SUMMARY We develop and verify an automated workflow for full-waveform tomography based on spectral element and adjoint methods. We choose the North Island, New Zealand as a study area because of its high seismicity, extensive seismic network, and the availability of a candidate ray tomography starting model. To assess the accuracy of this model, we simulated 250 regional earthquakes using a spectral element solver, and compared the resulting synthetics with recorded waveforms. In a 10–30 s passband, reasonable cross-correlation phase and amplitude misfits exist between data and synthetics, whereas at 2–30 s, waveform misalignment is severe enough that meaningful cross-correlation measurements are no longer possible. To improve the velocity model at these short periods, we created an automated inversion framework based on existing tools for signal processing, phase measurement, nonlinear optimization, and workflow management. To verify the inversion framework, we performed a realistic synthetic inversion for 3-D checkerboard structure and analyzed model recovery, misfit reduction, and waveform improvement. The results of this analysis show that the source–receiver distribution within the chosen domain is capable of resolving velocity anomalies in regions of sufficient data coverage, and of magnitudes comparable to those expected in a real seismic inversion. Along with this finding, the relative ease of use and reliability of the workflow motivates future efforts targeting a high-resolution (2–30 s), large-scale (>50 000 measurements) seismic inversion for the North Island. Updated models from such an inversion are expected to improve ground motion predictions, constrain complex velocity structures, and advance understanding of New Zealand tectonics.