Suppletion is extreme allomorphy in which a root is replaced by a form from a different etymological source in specific grammatical contexts. English go/went (from 'go' and archaic 'wend') is the classic example. Suppletion typically affects high-frequency words in inflectional paradigms (person, number, tense).
From your study of inflectional morphology, you know that regular inflection works by attaching morphemes to a stable root: *walk* → *walked*, *dog* → *dogs*, *fast* → *fastest*. The root stays constant; the affixes signal grammatical distinctions like tense, number, and degree. You also know from morpheme types that allomorphy — variation in the form of a morpheme — is common. The plural morpheme has allomorphs /s/, /z/, and /ɪz/ depending on the preceding sound (*cats*, *dogs*, *buses*), but the plural morpheme itself is still recognizably present in all three. Suppletion is what happens when this variation becomes so extreme that the root itself is unrecognizable: the connection between forms is paradigmatic (grammatically linked) but phonologically opaque.
The paradigm case is go/went. No phonological process derives *went* from *go* — the two forms share no phonological material. Historically, *went* is the past tense of *wend* (an archaic verb meaning to travel), which was borrowed into the *go* paradigm because *go* lacked a past tense form. This is suppletion: different etymological roots occupying different cells of the same inflectional paradigm. English has several more examples. *Be* is the most extreme: *am*, *is*, *are*, *was*, *were*, *be*, and *been* all fill cells of a single verb's paradigm while deriving from three different Proto-Indo-European roots. *Good/better/best* replaces the root entirely in the comparative and superlative rather than adding *-er* and *-est* to *good*. *Bad/worse/worst* does the same.
The frequency effect is striking and well-documented: suppletive forms are almost exclusively found among the most frequent words in a language. This is not coincidence. High-frequency words are acquired early, used constantly, and stored whole in the mental lexicon rather than being assembled from morphemes each time. Their irregular forms are reinforced through repeated exposure, so they don't get regularized by analogy (*I goed* is a child's error, not an adult's). Low-frequency words, by contrast, tend to be regularized because speakers encounter them rarely enough that the irregular form isn't entrenched in memory. When a language loses a word from active use, its suppletive forms are usually the first to disappear, as speakers regularize by analogy.
Suppletion matters theoretically because it challenges the view of morphology as a purely compositional, rule-based system. If words were always built by combining a root with affixes according to rules, suppletion would be impossible — the rules would require a stable root to work on. Instead, suppletion shows that paradigms (sets of grammatically related forms) can be stored and learned as wholes, with individual cells filled by whatever form is historically available, regardless of phonological relationship. This points toward a lexicalist view of morphology, where the lexicon stores not just roots but full paradigms for frequent and irregular items, alongside productive rules that generate regular forms for everything else.