To my knowledge, currently the best "motivation" for the Todd class comes from the so called "orientation theory" and the formal group laws associated to "oriented" cohomology theories. As far as I know, these ideas go back to Quillen's work on complex cobordism, although current formulation is mainly due to Panin-Smirnov's works (here, here and here). I suggest to read this exposition, which briefly summarizes the main ideas.
Let me summarize the explanation without going into every detail. We say that a cohomology theory $H\colon \mathrm{Var}_k\to \mathrm{Rings}$ is oriented if it has a theory of Chern classes with all usual properties one might expect (invariance under pullback, Withney sum and projective bundle formula, nilpotence...). (I am cheating here, people define orientations differently, but there is a theorem stating that both things, orientations and reasonable Chern classes, are in fact equivalently under reasonable conditions). It is worth noticing two facts about this definition:
the cohomology is not assumed to be graded (so that $K$-theory is such a cohomology)
For $L,L'\to X$ two line bundles $c_1(L\otimes L')$ need not to be equal to $c_1(L) + c_1(L')$. Such cohomologies are called additive. For example, $K$-theory satisfies that $c_1(L)=1-[L^*]$ so that $c_1(L\otimes L')=c_1(L) + c_1(L')-c_1(L)\cdot c_1(L')$, which is said to be multiplicative (the name comes from the multiplicative formal group law, where the set of coordinates is taken with respect to 1. That is to say, (1-x)(1-y)=(1-x-y+xy)). The target of the Chern character is always and additive cohomology.
The fun thing is that a given cohomology theory $H$ may have different orientations, in other words, different theories of Chern classes. However two orientations can be related concretely: if $H$ is oriented, with some first Chern classes $c_1$, and we have a new orientation, with a new first Chern classes $c_1^{\mathrm{new}}$, then it is easy to check that $$ c_1^{\mathrm{new}}(L)=F(c_1(L))\cdot c_1(L) $$ where $F(t)\in H(pt)[[t]]$ is an invertible series and with coefficients in the cohomology ring of a point. This $F$ is the (inverse) Todd series associated to the new orientation and can be applied, through the usual multiplicative extension, to any virtual vector bundle.
Let $\varphi \colon H\to H'$ be a morphism of cohomology theories (i.e., natural transformation between contravariant functor of rings) it is also easy to check that the orientation on $H$ one provides an orientation on $H'$, in other words, there are new Chern classes on $H'$ and therefore, an associated (inverse) Todd series.
Main example: Consider the Chern character $\mathrm{ch}\colon K\to H_{\mathbb{Q}}$, which is a morphism of cohomologies where $H_{\mathbb{Q}}$ is additive. Let us $c_1^K$ denote the Chern classes with values in $K$-theory, to avoid confussion. Let us check the associated (inverse) Todd series to $\mathrm{ch}$: $$ \mathrm{ch}(c^K_1(L))=\mathrm{ch}(1-[L^*]))=1-e^{c_1(L^*)}=1-e^{-c_1(L)}=\frac{1-e^{-c_1(L)}}{c_1(L)}\cdot c_1(L) $$ Hence, its associated series is $\frac{1-e^{-t}}{t}$, which is precisely the inverse of the classical Todd series.
Other examples: The universal map from cobordsim to any cohomology theory $\Omega\to H$ also has its associated (inverse) Todd series, but it is trivial (the series is $F(t)=1$).
On any oriented cohomology theory $H$ you can consider a new theory of Chern classes: $c_1^\mathrm{new}(L):=c_1(L^*)$, its associated (inverse) Todd series is $F(t)=-1$.
-----0-----
In the aformentioned exposition you have these kind of arguments with details and "without cheating" (orientations are defined properly, and not just as a "theory of Chern classes"). The Riemann-Roch theorem in this context is as corollary of the universal property of $K$-theory (which is the universal multiplicative oriented cohomology theory).