[guided]For a maximal ideal $\mathfrak m \subset A$ and an $A$-module $M$, write $M_{\mathfrak m}:=A_{\mathfrak m}\otimes_A M$ for localization at $\mathfrak m$; in particular, $I_{\mathfrak m}$ and $J_{\mathfrak m}$ are the localizations of $I$ and $J$. The source group multiplies fractional ideals, while the Picard group multiplies classes by tensor product. So the essential compatibility is the statement that the natural map $I\otimes_A J\to IJ$ is not just a surjection but an isomorphism.
Let $I,J\in \mathcal I(A)$. Multiplication in the field $K$ gives an $A$-bilinear map
\begin{align*}
\mu_{I,J}:I\times J\to IJ,\qquad (x,y)\mapsto xy.
\end{align*}
Because $\mu_{I,J}$ is $A$-bilinear, the [universal property of the tensor product](/theorems/3971) gives a unique $A$-linear map
\begin{align*}
\widetilde{\mu}_{I,J}:I\otimes_A J\to IJ,\qquad x\otimes y\mapsto xy.
\end{align*}
Why should this be an isomorphism? The point is that over a Dedekind domain, fractional ideals become principal after localizing at a maximal ideal. To see this, first clear denominators: choose $d,e\in A\setminus\{0\}$ with $dI\subset A$ and $eJ\subset A$. Then $dI$ and $eJ$ are ordinary nonzero ideals of $A$. After localizing at a maximal ideal $\mathfrak m\subset A$, they become nonzero ideals of the discrete valuation ring $A_{\mathfrak m}$, so they are principal. Multiplication by $d^{-1}$ and $e^{-1}$ inside $K$ converts these local generators for $(dI)_{\mathfrak m}$ and $(eJ)_{\mathfrak m}$ into local generators for $I_{\mathfrak m}$ and $J_{\mathfrak m}$. Thus, for the fixed maximal ideal $\mathfrak m\subset A$, there exist elements $u_{\mathfrak m},v_{\mathfrak m}\in K^\times$ such that
\begin{align*}
I_{\mathfrak m}=u_{\mathfrak m}A_{\mathfrak m}\quad\text{and}\quad J_{\mathfrak m}=v_{\mathfrak m}A_{\mathfrak m}.
\end{align*}
Localization commutes with products of fractional ideals, so $(IJ)_{\mathfrak m}=I_{\mathfrak m}J_{\mathfrak m}=u_{\mathfrak m}v_{\mathfrak m}A_{\mathfrak m}$. After localization at $\mathfrak m$, the map $\widetilde{\mu}_{I,J}$ becomes multiplication of two free rank-one $A_{\mathfrak m}$-modules:
\begin{align*}
u_{\mathfrak m}A_{\mathfrak m}\otimes_{A_{\mathfrak m}}v_{\mathfrak m}A_{\mathfrak m}\to u_{\mathfrak m}v_{\mathfrak m}A_{\mathfrak m},\qquad u_{\mathfrak m}a\otimes v_{\mathfrak m}b\mapsto u_{\mathfrak m}v_{\mathfrak m}ab.
\end{align*}
This map is an isomorphism. Its inverse sends $u_{\mathfrak m}v_{\mathfrak m}c$ to $u_{\mathfrak m}\otimes v_{\mathfrak m}c$, and this is well-defined because both sides are free rank-one modules over $A_{\mathfrak m}$ with the displayed generators.
Now $I$, $J$, and $IJ$ are finitely generated $A$-modules, since nonzero fractional ideals over a Noetherian domain are finitely generated after clearing denominators. An $A$-linear map between finitely generated modules is an isomorphism exactly when its localization at every maximal ideal is an isomorphism. Therefore $\widetilde{\mu}_{I,J}$ is an isomorphism globally.
Consequently the class of $IJ$ in $\operatorname{Pic}(A)$ is the tensor-product class of $I$ and $J$:
\begin{align*}
[IJ]=[I\otimes_A J]=[I]+[J].
\end{align*}
This proves that $\Phi:\mathcal I(A)\to\operatorname{Pic}(A)$ is a group homomorphism.[/guided]