We predict late-time optical/UV emission from tidal disruption events (TDEs) from our slim accretion disc model \citep{Wen20} and explore the impact of the black hole mass $M_\bullet$, black hole spin $a_\bullet$, and accretion disc size. We use these synthetic spectra to successfully fit the multi-band \emph{Swift} observations of ASASSN-14li at >350 days, setting only the host galaxy extinction and outer disc radius as free parameters and employing the $M_\bullet$, $a_\bullet$, disc inclination, and disc accretion rates derived from fitting 10 epochs of ASASSN-14li's X-ray spectra with the slim disc. To address the nature of the \emph{early}-time optical/UV emission, we consider two models: shock dissipation and reprocessing. We find that (1) the predicted late-time optical/UV colour (e.g., $u-w2$) is insensitive to black hole and disc parameters unless the disc spreads quickly; (2) a starburst galaxy extinction model is required to fit the data, consistent with ASASSN-14li's post-starburst host; (3) surprisingly, the outer disc radius is $\approx$2$\times$ the tidal radius and $\sim$constant at late times, showing that viscous spreading is slow or non-existent; (4) the shock model can be self-consistent if $M_\bullet \lesssim 10^{6.75}$M$_\odot$, i.e., on the low end of ASASSN-14li's $M_\bullet$ range ($10^{6.5-7.1}$M$_\odot$; 1$σ$ CL); larger black hole masses require disruption of an unrealistically massive progenitor star; (5) the gas mass needed for reprocessing, whether by a quasi-static or an outflowing layer, can be $
15 pages, 10 figures. Accepted for publication in MNRAS