Abstract Background Following CT, guidelines for staging oesophageal and gastro-oesophageal junction (GOJ) cancer recommend endoscopic ultrasonography (EUS), PET–CT and laparoscopy for T3–T4 GOJ tumours. These recommendations are based on generic utilities, but it is unclear whether the test risk outweighs the potential benefit for some patients. This study sought to quantify investigation risks, benefits and utilities, in order to develop pragmatic, personalized staging recommendations. Methods All patients with a histological diagnosis of oesophageal or GOJ cancer staged between May 2006 and July 2013 comprised a development set; those staged from July 2013 to July 2014 formed the prospective validation set. Probability thresholds of altering management were calculated and predictive factors identified. Algorithms and models (decision tree analysis, logistic regression, artificial neural networks) were validated internally and independently. Results Some 953 patients were staged following CT, by [18F]fluorodeoxyglucose PET–CT (918), EUS (798) and laparoscopy (458). Of these patients, 829 comprised the development set (800 PET–CT, 698 EUS, 397 laparoscopy) and 124 the validation set (118 PET–CT, 100 EUS, 61 laparoscopy). EUS utility in the 71·8 per cent of patients with T2–T4a disease on CT was minimal (0·4 per cent), its risk exceeding benefit. EUS was moderately accurate for pT1 N0 disease. A number of factors predicted metastases on PET–CT and laparoscopy, although none could inform an algorithm. PET–CT altered management in 23·0 per cent, and laparoscopy in 7·1 per cent, including those with T2 and distal oesophageal tumours. Conclusion Although EUS provided additional information on T and N category, its risk outweighed potential benefit in patients with T2–T4a disease on CT. Laparoscopy seemed justified for distal oesophageal tumours of T2 or greater.