) &ffi_size, 2},
   {"ffi_list_sizes",                            (DL_FUNC) &ffi_list_sizes, 2},
   {"vctrs_dim",                                 (DL_FUNC) &vctrs_dim, 1},
diff --git a/tests/testthat/test-deduplicated.R b/tests/testthat/test-deduplicated.R
new file mode 100644
index 000000000..cf59259f5
--- /dev/null
+++ b/tests/testthat/test-deduplicated.R
@@ -0,0 +1,73 @@
+# group_id_and_loc ----------------------------------------------------------------
+
+expect_matches_separate_calls <- function(x) {
+  expect_equal(
+    as.numeric(vec_group_id_and_loc(x)),
+    as.numeric(vec_group_id(x))
+  )
+  expect_equal(
+    attr(vec_group_id_and_loc(x), "unique_loc"),
+    vec_unique_loc(x)
+  )
+}
+
+test_that("vec_group_id_and_loc matches vec_group_id and vec_unique_loc", {
+  x <- c(2, 4, 2, 1, 4)
+  expect_matches_separate_calls(x)
+})
+
+test_that("vec_group_id_and_loc works for size 0 input", {
+  expect <- structure(integer(), n = 0L, unique_loc=integer())
+  expect_equal(vec_group_id_and_loc(NULL), expect)
+  expect_equal(vec_group_id_and_loc(numeric()), expect)
+})
+
+test_that("vec_group_id_and_loc works on base S3 objects", {
+  x <- factor(c("x", "y", "x"))
+  expect_matches_separate_calls(x)
+
+  x <- new_date(c(0, 1, 0))
+  expect_matches_separate_calls(x)
+})
+
+test_that("vec_group_id_and_loc works on data frames", {
+  df <- data.frame(x = c(1, 2, 1, 1), y = c(2, 3, 2, 3))
+  expect_matches_separate_calls(df)
+})
+
+test_that("vec_group_id_and_loc works on arrays", {
+  x <- array(c(1, 1, 1, 2, 4, 2), c(3, 2))
+  expect_matches_separate_calls(x)
+})
+
+test_that("vec_group_id takes the equality proxy", {
+  local_comparable_tuple()
+  x <- tuple(c(1, 2, 1, 1), c(1, 1, 1, 2))
+  # Compares on only the first field
+  expect_matches_separate_calls(x)
+})
+
+test_that("vec_group_id takes the equality proxy recursively", {
+  local_comparable_tuple()
+
+  x <- tuple(c(1, 2, 1, 1), 1:4)
+  df <- data_frame(x = x)
+  expect_matches_separate_calls(df)
+})
+
+
+# vec_deduplicate ---------------------------------------------------------
+
+test_that("vec_deduplicated(f) runs only on deduplicated values", {
+  ncalls <<- 0
+  f <- function(ii) for(i in ii) ncalls <<- ncalls + 1
+
+  x <- c(1, 1, 1, 2, 3)
+  vec_deduplicated(f)(x)
+  expect_equal(ncalls, 3)
+
+  ncalls <<- 0
+  x <- 1:5
+  vec_deduplicated(f)(x)
+  expect_equal(ncalls, 5)
+})